Wikipedia:AI noticeboard

"WP:AIN" redirects here; not to be confused with Wikipedia:Administrators' noticeboard or Wikipedia:Administrators' noticeboard/Incidents.

AI noticeboard

This page is for reporting issues of AI misuse.

• For issues concerning a single article, please refer to the guide on how to deal with AI-generated content before posting here. The guide consists of advice and resources about AI cleanup in general, and may be useful for editors interested in participating here.
• More general discussion about AI use and policies should be brought up at the village pump.
• You can inform an editor of a discussion involving them with {{subst:AINB-notice}}, but this is not required.
• {{subst:AIC status}} can be used to create a hatnote keeping track of a case's cleanup status.
• {{AINB status}} can be used to create a banner keeping track of the status of the user involved in a case.
• {{AINB article}} can be used to create a list of articles that may require cleanup.
• {{Userlinks}} can be used to provide informational links for a user, if not using one of the cleanup-specific templates.
• To view all open cases (including those whose reports have been archived), see Category:AI noticeboard open cleanup cases.

Start a new discussion

Backlog bonanza

Archives 1–6

Main page: Wikipedia:AI noticeboard/Backlog bonanza (Archives 1–6)

Welcome to the first instalment of Backlog Bonanza (click the link above, be prepared to be disappointed). All the LLMPRODs from Archives 1–6 are done (besides what's listed here). I haven't done any normal reverting, there's still loads of that in the archives if anyone fancies systematically going through them. At the linked page are users whose AINB reports got archived to Archives 1–6 or have a page in Category:AI noticeboard open cleanup cases. They're either recently editing and unresolved re whether to block, or consensus is needed re whether to LLMPROD their articles (and when to start from). Kowal2701 (talk, contribs) 20:32, 3 July 2026 (UTC)

for the masochistic there's always the list of user talk pages with uw-ai (not for LLMPROD, but for backlog) Gnomingstuff (talk) 21:02, 3 July 2026 (UTC)

Also probably worth going through G15 notices, only ~500 of them Kowal2701 (talk, contribs) 17:45, 8 July 2026 (UTC)

@Admins willing to patrol AINB: Can we please get a block for Schaumturducken2020 (talk · contribs), they were reported January 2026 and warned many times but ignored it, and are still editing. Their most recent creation was 29 June which has a bunch of AI signs like At least one source states the Principles comprised fifteen articles in editions circulating between 1933 and 1941, expanded to seventeen in the 1942 NDH-era Crljen edition. The combined 1933/1941 publication consulted by Tomasevich gives seventeen principles. Readers should consult Jareb's 2007 study (pp. 124–128) directly to resolve this discrepancy., em dashes, ...though it played an important political and psychological role in Croatian history., WP:AIATTR Kowal2701 (talk, contribs) 10:57, 6 July 2026 (UTC)

Blocked Schaumturducken2020 (talk · contribs) indefinitely from article space, as they have recently misused LLMs to generate article content without acknowledging previous talk page warnings.​ — Newslinger talk 21:30, 7 July 2026 (UTC)

@UtherSRG: It appears you are the mentor of Schaumturducken2020, who is apparently using AI to write, and has not acknowledged concerns placed on his user talk page. I don't see any interaction between the two of you since you posted the announcement of mentorship on the user talk page, but I wanted to know if you also support a mainspace block or some other administrative action. This is a collaborative project, communication is a critical element for collaboration, and Schaumturducken2020 is not communicating. ~Anachronist (who / me) (talk) 14:45, 6 July 2026 (UTC)

Yes, communication is crucial. If they are unresponsive, I have no problem with going down the block route. - In solidarity, UtherSRG (talk) 01:07, 7 July 2026 (UTC)

• I've made tracking pages for all the cases needing one in these archives, the ones left to potentially make I've listed at the above page (most of them are waiting on the LLMPRODs tbh) Kowal2701 (talk, contribs) 19:03, 5 August 2026 (UTC)

Archive 7

Main page: Wikipedia:AI noticeboard/Backlog bonanza (Archive 7)

Done all the LLMPRODs in Archive 7 other than those listed, same story as above. More cases from Archive 7 than there were from 1, 2, 3, 4, 5, and 6 combined. Kowal2701 (talk, contribs) 19:22, 6 July 2026 (UTC)

made all the tracking pages I could for cases which needed them, have listed ones I couldn't at the subpage Kowal2701 (talk, contribs) 13:33, 8 August 2026 (UTC)

Archives 8–9

Main page: Wikipedia:AI noticeboard/Backlog bonanza (Archives 8–9)

Same again. This should be the last big one from the archives since it's the last before WP:LLMPRV was passed. Enjoy Kowal2701 (talk, contribs) 13:35, 15 July 2026 (UTC)

Safeer Zaman (talk · contribs),

Previous AINB thread, they were warned 2 May to no response and were notified of their articles being PRODed and draftified explicitly for LLM use (also in May). (20 July) has chatbot communication. (1 August) has loads of AI signs. @Admins willing to patrol AINB: I think a mainspace block is needed, they've never edited their user talk Kowal2701 (talk, contribs) 22:06, 10 August 2026 (UTC)

Blocked from mainspace, and I reverted the obvious AI authorship on August 1. Cleanup can commence. ~Anachronist (who / me) (talk) 01:41, 11 August 2026 (UTC)

(just noting here) Previous AINB thread in May, they received a thorough warning but went inactive, only to return 2 July to make an AI-generated draft Kowal2701 (talk, contribs) 13:59, 12 August 2026 (UTC)

What an incredibly ironic article topic to generate. Btw, it was submitted to article space and then draftified. ‑‑gurkubondinn 14:04, 12 August 2026 (UTC)

tracking pages done for these, only one I wasn't sure on (the others are waiting on LLMPRODs) Kowal2701 (talk, contribs) 22:14, 17 August 2026 (UTC)

ANI edition

Main page: Wikipedia:AI noticeboard/Backlog bonanza (ANI edition)

This is a bit of a monster. I've included cases from the ANI archives up to 22 March 2026 (around the time NOLLM was passed) that need cleanup, there's (at time of writing) 50 cases that need more analysis, and an additional list of cases that need LLMPRV applied (most are no-brainers but still use judgement). I might add to this at some point, probably up until early May 2026 when LLMPRV was passed. I am dreading knocking on SPI's door ngl Kowal2701 (talk, contribs) 11:34, 21 July 2026 (UTC)

When an active editor like Shaumturducken2020 is blocked and then goes silent immediately with no activity for weeks, it indicates either that the editor gave up and left, or resumed the same activity as a sockpuppet. Knocking on the SPI door is unlikely to yield anything unless you have a suspected sockpuppet candidate, because checkusers aren't allowed to engage in fishing expeditions. Once you do have a candidate, a checkuser often does find related activity, however. ~Anachronist (who / me) (talk) 12:51, 21 July 2026 (UTC)

Yeah it was more to see if there were any prolific LLM-using sockmasters or old cases still needing cleanup, and whether we can setup some sort of system (idk). Having gone through the ANI archives, like half of LLM cases turned out to be socks, and while most of their creations were WP:G5'd, obv some were left and not tagged w {{AI-generated}} and normal edits weren't always reverted. AFAIK it's usually up to the CU to do the cleanup, which seems like a pretty terrible system, but idk, will probably message Asilvering about it Kowal2701 (talk, contribs) 14:03, 21 July 2026 (UTC)

there absolutely are, but a lot of them aren't identified as AI Gnomingstuff (talk) 06:45, 26 July 2026 (UTC)

I don't think it should be up to a CU or any other admin to clean up everything; they should do whatever is needed that normal editors cannot do themselves (block accounts, protect pages, and delete pages), tag AI-generated pages for socks that cannot be publicly disclosed, and the rest of the community can assist in the cleanup. ~Anachronist (who / me) (talk) 13:49, 26 July 2026 (UTC)

PanoptikEU

Cleanup efforts are ongoing.

• PanoptikEU (talk · contribs · deleted contribs · logs · filter log · block user · block log)

The user has good intentions and probably created all their articles before WP:LLM was prohibited but still we need to do a lot of cleanup. Most of their articles were created rapidly without any human review. Maybe convert notable Estonian articles into one-line stubs. ~2026-41256-48 (talk) 12:13, 25 July 2026 (UTC)

129 creations in 233 edits Kowal2701 (talk, contribs) 18:38, 28 July 2026 (UTC)

Their latest creations are obviously LLM-generated. Looking at their early articles, they seem decent quality compared to what we normally see and there aren't really enough meat on them for AISIGNS, but: at Sirje Olesk ref #1 doesn't support She is a recognized authority on the Finnish-Estonian writer Aino Kallas and the poet Lydia Koidula., Konrad Duden Prize has WP:AIREDCAT, and in Sven-Erik Soosaar ref #3 appears hallucinated Kowal2701 (talk, contribs) 21:22, 29 July 2026 (UTC)

Tbh I'm not 100% sure these are all AI, would appreciate a second opinion Kowal2701 (talk, contribs) 21:30, 29 July 2026 (UTC)

I disagree, don't think these three are AI - Sirje Olesk source kinda supports it thru "She has mainly dealt with various aspects of Estonian poetry, Estonian-Finnish literary relations, the life and work of Aino Kallas and Lydia Koidula, studied the history of the Estonian Writers’ Union and literary life in Soviet Estonia". Soosaar's ref #3 did in fact exist but the page went down - see https://web.archive.org/web/20190915191335/https://www.ut.ee/et/uritused/sven-erik-soosaar-eesti-keele-kultuursonavara-arengujooni-eesti-keel-piirikeel, an archived copy from 2019 (display is broken, but if you scroll down a bit you find the article).

And as for the red category, i know (especially before I got HotCat) I've occasionally put in categories assuming they were functional categories and then checked, noticed they weren't, and deleted - i find it harder to believe this is AI than that this is a user forgetting to check whether a reasonable category they inputted actually exists. Jishara (talk) 01:58, 1 August 2026 (UTC)

I've put a few into Pangram and they're consistently coming back as AI-assisted, I think this meets the threshold for LLMPROD but does anyone object? Kowal2701 (talk, contribs) 12:14, 11 August 2026 (UTC)

No objections from me, I support using LLMPRV for cleanup. ‑‑gurkubondinn 23:10, 19 August 2026 (UTC)

I looked at Avo Keerend and support LLMPRV. WP:AIATTR and typical shifting of facts among sources. M kuhner (talk) 16:52, 20 August 2026 (UTC)

Right. I’m in favour. LLMPRV it is. WorldClassChampion (talk) 17:03, 20 August 2026 (UTC)

Sorry guys, this one's too strange, don't feel comfortable doing LLMPRODs at this point. Some articles were created within 2 minutes of each other, which seems quick even for chatbots, yet they're not the obvious slop we'd expect and Pangram gives them as AI-assisted. The editor's first edits include changing a deprecated infobox parameter , and their edit summaries appear AI-generated (Create article on Karl Pajusalu and add sourced biography, academic career, research, works and honors., Reflecting marriage name change and birth of children.), but there aren't many AI signs, and the articles I checked for WP:V (the first two IIRC) checked out. Kowal2701 (talk, contribs) 16:17, 21 August 2026 (UTC)

it would be great to hear something from you, @PanoptikEU. could you tell us if you used AI for these articles? ‑‑gurkubondinn 16:49, 21 August 2026 (UTC)

Thank you for the ping, and sorry for the delayed response.

I used AI assistance to translate material from Estonian Wikipedia into English. I personally reviewed the articles and checked their sources before publication. My aim was to make information about Estonian scholars, artists and cultural figures more accessible to English-language readers.

I would like to correct the statement that these articles were published without human review. That does not describe my process. I cannot reconstruct every individual drafting step now, so I am not claiming that every passage was solely a translation, but I do remember reviewing the work and checking the sources. If a factual or citation error survived that review, I take responsibility for correcting it.

I would ask that the translated material be assessed under WP:LLM-assisted translation, with any other additions considered separately. Thank you to Jishara for finding the archived Soosaar reference, and to Kowal2701 for checking some of the articles and reporting that those source checks passed. I appreciate those checks as well as the concerns other editors have raised.

M kuhner, could you identify the particular passages and source mismatches you found in Avo Keerend? I would like to respond to those examples directly and correct any problems, including missing translation attribution where applicable.

I would prefer to preserve properly sourced, compliant material and make targeted corrections where needed, rather than reduce all the articles to one-line stubs. PanoptikEU (talk) 13:19, 6 September 2026 (UTC)

Thanks @PanoptikEU for the response. Really I should've thought to check whether these were translations, apologies. Translations from another Wikipedia need to be attributed (see WP:TFOLWP), can you please do that for your creations now? Another issue I'm seeing is how Avo Keerend has content that wasn't in the Estonian article, as does Jaan Elken (Estonian article) and likely others. Can you please make sure that wherever this is the case, the content adheres to WP:V and WP:WTW/WP:NPOV? Otherwise it'd have to be removed as it doesn't fall under the exception in WP:NOLLM for translation Kowal2701 (talk, contribs) 14:27, 6 September 2026 (UTC)

I did not take notes, unfortunately. Here are a few issues.

The lead is sourced to (1) and (2). Here is the key sentence: He was known in post-war Estonian graphic art especially for his linocuts, book illustration, ex libris work, and later geometric-abstractionist prints.

(2) is a list of exhibitions. It mentions three joint exhibitions that had bookplates ("ex libris work") and book illustrations, but I don't think we can go from that to "was known especially for".

(1) does not mention book illustrations, bookplates, or abstractionism. I also can't get the timeline implied above out of this source.

The paragraph starting "During World War II" has a lot of very close paraphrase of (1). However, In 1943 he studied at a drawing studio for war correspondents in Berlin. is sourced to (1) and (3) but (1) has nothing about this at all and should not be cited here.

"Art-historical scholarship places" and "has been connected by museum curators" are WP:AIATTR.

The paragraph starting "The Art Museum of Estonia" is a too close paraphrase of (1). I stopped here as I think the pattern is clear.

I would describe these as typical issues for recent LLMs: the claims are often more or less true, but not supported by the given source, and there is a lot of too-close paraphrase. M kuhner (talk) 05:22, 9 September 2026 (UTC)

I think converting into one-line stubs is much better. ~2026-46891-52 (talk) 15:08, 28 August 2026 (UTC)

User:474D6F756C617264

 474D6F756C617264 (talk · contribs · logs · block log) (13 August 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. Needs to be steered away from pasting LLM output and towards constructive contributions.

Has created LLM-generated articles, some of which may be unattributed machine translations. Some of the smoking guns are AbuseLog/44862535 and AbuseLog/44862596.

Seems to be someone that wants to contribute constructively, they just need to do it themselves instead of pasting LLM output into the project. ‑‑gurkubondinn 12:07, 13 August 2026 (UTC)

Okay, I have taken note of the issue; I have created the necessary pages in my drafts and will rework their format. I will go through the "submission for review" process before publishing anything new.

My apologies for the wasted effort. 474D6F756C617264 (talk) 14:02, 13 August 2026 (UTC)

I have no idea what the best solution to this problem is, but this revert by @Zackmann08: at Tetraclass Model (edit | talk | history | protect | delete | links | watch | logs | views) changes the subject of the article from Tetraclass Model to a duplicate of Melk concentration camp (edit | talk | history | protect | delete | links | watch | logs | views) (also created by 474D6F756C617264). The obvious solution would be to make Tetraclass Model about Tetraclass Model again, but I'm wary of restoring anything questionable. So draftify? FDW777 (talk) 14:13, 13 August 2026 (UTC)

The problem with draftification is that the article is over 90 days old at this point. But this might be different if it's done in cooperation with the user that created it?

I have two questions though. Why did the page history contain an article on a different subject, and was any of those article AI-generated? I'm encouraging a good faith answer and reply from @474D6F756C617264 here. ‑‑gurkubondinn 14:27, 13 August 2026 (UTC)

Hello,

I apologize for complicating your work. Some time ago, I published an article about the Melk concentration camp based on my draft.

Later, using that same draft, I created the article on the tétraclasse model. @Zackmann08 restored information regarding the Melk camp, where my grandfather was deported. I am not quite sure of the proper procedure, and I certainly do not want to commit any act of vandalism. I received a comment on my talk page about this from @Quadrupletime. What should be done? Should the "tétraclasse" page be deleted? I could recreate it in the future to include the information on the "tétraclasse" model. As it stands, the situation cannot continue; the page title bears no relation to its content. Please accept my apologies for the inconvenience caused by this review, and thank you in advance for your understanding.

474D6F756C617264 (talk) 14:55, 13 August 2026 (UTC)

@Gurkubondinn it's been in mainspace for less than two days, you have misread the history. FDW777 (talk) 15:00, 13 August 2026 (UTC)

@FDW777, fair point and an entirely correct diagnosis. I've indeed misread the page history. Draftification seems both appropriate and correct then! It can always be moved back after it has rewritten with without using AI and this has been fixed. ‑‑gurkubondinn 15:13, 13 August 2026 (UTC)

@User:Gurkubondinn I would suggest you edit the above comment. There's an unintended second meaning which implies that you want 47D6 to fix their article with AI, and then move it back to mainspace. Somepinkdude (talk | contribs), in solidarity 22:11, 13 August 2026 (UTC)

Yikes, thats an unfortunate typo! Thanks for pointing that out, Somepinkdude, I have fixed and clarified it. ‑‑gurkubondinn 22:18, 13 August 2026 (UTC)

For my part, I mentally corrected it while reading. I have to say, it made him smile. 474D6F756C617264 (talk) 00:41, 14 August 2026 (UTC)

The 90 day limit doesn't apply in all cases. Countless articles are created every day, more than new page reviewers can keep up with. If anything meeting speedy-deletion criteria (unambiguous promotion, attack pages, hoaxes, patent nonsense, AI hallucinations, AI communication to the user, AI markdown indicators, etc.) manages to survive more than 90 days, it's fair game to delete when discovered, regardless of age. Draftification is a milder alternative, and something I do instead of speedy-delete if it looks like the topic could be notable but not currently up to mainspace standards. That's why so much AI slop gets draftified rather than outright deleted. ~Anachronist (who / me) (talk) 20:56, 14 August 2026 (UTC)

noting User talk:474D6F756C617264#Your submission at Articles for creation: Tetraclass Model (August 23) Kowal2701 (talk, contribs) 16:44, 13 September 2026 (UTC)

The draft still looks like LLM. The editor did a rewrite of one section and resubmitted, but the LLM text remains. I will give a warning. M kuhner (talk) 23:53, 15 September 2026 (UTC)

User:DJ71park

Cleanup efforts are ongoing and being tracked at /2026-09-16_DJ71park, 106 pages need review.

DJ71park (talk · contribs · deleted contribs · logs · filter log · block user · block log)

I believe DJ71park has been using AI quite a bit. Beyond the accusations on their talk page, including discussions of AI art usage, their writing style feels quite consistent with LLMs, and the way they source a lot of their long stretches of writing exclusively with articles from newspaper.com in some places feels particularly odd with how the information within the newspaper articles has been expanded upon for the information in the articles. I noticed these oddities while going through the page for L.W. de Laurence, where I believe it is somewhat clear that AI was used. I am making this submission because I am somewhat new to Wikipedia editing, so I am not fully confident in my ability to label this as AI myself, and I am not entirely sure from the Wikipedia resources I have read how exactly to make the necessary changes if others concur that problematic AI use is in play here.

TThorn1 (talk) 22:37, 17 August 2026 (UTC)

Thanks for the report! (May 2025) had utm_source=chatgpt Operation Fitzwilliam (Feb 2025) has loads of WP:AISIGNS. I have no idea what was going on with their latest creation Henry Clay Hodges (industrialist) (June 2026). Have given them a final warning (cleanup will still be needed) Kowal2701 (talk, contribs) 22:57, 17 August 2026 (UTC)

Thank you for checking! I'm probably going to try to clean up the L. W. de Laurence page. I couldn't find a guide to cleaning up AI text, so if you have one please let me know. I don't know if common practice is to simply rewrite things, destroy everything and start anew, or what else, so I'd appreciate some pointers if that's not inconvenient. I do think I'll for sure have to do some significant reformatting as the problematic sections read to me more like video essays paraphrasing and exaggerating a Wikipedia page rather than a Wikipedia page. TThorn1 (talk) 23:11, 17 August 2026 (UTC)

There's WP:AICGUIDE, but yeah, basically remove/rewrite or try to keep the basic facts, attempting to preserve LLM-generated is usually very time-consuming. This link is all their significant edits to that page, let me know if you have any questions or need help with anything (and also if there's anything in the guide that's unclear), thanks Kowal2701 (talk, contribs) 23:46, 17 August 2026 (UTC)

I recommend picking out the three or four best sources, reading them carefully (you cannot trust a word of what LLMs have said about them), and cutting the article down to basics. You want the three solid sources so that it doesn't end up being deleted, but the bulk of the LLM text is best removed. I have fixed these, and I'm currently walking someone through fixing their own, and it's kind of awful--the issues are subtle and pervasive and honestly it's easier to write new text.

If you look at the Syzygy Rebel case on this noticeboard, you'll see what happens when someone is sure they can just tweak things and solve the problem. For some of their articles we've gone three rounds of me calling out source/text mismatches and them rewriting again. I am positive it would have been easier to reduce to a stub and rebuild, but I couldn't convince them of that. M kuhner (talk) 00:08, 18 August 2026 (UTC)

That makes a lot of sense to me. I reckon I'll start by wiping most of everything from the LLM and research to confirm the essentials (times of birth/death, that kind of stuff) and then redo (or reinstate old) notable information on his plagarism and overall career path, and finish by writing a section or two about a few major incidents I think are still necessary to retain in the article due to their significance to his legacy.

Upon reviewing the article again, I realized without the AI fluff, most of the huge article can be brought down to a few paragraphs probably. There's a lot of detail and superfluous descriptions that work fine enough for a story, but are pretty unrelated to the subject of the article itself, and would make much more sense in an article about the cults he founded or about the police raids on them rather than on his page itself.

I may also go back eventually to use those old newspaper sources and create a new main article on those events, or just write an article somewhere else, because there is some really interesting information taken from the newspapers that is nigh impossible to find any information or a primary source for on the internet without specifically going through old newspaper records. Failing to find any source discussing the Order of the White Willow initially is what drew my attention to the fact that the text may be dubious in the first place (the information is technically correct in the LLM article, it's just not placed next to the proper source currently since DJ71park's later edits shifted the sources over to the incorrect place).

Your advice is going to be immensely helpful, thank you! TThorn1 (talk) 00:30, 18 August 2026 (UTC)

It's we who should thank you! We deal with dozens to hundreds of articles a day, and seldom have the resources to do anything more elegant than ripping out the bad text or deleting wholesale. It's really nice to have someone who can give an individual article the care it deserves. M kuhner (talk) 00:34, 18 August 2026 (UTC)

Citing Wikipedia and played a pivotal role in AbuseLog/40059058, early 2025. Probably ChatGPT, and utm_source=chatgpt.com shows up in some edits. ‑‑gurkubondinn 01:13, 18 August 2026 (UTC)

Bumping thread. Kowal2701 (talk, contribs) 21:09, 13 September 2026 (UTC)

They started editing Jan 2025 and your first log above Feb 2025. I'm just making a tracker for the whole period. They have not edited in mainspace since early August but are building something in user space; at first glance it didn't look bad. M kuhner (talk) 02:29, 16 September 2026 (UTC)

June creations need some analysis to justify LLMPRODs, they're too unorthodox for me to do rn Kowal2701 (talk, contribs) 13:39, 16 September 2026 (UTC)

User:BiscuitsBeforeBias

LLMPRODs are done and drafts tagged

 Cleanup has been requested and is being tracked at /2026-09-16_BiscuitsBeforeBias, 83 pages need review.

BiscuitsBeforeBias (talk · contribs) created Draft:HMS Pentstemon (K61) (originally in mainspace) with the red flag of The subsequent history of Rosa Vlassi is not sufficiently well documented in the sources used for this article to establish her ultimate fate with certainty at the end. FDW777 (talk) 04:57, 19 August 2026 (UTC)

Editor claims here that It was run through a AI polishing route no more, which is somewhat at odds with the sentence I mentioned, as well as the broken reference formatting in the article's initial version. FDW777 (talk) 22:31, 19 August 2026 (UTC)

edits in July and August 2025 (when their articles were created) had markdown Kowal2701 (talk, contribs) 18:33, 23 August 2026 (UTC)

Editor claims at User talk:BiscuitsBeforeBias#August 2026 and here, that their AI use was for polishing and denies broader use, which isn't particularly credible given the sentence I mentioned, red-linked category, broken references (which they didn't have at a previous creation) and a couple of additional things. Declined to give a fuller explanation when asked, claiming the sentence I mentioned was added by mistake. FDW777 (talk) 19:43, 26 August 2026 (UTC)

Bumping thread. Kowal2701 (talk, contribs) 21:20, 13 September 2026 (UTC)

It looks to me like the issue goes all the way back. Certainly by July 2025 one can see AISIGNS in text and edit summaries very different from article text. I am making a tracker for the whole thing. M kuhner (talk) 05:06, 16 September 2026 (UTC)

will wait a week for LLMPRODs Kowal2701 (talk, contribs) 13:23, 16 September 2026 (UTC)

week waited Kowal2701 (talk, contribs) 21:04, 23 September 2026 (UTC)

User:Mellangoose

LLMPRODs are done and drafts tagged

 Mellangoose (talk · contribs · logs · block log) (19 August 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Cleanup efforts are ongoing and being tracked at /2026-09-27_Mellangoose, 44 pages need review.

They created Thomas Hawkins of Trewithen with a markdown codeblock in Diff/1370165057 today, which has since been G15 deleted (could ask an admin to undelete it if needed). I asked them about their AI use in § August 2026 on their talk page, but they haven't edited since. In the meantime, Ike Lek has raised concerns about the Hawkins Family of Trewithen article which was also created today, so I'm creating this thread before I loose track of this and notifying Mellangoose of it. ‑‑gurkubondinn 21:23, 19 August 2026 (UTC)

Mellangoose has replied with Diff/1370310637 in User talk:Mellangoose § August 2026, but not quite committed to stop using LLMs on Wikipedia. I've asked them to reply here on the noticeboard. ‑‑gurkubondinn 13:33, 23 August 2026 (UTC)

Bumping thread. Kowal2701 (talk, contribs) 21:21, 13 September 2026 (UTC)

Created Treludderow and Walter Borlase within about 10 minutes of each other in Diff/1370140022 and Diff/1370141324, one has AIREDCATS and the other has AIBOLD and messed up citation templates. This was before this thread was created, because the user seems to have stopped editing when confronted about their AI use. ‑‑gurkubondinn 11:19, 14 September 2026 (UTC)

Special:AbuseLog/39931319 (Feb 2025) cited Wikipedia, will do analysis for LLMPRODs in a bit Kowal2701 (talk, contribs) 22:48, 23 September 2026 (UTC)

Start date wrt creations appears to be Caving in the Balearic Islands (4 August), it isn't a translation and has AIREDCATs, AIATTR + Pangram partial-positive . There's Voluntary tubal opening (21 July) which appears to be an LLM translation, creations prior appear to be human-written or non-LLM machine translations (errors and hyphens instead of dashes). LLMPRODing from 4 August, start date for normal edits may be earlier (there's a gap between 31 March and 21 July for creations) Kowal2701 (talk, contribs) 10:51, 24 September 2026 (UTC)

I think April 18, which starts having edits very close together and has a bold list. Making tracker accordingly. M kuhner (talk) 04:22, 27 September 2026 (UTC)

Also the "should be treated as X rather than Y" didactic disclaimer/negative parallelism combo common in 2026 LLM text Gnomingstuff (talk) 02:07, 28 September 2026 (UTC)

User:M.A.LasTroniN910

 Cleanup has been requested.

• M.A.LasTroniN910 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

High-volume AI edits/article creations (may be AI translation). Usual slate of AI signs (first time seeing "the sources reviewed" in the wild). Seems to have started August 2026 after an editing break Gnomingstuff (talk) 03:36, 21 August 2026 (UTC)

What a ferocious rate of article creation. :( I share some interests with this editor—it would be a shame to ‘lose’ the articles they’ve created from enwiki if they are AI-generated (putting lose in quotation marks, since we don’t really have them in that case). If it comes to cleanup, I will try to help out here. Suðurhafsljósæta (talk) 07:37, 21 August 2026 (UTC)

This editor has continued editing without responding here, although M.A.LasTroniN910 did acknowledge the warning on their talk page. @M.A.LasTroniN910 can you please clarify how you used LLMs? Helpful Raccoon (talk) 02:09, 24 August 2026 (UTC)

Just want to note Wikipedia:Administrators' noticeboard/IncidentArchive1132#User:M.A.LasTroniN910 and copyrighted plot descriptions and Wikipedia:Administrators' noticeboard/IncidentArchive1155#M.A.LasTroniN910 adding copyrighted material to articles past last warning. GreenLipstickLesbian💌🧸 02:39, 24 August 2026 (UTC)

Hey, sorry for the late reply! Regarding LLMs, I only used it to help me find sources. Everything else, including reading and evaluating the sources and gathering the relevant information, I did myself through careful research and reading. Since the sources were in other languages, I only used it to help locate them. I would never use AI-generated material directly in an article.

Also, just to clarify, I’m now an experienced editor when it comes to manga-related articles. I’ve been working on Wikipedia for years. The mistakes that GreenLipstickLesbian pointed out were things I made early on when I was still learning, and I took those criticisms seriously and learned from them. It’s not that I’m still making the same mistakes today.

You can check my newer articles and their sources yourself and see that I’ve done the research and source verification carefully, with the goal of making sure they meet Wikipedia’s standards for reliable sourcing and GA-quality articles. M.A.LasTroniN910^t@lk 12:23, 24 August 2026 (UTC)

You didn't use LLMs at all to write the text in Chen Dong (writer) (created just over a week ago)? It contains both WP:AIATTR (e.g. Contemporary profiles have described him as a Beijing native and noted his interest in wuxia fiction from childhood) and WP:SUPERFICIAL (He was among ten web novelists recognized that year, reflecting the increasing institutional recognition of online literature within China's literary organizations). Helpful Raccoon (talk) 19:45, 24 August 2026 (UTC)

You were right to raise the concern. I went back through the Chen Dong article carefully and reviewed the wording and sourcing more closely. English is not my first language, and I mistakenly thought some of the phrasing was natural and appropriate for Wikipedia when, in retrospect, it was too indirect and at times overly interpretive.

I accept that mistake, and this has been a useful lesson for me. I have been going back through the article and removing or rewriting phrases that sound generic, formulaic, or more interpretive than the sources support. I’m also being more careful to distinguish between what a source actually says and what can reasonably be inferred from it.

Thank you for pointing it out. I appreciate the criticism, and I’ll take more care with the wording and sourcing going forward. M.A.LasTroniN910^t@lk 04:49, 26 August 2026 (UTC)

@M.A.LasTroniN910, do you speak Japanese or Mandarin? Kowal2701 (talk, contribs) 22:56, 13 September 2026 (UTC)

Japanese. M.A.LasTroniN910^t@lk 01:23, 14 September 2026 (UTC)

several of your creations are based entirely on sources in Mandarin, how are you ensuring that WP:V is met? Kowal2701 (talk, contribs) 10:02, 14 September 2026 (UTC)

It’s true that the sources are in Mandarin, but I manually checked every single source carefully and verified the information before using it. I spent a considerable amount of time reviewing them to make sure the claims in the articles were properly supported.

I did use LLMs to help locate potential sources, and I acknowledge that I initially moved too quickly when creating some of the articles. However, I have since reviewed the sourcing much more carefully, and as far as I’m aware, all of my newer articles now comply with WP:V.

Please feel free to review the sources as well. If you find any unsupported claim or an issue with a source, please let me know and I’ll go back and check it again. M.A.LasTroniN910^t@lk 15:19, 14 September 2026 (UTC)

If you can't read Mandarin, how did you check the sources verified the information? Kowal2701 (talk, contribs) 15:21, 14 September 2026 (UTC)

When I open the source, I use a Google Translate browser extension to translate the page and view the English version. Then I carefully check the translated information against the original source to make sure I’ve understood and verified it accurately. M.A.LasTroniN910^t@lk 08:50, 15 September 2026 (UTC)

@M.A.LasTroniN910 You are essentially using LLM to check LLM then? Moreover to use LLM under one of the narrow WP:NOLLM requirements, then you need to know both the source and end language well enough to be sure that the translation is suitable for Wikipedia. Moreover, strictly speaking this is just for inter-language Wiki translations, not general sourcing. ChrysGalley (talk) 09:10, 15 September 2026 (UTC)

I’m not using an LLM to verify another LLM. I’m using a Google Translate browser extension to translate the Mandarin page into English so I can read and understand the source. I then carefully check the information in the source and cross-reference it with other references where possible.

I’m aware that this does not replace actual proficiency in Mandarin, and I’m not presenting machine translation as authoritative. It is simply a tool I use to understand the source and verify the relevant information. M.A.LasTroniN910^t@lk 09:24, 15 September 2026 (UTC)

"Translating" from a language you do not know is a violation of the LLM-assisted translation rules, and Google Translate is now LLM-powered. —ClaudineChionh (she/her · talk · email) 10:20, 15 September 2026 (UTC)

Then, based on what you’re saying, I understand that I have violated the relevant rules, and that the affected content should undergo cleanup. M.A.LasTroniN910^t@lk 10:30, 15 September 2026 (UTC)

Regarding the Genjo Sanzo article, @Gnomingstuff, you removed all of my edits. I worked on every section, including the Creation section and the Anime casting and reception section. How can all of that be considered AI-generated? Most of the reviews can already be found in the Saiyuki article.

For the Creation section, I read the source PDF myself and wrote the section based on it. I had already mentioned that I can read Japanese. For the reception section, I gathered all the sources myself and manually wrote the content based on them. How can the Reception section be considered AI-generated as well?

Also, this article is about a Japanese character, not a Chinese one. M.A.LasTroniN910^t@lk 12:37, 25 September 2026 (UTC)

Also noting that M.A.LasTroniN910 removed an AI tag from Chen Dong (writer) after making very minor changes. Helpful Raccoon (talk) 02:19, 24 August 2026 (UTC)

User:Cytoscape

LLMPRODs are done, no drafts to tag

 Cytoscape (talk · contribs · logs · block log) (21 August 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Cleanup efforts are ongoing and being tracked at /2026-09-19_Cytoscape, 236 pages need review.

Larger and somewhat rapid edits. There is an chatbot codeblock in Diff/1370565031, which is how I found this user.

They were first warned in § July 2025, then asked about their AI use in § AIs, short descriptions on their talk page in September 2025, but replied to neither message (removed the first warning from their talk page so they can be presumed to have read it). My notice and questions in § August 2026 were also promptly ignored, because they went and pasted more AI-generated text into article space in Diff/1370572260 instead of replying.

@Admins willing to patrol AINB: I believe that a block from article space is required to force communication. I will notify Cytoscape of this thread. ‑‑gurkubondinn 23:11, 21 August 2026 (UTC)

My mistake, read the timestamps wrong. The second edit also came before I wrote on their talk page. Let's see if they communicate. ‑‑gurkubondinn 00:34, 22 August 2026 (UTC)

so... did they respond? | One Reaction was here. Got a complaint? 13:24, 9 September 2026 (UTC)

No. They haven't edited since. Asking for consensus for cleanup. M kuhner (talk) 05:25, 18 September 2026 (UTC)

Support Cleanup There is no constructive way to edit with AI. It is proven incapable of following our guidelines on notability and copyright. | One Reaction was here. Got a complaint? 11:35, 18 September 2026 (UTC)

Support cleanup. Suðurhafsljósæta (talk) 11:53, 18 September 2026 (UTC)

Support cleanup via LLMPRV. ‑‑gurkubondinn 20:20, 19 September 2026 (UTC)

Made tracker from July 1 2025 based on warning: if someone has a better date, let me know. M kuhner (talk) 16:07, 19 September 2026 (UTC)

their first creation is Special:Permalink/1294266197 (6 June 2025) which has markdown for links, loads of red cats in the ones that follow. Confident that's the exact start date, they took a long break before that and their editing is minimal Kowal2701 (talk, contribs) 11:03, 24 September 2026 (UTC)

(@M kuhner) Kowal2701 (talk, contribs) 11:04, 24 September 2026 (UTC)

 Done Updated tracker to start 6 June 2025. M kuhner (talk) 05:03, 27 September 2026 (UTC)

User:FerIow

 FerIow (talk · contribs · logs · block log) (22 August 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

 Cleanup has been requested and is being tracked at /2026-09-25_FerIow, 45 pages need review.

Small potatoes, but should be dealt with early. FerIow pasted AI-generated text to their sandbox page in Diff/1370670820 with a chatbot block and an AIREDCAT, and then removed the tell-tale signs in Diff/1370671811 shortly afterwards. The sandbox edits are related to the 2026 Fagersta school stabbing page, which they had edited shortly before the sandbox and shortly after it. My notice in User talk:FerIow § August 2026 was quickly removed (and the sandbox page was edited to say yeah sure whatever in Diff/1370673183), so they can be presumed to be aware of WP:NOLLM at this point, but they reinstated their changes in Diff/1370672978 with an edit summary denying AI use. I speak Swedish, and both the sources that were added do support the claims but one of them is the school's own website which isn't ideal. While this particular edit on it's own is not a great concern, I think that it is probably better to try to encourage a course correct sooner rather than later before they have made more AI-generated edits to the project. I'm notifying them of this thread. ‑‑gurkubondinn 13:00, 22 August 2026 (UTC)

Bumping thread. Kowal2701 (talk, contribs) 23:03, 13 September 2026 (UTC)

Prodding it a bit more to see if it still moves. | One Reaction was here. Got a complaint? 11:29, 22 September 2026 (UTC)

I sent a warning to test if they stopped. They can clearly see their talk page, as said warning was removed by them this morning. | One Reaction was here. Got a complaint? 16:31, 23 September 2026 (UTC)

@OneReaction5890, can you stop sending them warnings please? They're not required to respond if they don't continue editing Kowal2701 (talk, contribs) 12:58, 24 September 2026 (UTC)

Stopped sending warnings in accordance. They seem he have stopped, but that doesn't mean I'm taking my eyes off just yet. | One Reaction was here. Got a complaint? 12:53, 25 September 2026 (UTC)

Made a tracker to spot-check for issues; don't know whether cleanup is required yet. M kuhner (talk) 21:22, 25 September 2026 (UTC)

Dionysodorus found Klaus Fehling created Jan 2025 as an unacknowledged translation from German Wikipedia, but with a totally wrong title (was called Andre Erlen). Dunno if that is an LLM error but it is certainly problematic! M kuhner (talk) 21:27, 25 September 2026 (UTC)

Article titles for German sources have been translated into English. Pretty sure this is unreviewed machine translation; don't know if it's LLM but the human does not seem to have been in the loop. M kuhner (talk) 21:30, 25 September 2026 (UTC)

User:Alfiechidzik denying blatant LLM usage

I just declined Draft:Instro Precision by User:Alfiechidzik, which has numerous instances of both WP:AIATTR (Corporate filings show that ownership of Instro Precision is held by Elbit Systems UK Limited, which controls the company through a majority shareholding) and WP:AITREND (The acquisition formed part of Elbit's wider expansion within the United Kingdom defence sector, These protests formed part of a wider campaign directed at Elbit Systems UK subsidiaries and facilities across the United Kingdom). Their talk page response was to say they didn't use LLMs. Alfiechidzik has also expanded and created several other articles in the WP:A-I topic area, which also show AISIGNS, e.g. "Statements from the Belgian government and media reports identified one of the seized shipments as coming from Moog's Wolverhampton factory and bound for Israel's M-346 program". Helpful Raccoon (talk) 00:02, 24 August 2026 (UTC)

I have not formed an opinion at the moment on whether the draft is AI, but I note for the record here that this editor has previously admitted to using AI to revise article text and committed to not doing it anymore. ^purring_maggot (talk) 04:07, 24 August 2026 (UTC)

The draft was submitted a month before the ANI discussion, but the Elbit Systems edit I highlighted was weeks after the discussion. Helpful Raccoon (talk) 04:21, 24 August 2026 (UTC)

Ahh, I totally missed that the draft was a few months old. Sorry, I got distracted by what looks like a bunch of copyvio images this editor uploaded to Commons. ^purring_maggot (talk) 04:26, 24 August 2026 (UTC)

(Parts of the Elbit Systems edit appear to be copied+modified from the bottom of Draft:Instro Precision) Helpful Raccoon (talk) 04:28, 24 August 2026 (UTC)

You left a couple of comments but then deleted them so not sure if this is the right place to post a reply.

You're absolutely right to call me out on the on the use of AI for wording back in June, as I admitted to it, and have taken the feedback on board and not used it since. The Instro Precision article draft does date from that period, so I can understand why you flagged it, and on reflection it makes sense to delete it so that it can be rewritten from scratch so that any future draft is untainted by AI concerns.

What does concern me though is that the cited instances in this case of problematic wording is actually just not AI usage, e.g. "Statement from the Belgian government and media reports ..." I'm struggling to think how I can express that differently other than directly quoting RTL and RTBF (which runs the risk of over attribution) or quoting directly from Walloon documents (which is primary research). I'm not trying to be difficult here - I'm just genuinely at a loss. Similarly with "Press commentary before the trial ..." - should I be directly quoting the specific sources in this instance?

Anyway, I apologise for my rather hasty reaction - I'm just tired and irritated as I've been working all weekend. Alfiechidzik (talk) 09:36, 24 August 2026 (UTC)

Sorry, I originally misinterpreted your below comment. Yes, it's more plausible that the Belgian text and later text is not LLM-generated; it seems more like you picked up similar habits just from reading LLM output. Overattribution here is unnecessarily attributing claims to sources when they can typically be omitted, especially vague sources like "media reports" which should at least be made more specific. Helpful Raccoon (talk) 10:03, 24 August 2026 (UTC)

See WP:INTEXT for when in-text attribution should be used. Helpful Raccoon (talk) 10:14, 24 August 2026 (UTC)

OK - that's helpful to know. You're right that it's a habit that i've picked up - use the specific citations as footnotes and then refer to them as "press commentary" rather or "media reports" in order to avoid having to either repeat the specific attribution in the text, or assert it as fact.

I've read WP:INTEXT although the examples it gives are rather blatant extremes so not that helpful when dealing with nuanced, middle-ground scenarios. I know what I would write if I was writing an academic paper, but that seems to be putting too much of an authorial voice - and potentially non-neutrality - into the article which is also in contravention of Wiki guidelines.

In any case, I will try to be more thoughtful and more specific when referring to sources inside the text itself going forward. Alfiechidzik (talk) 10:20, 24 August 2026 (UTC)

"Press reports" or "government statements" just add nothing. You might as well say it in Wikivoice, with citation, because the reader can't use what you're saying to determine degree of uncertainty. In-text attribution is only worth doing when it carries additional information, and "press"/"government" are far too vague to do that. (And to anyone with a nose for LLM, it's a major indicator which you really don't want to copy.)

An example that actually carries information would be "Peruvian press reports said X [1][2][3] while Venuzuelan reports said Y [4][5][6]". This helps provide context for the reader. Be sure there are actually multiple citations in both sets, though, or you're back at WP:AIATTR. (And be careful of WP:SYNTH.) M kuhner (talk) 17:56, 19 September 2026 (UTC)

Hmm, Alfiechidzik's more recent creation Moog 4 doesn't have the same numerous AISIGNS. There is WP:AIATTR with Press commentary before the trial raised concerns that... that isn't copied from anywhere. There's also Belgian media reported that the cargo was inspected... but that seems copied from Draft:Instro Precision. Helpful Raccoon (talk) 05:01, 24 August 2026 (UTC)

Im struggling to see what the AI signs are with regard to the Belgian text. I literally looked up the RTL and RTBF articles myself having been sent them by a colleague in Belgium, and likewise the Middle East Eye, Declassified and Private Eye reporting - the Private Eye reporting isn't even on the internet and I had to physically go and buy paper copies of the magazine to get the citations for these. I even emailed the Wallonian government press office myself to get their own confirmation of the facts and what they publicly released so that I could confirm this for myself (as there appears to have been some initial confusion about exactly what was seized). Alfiechidzik (talk) 07:59, 24 August 2026 (UTC)

User:Cristõvão Campos

 Cristõvão Campos (talk · contribs · logs · block log) (25 August 2026)
Editor has been blocked from editing the mainspace (articles only) for AI/LLM violations. 

Large rapid AI-generated edits, article creations that might be unattributed translations, lots of markdown. Mostly ignores talk page messages or notices, and just keep going instead. Has occasionally replied to the CS1 notices posted by the bot, asked for help (in somewhat broken English) once, and apologised for contributing in the wrong language once. I've draftified one of their more recent creations a reverted large changes on one article, but they just ignored me and carried on like nothing happened.

My guess is that this is someone without a sufficient command of English to understand what people are trying to tell them. ‑‑gurkubondinn 00:41, 25 August 2026 (UTC)

These are 100% unattributed translations. JTtheOG (talk) 19:36, 29 August 2026 (UTC)

Still going, created Diogo Valsassina in Special:Diff/1374472988 yesterday. It's already been sent to AfD by JTtheOG: Wikipedia:Articles for deletion/Diogo Valsassina (just missing the WP:DSAIGEN deletion sorting). ‑‑gurkubondinn 12:49, 13 September 2026 (UTC)

Also, does Bronco test look human-written to you? JTtheOG (talk) 19:30, 13 September 2026 (UTC)

yes, see Kowal2701 (talk, contribs) 23:17, 13 September 2026 (UTC)

Pretty sure that you meant to say "no", as in "yes this is AI-generated". ‑‑gurkubondinn 10:41, 14 September 2026 (UTC)

Y Kowal2701 (talk, contribs) 12:05, 14 September 2026 (UTC)

created Beverly Randolph on 18 Sept, currently at AfD, is mostly unsourced and several potential BLP vios. They've not responded to any of the last batch of warnings and messages, in fact they've only ever made 8 edits to their own talk (none to article talk pages) which are all very short, mostly saying "thank you" or "sorry" except for Special:Diff/1325986424, @Admins willing to patrol AINB: I think a mainspace block on CIR grounds may be warranted? Kowal2701 (talk, contribs) 13:12, 24 September 2026 (UTC)

Blocked from mainspace. ~Anachronist (who / me) (talk) 14:43, 24 September 2026 (UTC)

Paulo Vilhena created November 23, 2025 is an unattributed translation from the Portuguese Wikipedia, and clearly has not been reviewed as it alternates referring to the actor as "he" or "she." However it is otherwise very close to the Portuguese original so probably not LLM; I'd have expected LLM to embroider more. I will edit-summary it as a translation. M kuhner (talk) 14:47, 28 September 2026 (UTC)

User:Thomson Walt

 Cleanup has been requested.

 Thomson Walt (talk · contribs · logs · block log) (22 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Thomson Walt (talk · contribs · deleted contribs · logs · filter log · block user · block log)

The article that first caught my attention was Pu xie qi yao. The tone of the article and the sourcing seemed irregular, but not a degree I could say with confidence it was AI. Looking at this editor's recent contributions, they have made a lot of large content additions in short spans of time recently, and while sometimes I myself wait to release big edits on similar topics at the same topic to help with page linking, the back-to-back edits this user makes on the same page at times lead me to believe these aren't saved edits being pushed out at once. As this editor reports being a native speaker of Isan, I think it is plausible there is some AI translation going on here; however even then I am skeptical of it being handled responsibly. I have already left a notice on the user's talk page about their marking of every edit as minor. I also thing the peculiar way they cite page numbers could be a human error, but warrants looking into. @Gurkubondinn and @Athanelar both weighed in on this while I was deciding whether to bring this to the noticeboard, so I'm pinging them so they can share their thoughts. – Ike Lek (talk) 15:39, 26 August 2026 (UTC)

My thoughts (and Anthanelar's) are on User talk:gurkubondinn § You free to give me a second opinion here?, can't say for sure if this is AI-generated or not, though I'm pretty sure that it is involved one way or another. The rapid large edits are concerning for sure, and this may be someone that isn't aware of the NOLLM guideline. ‑‑gurkubondinn 16:38, 26 August 2026 (UTC)

On the fence here, I think translation is the most likely here. The phrasings are a little too idiosyncratic.

Special:PermanentLink/1046670519 is an example of their older writing style, it's different from the comparable 2026 article Special:PermanentLink/1368964775 but not drastically so. Gnomingstuff (talk) 07:10, 27 August 2026 (UTC)

Special:AbuseLog/43164411 had utm_source=chatgpt, and they're making large edits at an implausibly fast rate. @Thomson Walt, can you please respond here? Kowal2701 (talk, contribs) 13:06, 2 September 2026 (UTC)

still going, some 1000+ byte edits were in less than a minute Kowal2701 (talk, contribs) 17:41, 7 September 2026 (UTC)

@Admins willing to patrol AINB: could we get a mainspace block to stop the flood? Editor has not responded and is actively and rapidly editing. M kuhner (talk) 03:39, 22 September 2026 (UTC)

looking⇒SWATJester ^{Shoot Blues, Tell VileRat!} 04:05, 22 September 2026 (UTC)

I'm not seeing anything more recent than 3 days ago on their talk page, or anything beyond pointing them to this noticeboard conversation; and nothing about the edits I saw on the most recent contributions (I spot checked 3x edits that were over 1000 bytes added) immediately screams LLM. The rate of editing isn't static, and isn't so fast (again unless I'm missing something) that it couldn't be human. They most certainly were using them before, but I'm not sure I can justify a mainspace block based on their behavior *today* unless I'm missing something. ⇒SWATJester ^{Shoot Blues, Tell VileRat!} — Preceding undated comment added 04:11, 22 September 2026 (UTC)

Hi everyone. I want to be open about this. Yes, I do use AI/LLMs as part of my workflow. As I’ve mentioned on my user page, my English is around an intermediate level, so I rely on these tools to help me translate source materials and clean up my grammar, wording, and consistency. Because I use a mix of translator tools depending on the usage limitations and context, my writing style might shift over time. To clear up the confusion about the specific article mentioned above. It’s about a ruler found in a Chinese text. Previously, the info was mixed up with another figure from the Thai chronicles, Siddhijaya Brahmadeva. Since there isn't any established research linking the two, I separated the material and translated it from the Thai paper. The article was initially a bit messy, but I've gone ahead and fixed the issues mentioned.

And why I usually add over 1,000 bytes so quickly. That's just how my usual workflow goes. When I read a research paper, I usually extract the relevant points and apply them to several related articles at once. For example, while reading Michael Vickery’s paper on the old cities of Chaliang, Si Satchanalai, and Sawankhalok, I found sections discussing the different historical debates around these locations. This information directly applied to three articles, its ruler Sri Naw Nam Thum and the sites Mueang Chaliang & Si Satchanalai Historical Park. Once I’ve fully processed the source, I can easily jump between the tabs and make those edits within about ten minutes. Since each edit needs its own context and full citation, an edit can naturally exceed 1,000 bytes. I usually repeat this process for every relevant point I find in each paper. On working days, I might get through about 3 papers, but on my days off, I can sometimes manage up to 10. About that utm_source=chatgpt parameter. I actually use the tool like a search engine to hunt down specific papers. Some sites automatically track where you came from and tack on those UTM parameters to the link. That tag just shows how I found the source, not that the content was generated by AI. Hope this helps clear things up. Thank you. Thomson Walt (talk) 15:30, 22 September 2026 (UTC)

User:Cyberkiborg

 Cyberkiborg (talk · contribs · logs · block log) (27 August 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Has created some articles, including the Transparency in the software supply chain article that is currently at AfD: WP:AFD/Transparency in the software supply chain. There's an excellent source analysis in § Verifiability issues on the talk page. The usual rapid editing in f.ex. History/Cyber threat intelligence. ‑‑gurkubondinn 11:53, 27 August 2026 (UTC)

Thank you for your work. No irony intended—thank you. AI-generated hallucinations and the distortion of facts undermine the credibility of Wikipedia articles and, consequently, their purpose. So, as Dreamyshade recommended in the discussion you mentioned, I’ll refrain from using language-checking services and dictionaries (though Deepl, for example, seems to be acceptable as a tool), as well as from searching for sources and editing. I haven’t used LLMs extensively when writing articles; the examples from the analysis you mentioned—the one Dreamyshade did—are more a result of me seeking confirmation of my own conclusions and generalizations. As for “rapid editing”—this is because I compile big edits in Sublime, Google Docs, or Obsidian, and then transfer them to Wikipedia. I also break them down into self-contained sections along the way so that if any comments come up, they can be pinpointed, rather than having to undo my edits to the entire article. Cyberkiborg (talk) 20:10, 27 August 2026 (UTC)

(though Deepl, for example, seems to be acceptable as a tool)

No it isn't.

I haven’t used LLMs extensively

This isn't about how you haven't used them. How have you used them? Gnomingstuff (talk) 15:40, 30 August 2026 (UTC)

I'm sorry, but I'm completely confused after these discussions and your comment—could you help me figure this out? If I understand WP:AITALK correctly,it states that you can use a translator, for example, but notes that it can make mistakes. I interpreted this to mean that you need to carefully verify the results when using it. Is that not the case? WP:LLMS allows for basic editing as long as the text and suggested sources are thoroughly verified. WP:LLM only mentions basic editing and prohibits the use of LLMs for any other tasks. In WP:LLMRESP, broader editing and source research are again permitted under certain restrictions. In fact, I’ve used tools like Google AI or ChatGPT to condense large amounts of my own text, find sources, and provide critical feedback—always verifying everything suggested. Dreamyshade’s comments on the article “Transparency in the Software Supply Chain” are correct, and I agree with every point. But in terms of LLM, they pertain to my own choice of sources, their placement in the text, and the degree of generalization of the information. I have already concluded that it is necessary to avoid getting carried away when presenting content and to present information exclusively in a neutral, concise manner, without generalizations ```` Cyberkiborg (talk) 16:45, 31 August 2026 (UTC)

The "basic copyediting" carveout in WP:NOLLM basically means things your average spell checker could do, not condensing text or "generalization of information" which are definitely functions that sound like they probably contributed to the source-text issues in Transparency in the software supply chain. ^purring_maggot (talk) 00:27, 2 September 2026 (UTC)

The reason for WP:AITALK is better explained by WP:LLMCOMM.

‑‑gurkubondinn 13:34, 2 September 2026 (UTC)

Got it. No LLMs when writing articles or participating in discussions. I apologize for relying on the essay where rules clearly forbid it. Honestly, I’m glad my first attempt failed—and not, say, my tenth—at trying to do something beyond a basic search. The rapid editing was a direct result of my process, where I tend to dump and structure everything in Obsidian first. Cyberkiborg (talk) 16:33, 4 September 2026 (UTC)

Bumping thread. Kowal2701 (talk, contribs) 20:31, 24 September 2026 (UTC)

In case anyone missed it: Diff/1372411489. ‑‑gurkubondinn 17:06, 29 September 2026 (UTC)

Special:Diff/1376870635 (26 Sept) looks suspect Kowal2701 (talk, contribs) 17:43, 29 September 2026 (UTC)

User:AuthorAuthor3

 AuthorAuthor3 (talk · contribs · logs · block log) (27 August 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Created the WildBlue Press article, currently listed at AfD: WP:AFD/WildBlue Press. There is user communication in Diff/1371487695 and Diff/1371505099 (Use code with caution, renders as ... college friend Michael Cordova, Use code with caution. whose wife, ...) and then added faux-inline references in Diff/1371540058 after my notice in § August 2026 on their talk page. Actively editing and has not addressed concerns over the AI use. ‑‑gurkubondinn 12:09, 27 August 2026 (UTC)

I did a search on Google when I looked for articles about the publisher. I didn't realize I was searching AI. AuthorAuthor3 (talk) 18:08, 27 August 2026 (UTC)

Could you tell me what tool you were using to generate your citation templates? ^purring_maggot (talk) 18:54, 28 August 2026 (UTC)

I found and read the sources, and I copied the citation format from other Wikipedia articles. I don't remember which ones. At one point I did a Google search asking if a citation was formatted right, and I've since learned that Google may answer questions with AI. AuthorAuthor3 (talk) 02:35, 31 August 2026 (UTC)

OK, thanks. When you asked Google about the citations, could you have copied wiki code from there? The message "Use code with caution." wouldn't have come from copying templates from another article.

There are loads of tools to generate citations, including right inside the Visual Editor (Help:Introduction to referencing with VisualEditor/2). Personally I use Citer for citations.

The WildBlue article isn't the only cause for concern. For example, this diff was reverted by another editor because it constitutes close paraphrasing from the source, which is considered a type of copyright violation. This is also exactly the sort of output a chatbot would produce. If you've copied text from anywhere else in order to make your edits (forget about the citation templates for now, I'm asking about article text), it would be best if you could let us know. ^purring_maggot (talk) 03:13, 31 August 2026 (UTC)

User:Stardust20

LLMPROD is not needed in this case

 Cleanup has been requested.

Stardust20 (talk · contribs · deleted contribs · logs · filter log · block user · block log) A lot of this users edits have been made using WP:LLM and need cleanup. I noticed it on Rothesay (insurer) and trying to cleanup the slop. Thanks, ~2026-46891-52 (talk) 15:02, 28 August 2026 (UTC)

not triggered any relevant edit filters nor tags and no creations, lots of large edits made implausibly quick, but edits need more analysis. They started editing properly in Oct 2025, at which time they were adding non-existent templates . I'm gonna leave this for someone else (sorry) Kowal2701 (talk, contribs) 23:56, 13 September 2026 (UTC)

User:2006mar

 2006mar (talk · contribs · logs · block log) (29 August 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Created Ahmed Choukri Stadium (Diff/1367158105) and Moulay Rachid Stadium (Casablanca) (Diff/1369115621) recently, both with AIREDCATS. Markdown in AbuseLog/44812363, AbuseLog/44811935, AbuseLog/44811812, AbuseLog/44784948. Attempts to cite Wikipedia in AbuseLog/44805142, AbuseLog/44796124, AbuseLog/44796120, and AbuseLog/44784947.

Has been warned several times, ignored it every time. Was asked about User talk:2006mar § Fake citations by JTtheOG, ignored that too. Asked Problem? in Special:Diff/1365377903 when an AfC draft was declined. Mainly edits Moroccan sports-related articles, so there might be a language barrier at play. ‑‑gurkubondinn 14:58, 29 August 2026 (UTC)

Still editing, recent edits are smaller. Diff/1374407716 introduced incorrect grammar, but didn't see any AISIGNS in the ones that I spot checked. ‑‑gurkubondinn 18:36, 17 September 2026 (UTC)

User:Middleground1

 Cleanup has been requested.

• Middleground1 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

The usual. AI edits going back at least until 2025. Multiple warnings on talk page and no change in behavior. When will it end. Why does no one listen. Gnomingstuff (talk) 15:38, 30 August 2026 (UTC)

Countless filter log special abuse triggered. Can't go unnoticed for long. ~2026-47519-49 (talk) 12:47, 1 September 2026 (UTC)

Yep, loads of AI-sourced citations , creations from 2024 have lots of signs, worth doing a proper analysis (I'll try in a bit if no-one beats me to it) Kowal2701 (talk, contribs) 22:54, 3 September 2026 (UTC)

Special:AbuseLog/39603430 (Dec 2024) is the first chatgpt-derived link. John Theurer Cancer Center (June 2023) mixes the use of emdash and double hyphen, its "Notable accomplishments" section is very promotional but this appears human-written, similar to Filip Swirski (March 2024) except for His writings reflect translational and fundamental cardiovascular and neurodegenerative science research... (Pangram says no ). There's mass creation and lots more signs in late 2025, looks like that may be where the start date is, but more analysis needed Kowal2701 (talk, contribs) 12:42, 14 September 2026 (UTC)

New York State Housing Trust Fund Corporation (Nov 2025) had AIDASH, overreliance on primary sources, AIATTR, Pangram says AI-assisted (idk why, the prose is pretty obvious). They denied AI use and said they run stuff through MS Word, despite having a G15 in August. Recently editing, mostly in sandbox Kowal2701 (talk, contribs) 11:25, 25 September 2026 (UTC)

MeloMaster

Cleanup efforts are ongoing and being tracked at /2026-09-14 MeloMaster, 55 pages need review.

LLMPRODs are done and drafts tagged

 MeloMaster (talk · contribs · logs · block log) (2 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

MeloMaster (talk · contribs · deleted contribs · logs · filter log · block user · block log)

I ran into this person's articles through seeing The Early Spring on the NPP popular unreviewed list, and it had characteristic AI phrasing and source-text integrity issues. This is a bummer because it's a very popular show, so lots of people are reading the article. I cleaned up some of it, but really it needs more work.

The person also made a bunch of other articles that need review for AI issues and likely reworking - see User:MeloMaster/Articles - including Oh Eun-seo, Generation to Generation, Dear X (webtoon), Love for You, Love Between Lines, and Speed and Love. I left a notice on their talk page a few days ago about LLM use, but it looks like they may not have received a warning before that. They haven't edited for a few weeks. Dreamyshade (talk) 04:49, 1 September 2026 (UTC)

LLMPROD needed asap ~2026-47519-49 (talk) 12:44, 1 September 2026 (UTC)

I just spot-checked a couple of their articles and spotted AI signs and source-text integreity issues in them all. I'd support an LLMPROD. InfernoHues (talk) 21:13, 2 September 2026 (UTC)

doing Kowal2701 (talk, contribs) 21:31, 2 September 2026 (UTC)

Stripped "AINB article" lines as redundant with tracker (tracker is easier to use) and had not been updated. M kuhner (talk) 15:19, 25 September 2026 (UTC)

User:Magentic Manifestations

 Cleanup has been requested.

Magentic Manifestations (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Magentic Manifestations has recently created or edited a number of similar pages, and I have noticed several things that seem worth looking into.

Checked some paragraphs of different articles with AI detection tool and each of them flagged as 100% AI-generated. One thing that stands out is that some of the articles were created as a single large edit, with large amount of structured and technical content appearing all at once with a single large first edit. This is also the case with Pyroclastic rocks, Peddapalli. The article contains detailed geological descriptions and explanations from the beginning.

There are also some inconsistencies in the article. Some of the wording appears more detailed than what is supported by the cited sources, and some statements seem to go beyond what the references establish. There are also generic explanations of terms which seem unnecessary for the article and make parts of it read more like a generated summary.

Specialabuse filters have been noticing this since a very long time. Check filter log.

Tagged Pyroclastic rocks, Peddapalli for now ~2026-47519-49 (talk) 12:42, 1 September 2026 (UTC)

If you are making a claim against an editor, then it would be better if you could provide concrete proof of the allegation being leveled so as to address them. New articles are generally created from scratch in one go and nothing is wrong with that. The mentioned article in question (Pyroclastic rocks, Peddapalli) adheres to the MOS and has three clearly sourced paragraphs. You have tagged three other articles as well blindly as AI-generated with no clear reason. So, unless you have something concrete here, I am going to remove the tags and ignore this discussion ever happened. Thanks.

Note: The only edits from the IP address in question seems to have been leveling these allegations. Seems suspect and motivated. Magentic Manifestations (talk) 15:03, 1 September 2026 (UTC)

Ironically, Pangram says that the report is 100% AI-generated, which does indeed seem likely. Kisseran (talk) 15:21, 1 September 2026 (UTC)

Im not seeing any huge signs of LLM use in that article, but it's hard to tell with such a small text. Everything looks attributed correctly, the notability seems suspect but that's not what I focus on. Hooples (talk) 15:49, 1 September 2026 (UTC)

I'm concerned by very close paraphrase of source 1. The archiving prevents my copy-vio software from running on this source, but eyeballing shows near-identical passages of considerable length. For example: source has The outcrop is reported as welded agglomerate consisting rounded to angular, large fragments of granite, granite gneiss, basalt and banded ferruginous quartzite set in a matrix of ignimbrite. The acidic fragments are ripped off from the country rocks during violent eruptions. The rock characteristics indicate ignimbritic pyromagma which has contributed to the glowing avalanche type of volcanic eruption.

In contrast, article has Pyroclastic rocks at Peddapalli comprise a welded agglomerate of various igneous rocks such as granite, gneiss, basalt, and banded ferruginous quartzite, set in a matrix of ignimbrite. [...] These rocks were formed when fragments were ripped from the country rocks during violent submarine volcanic eruptions. The rock characteristics indicate ignimbritic magma associated with a glowing-avalanche type of eruption.

The wording "country rocks" is sufficiently odd to attract notice here.

I am also concerned about the last paragraph, which seems to be a random assortment of close paraphrases from various sources with no clear topic. Why are pyroclastic rocks defined here, and not when mentioned above? What is the use of telling us that they are classified by size and abundance, without venturing any statement about the ones at Peddapalli specifically? Why is tephra defined?--it is never discussed anywhere in the article. (It is sourced to 1 and 4 but the sentence is close paraphrase of 1 and does not really agree with 4.)

The article is short and I don't feel 100% sure it is LLM, but these flaws are quite typical of LLM.

M kuhner (talk) 22:25, 1 September 2026 (UTC)

If you see that please tag it with Template:Close paraphrasing, whether or not it was llm. CMD (talk) 22:41, 1 September 2026 (UTC)

The phrase "country rocks" refers to the rock type as exactly as mentioned in source #1. The geology section follows the structure as given in the source (initial description of the Pedappalli rocks followed later by the definition of pyroclastic rocks and the definition of tephra (which is also stated in source #4)).

So, the problem here is the dearth of sources and that majority of the content derives from a single source (#1 INTACH). If we still want to go by tools, both GPTZero and Quillbot rate this as 100% human.

However, I do agree with CMD that there might be some close paraphrasing given that it was derived from a single source. However, being geology, I often find it difficult to rephrase them, and I am not sure rewording might serve any purpose. I have addressed the coherence part. Anyways open to comments. Magentic Manifestations (talk) 08:51, 2 September 2026 (UTC)

If there is only one usable source, that's not enough for an article. I do sympathize with the difficulty of rephrasing technical descriptions: biologists hit that too, and there is some leeway for close paraphrase in these situations. But I really don't think there are enough sources here. Is there any possibility of finding more? Otherwise it should probably go to Articles for Deletion. M kuhner (talk) 19:35, 4 September 2026 (UTC)

User:Indinfokmr

 Indinfokmr (talk · contribs · logs · block log) (1 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Communication from a chatbot in AbuseLog/42213058, doesn't seem to have stopped using AI. One particle is at AfD, I've sent another one to draft. Both were full of AISIGNS, and the article that I draftfied was created with one reference that returned 404 not. Had never been warned for AI use until today. ‑‑gurkubondinn 15:01, 1 September 2026 (UTC)

looks like language-proficiency issues , I'll leave them a message Kowal2701 (talk, contribs) 23:02, 3 September 2026 (UTC)

didn't respond and are currently editing, but they appear to have stopped using AI (eg. ). However, they've submitted the article you drafted (Draft:Kashmiri Coppersmithing) to AfC while it still has the AI tag... Kowal2701 (talk, contribs) 12:54, 14 September 2026 (UTC)

Went ahead and declined the draft at AfC. They're clearly aware of user talk pages existing because they posted User talk:Qwerfjkl § Question from Indinfokmr on Osvaldoginella gomezi (04:23, 14 September 2026) today.

Indinfokmr, can you please respond to these concerns? ‑‑gurkubondinn 13:06, 14 September 2026 (UTC)

Need consensus for cleanup; doesn't seem like we will get a response. M kuhner (talk) 15:21, 25 September 2026 (UTC)

Support using LLMPRV for cleanup. They're still editing though, but it doesn't look like we'll get an answer. ‑‑gurkubondinn 15:27, 25 September 2026 (UTC)

IamMKO

IamMKO (talk · contribs), XTools

AI-sourced citations , draft declined for AI use , several AI signs in Yoruba Language Center, University of Ibadan (June 2026), seems good-faith, I've left them a message Kowal2701 (talk, contribs) 11:18, 2 September 2026 (UTC)

LLM use by User:Mitchell2526

There are definite indications of LLM use by User:Mitchell2526. Some of this is in Draft:Timothy Cover and Neil Osheroff where there were tags showing that the refs were generated using ChatGPT (since removed). For certain Draft:Timothy Cover has several sources which are both incorrectly formatted and do not verify the statements, strongly suggesting AI hallucination. The user has denied any use of LLM at User talk:Mitchell2526.

Courtesy ping of MadelynnSienna and gurkubondinn. Ldm1954 (talk) 14:59, 2 September 2026 (UTC)

AI-sourced citations and markdown . After the warning they created Special:Permalink/1373055436 (3 Sept), I'm not seeing any signs, the prose appears human but spotchecks of refs and text-source would make certain. Ofc edits before 2 Sept need cleanup regardless Kowal2701 (talk, contribs) 23:14, 3 September 2026 (UTC)

User:Noboyo

 Cleanup has been requested.

 Noboyo (talk · contribs · logs · block log) (14 September 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. Autopatrolled permission has been removed from editor's account.

Noboyo (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Another case of high-volume AI editing, this time to musical articles with the usual slate of music article-specific AI indicators. Also, many hits in filter log for ChatGPT params. Not sure how far it goes back, there's no clear breakpoint in the edit history timing, but the earliest filter log hit for AI params is May 2025 so it can be assumed that all of the (many, many, many) edit since then are also ChatGPT Gnomingstuff (talk) 22:54, 2 September 2026 (UTC)

yep, Special:AbuseLog/40858164 (May 2025) had utm_source=chatgpt, and they triggered Special:AbuseFilter/1325 6 times between Nov 2025 and April 2026 . Creations are cleaner than your average LLM articles and music articles are tough to detect, these probably need spotchecks unfortunately Kowal2701 (talk, contribs) 18:18, 4 September 2026 (UTC)

music articles are actually some of the easiest to detect imo, the formula is incredibly rigid and idiosyncratic Gnomingstuff (talk) 02:06, 5 September 2026 (UTC)

in what ways are they structured differently from human-written ones? I find SUPERFICIAL doesn't really stick out in stubby artistic topics Kowal2701 (talk, contribs) 15:32, 5 September 2026 (UTC)

the album blends whatever and whatever while maintaining blah

lyrically, the album explores themes of whatever

reviewer 1 highlighted blah blah blah, reviewer 2 emphasized whatever

there are other indicators but this stuff shows up over and over in AI articles and almost nowhere else; there is so much of it because at this point our music articles are primarily AI thanks to a handful of individuals Gnomingstuff (talk) 16:33, 5 September 2026 (UTC)

have you got the time to determine a start date here? I don't see any pattern changes in creation timing or activity (btw only 0.8% of their edits are to talk pages)

Also, @Admins willing to patrol AINB: can autopatrolled be removed here please? Kowal2701 (talk, contribs) 18:07, 14 September 2026 (UTC)

Autopatrol revoked. ~Anachronist (who / me) (talk) 22:51, 14 September 2026 (UTC)

User:Utkarsh Nandan Oreya

 Cleanup has been requested.

Utkarsh Nandan Oreya (talk · contribs · deleted contribs · logs · filter log · block user · block log)

User created a number of lengthy articles (between 5,000 bytes and 22,000 bytes) in the last day, all from scratch with no additional edits, sometimes within ~10 minutes of each other. Running a couple of them through GPTZero returned an AI score of 1.0 Giulio 00:38, 4 September 2026 (UTC)

received a final warning this morning , I'll draftify yesterday's creations and leave them a message explaining things. Worth keeping an eye on Kowal2701 (talk, contribs) 18:39, 4 September 2026 (UTC)

their response was positive but potentially AI Kowal2701 (talk, contribs) 18:15, 14 September 2026 (UTC)

Looking at their history, it is very likely that they use LLM extensively to create articles and use AI for responses on their talk page. Dz5t 8O12 (talk) 19:12, 20 September 2026 (UTC)

they submitted Draft:Nrottam Das (poet) with minimal changes which got declined, tbf their response to that was human-written so maybe there's still hope Kowal2701 (talk, contribs) 22:08, 20 September 2026 (UTC)

User:Chiserc

 Cleanup has been requested.

Chiserc (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Came across an article by this user while patrolling, created in a single edit in draftspace with redcats, curly qoutes, and the category Category:Pending AfC submissions in article space. Sourcing with no links to Spanish language journals led me to look at their history.

• An article created in a single edit in draftspace, with redcats and in category Category:Pending AfC submissions in article space. The links to tsarevo.info are currently dead, however at archive.org they are live as of December 2025. . The next edit on the article is by Chiserc with the edit summary "Remove unverified excessive details"
• On August 20, created 3 long articles in single edits in draftspace, again with redcats and the Pending in article space cat. These draw on Lithunian, Italian, and Estonian sources.

There's a whole run of similiar article creation this summer. Same category issues, some unusual bolding, . That one relies on English source, but these all have a pattern of citations to pages in these books and article sources, within 12 hours of writing another article. That one has the most smoking guns: a non-existant template, the pending submissions in article space category, red see also links, and the depreciated cite journal parameter "ref=harv", some of the sources are in Greek and German.—BrechtBro (talk) 17:26, 4 September 2026 (UTC)

I have notified Chiserc of this discussion. —BrechtBro (talk) 17:47, 4 September 2026 (UTC)

Hello. I always follow same pattern: I create articles working offline, and then make the final single edit in draftspace. That's my style, if this is a problem, I won't do that, but I always did that and I don't see the problem here.

Regarding folklore articles, yes, I made similar folklore articles in similar categories, since I had finally time to make them now, after a lot of research in this topic; I didn't write them one after the other, I just had time to upload them together this summer. Also, since these are very major articles, with a lot of text, it is evident that they will be fixing issues, especially in citations and formatting, and I asked help for that, with no response. These are very major articles, and I assume a crystal-clear formatting is not expected. Of course, issues with wrong templates or wrong categories will be here, but again I don't see the problem here. Depreciated parameters or non-existant templates are errors that are everywhere; actually I reversed ref=harv when I saw that is now depreciated in some cases beforehand. My focus was on text, not so much on formatting (like red links, archived that were incorrectly mentioned as lived sites, or bolding that was for importance). Also, I had a very tough time to make links to pages, so in some cases, I didn't add a lot.

I normally verify text very carefully, with their sources, and I don't see issues here, or at least major ones. With formatting and templates, yes I always have difficult time to fix these issues, but there is no help. Lastly, Mario Jaén is a tough try to make an article for an actor, that I thought there is potential, but finally I found few sources as stub-article, and I just left it with the hope someone else will make it better. I don't see why everything is labelled, but in any case you are free to decide. Chiserc (talk) 21:55, 4 September 2026 (UTC)

Let me add some more as additional, last comment. These articles were my effort to finally make some in-depth articles; if you see, I requested a B-class rating in many of these articles (reassess), in order to finally have some good articles as a personal, let's say, trophy after many other articles. So, I pushed a lot to add sources, even much more as needed, to justify a good rating for my drafts; this justifies a lot the excessive citations in 1-2 cases mainly (Kuker is the most evident case). Also, bolding was added in some cases, since I didn't have many links to add, so yes, it may be added incorrectly for emphasis in absence of links. Lastly, other issues, especially quotations, and article space categories are errors that I always make, and especially the article space categories are always here as errors (in my last articles I finally fixed that). So, in general, what you describe is mostly my effort to push the articles to make them very detailed, after collecting and finalizing information for a set of articles in similar categories, to finally make them in B-class or finally good articles; I pushed a lot I think, with many issues (not major though, or AI) but I tried for the best.

As a final note to add, I think that some of the articles should be at draft status, since what I asked almost always as comment was support to improve with no response, and they are just being accepted. The AI label I think is quite misleading, but I won't intervene on that, you can decide on that. I don't have time for this now to make more edits, since the sources of text are mostly here, but any help and support from others is more than appreciated. Thanks again! Chiserc (talk) 22:32, 4 September 2026 (UTC)

A spotcheck of Chiserc's articles at wikipedia.gptzero.me all give 1.0 AI going back into December until Prishani Naidoo, David Chidester, and Rebecca Jensen-Clem which score part AI part human, but the earlier Hjallastefnan, Luise Klebs and Lee Haring are also rated at 1.0 for AI. Njálurefill which was accepted into mainspace a few days ago has a bunch of sourcing in Icelandic. User's talk is archived, but I don't see a prior AI warning. —BrechtBro (talk) 00:17, 5 September 2026 (UTC)

Hello. I don't have any warning at all (once a rejection for a very poorly made indeed draft of Hjallastefnan that looked like AI maybe but I made it better and accepted), and actually almost all my drafts are approved via AfCs, and the most experienced reviewers. I don't use AI in any kind of writing process or formatting the drafts/articles. What I do is only beforehand, when sourcing, I use any possible tool in order to translate the text of sources I find from other languages into English, to understand them on my own, and then I write and submit in total the whole article by myself, from scratch. Yes, I had some fluctuated writing experiences throughout this process, and sometimes I had found that I may have used a style that looked like AI, in earlier stages mostly, but honestly it wasn't (Hjallastefnan was the major example on this). This may explain this score, especially since I am a non-native English speaker, and I use multiple languages as sources, as you said before correctly, to create different articles, and not on same topics for diversity. The folklore articles which are on same topic, even these ones, are from different areas in this rationale. Check for this behavior like this: https://hai.stanford.edu/news/ai-detectors-biased-against-non-native-english-writers. I hope this explains a little bit (please don't just minimize all my work, this is quite harsh and unfair). For female researchers especially I made significant effort to find under-represented or young ones, like you mentioned Prishani Naidoo, Rebecca Jensen-Clem, Luise Klebs, and many others (7 women scientists so far).

Apart from this, for the current drafts you moved back in draft space, most of your concerns, I guess if not all, are cases of over-detailed efforts, especially White kuker of Strandzha and Cheshire soul-caking play, or even Estonian nature writing which is a more broad topic, to improve the topics with as much depth as possible (mentioned in comments I left before review in each draft). Kalogeros (folk custom) had issues of some excessive citations, especially in introduction, which I improved a lot in subsequent articles, similarly the others in some parts only, but I don't see anything as major issue, apart from that.

I recommend keep drafts probably White kuker of Strandzha and Cheshire soul-caking play, which are the most evident efforts for over-detailed writing (excessive strings of citations, some red links, and all others you mentioned before), while I don't see issues (at least major ones) for others. It's the best I could for this concern as honest explanation. Thanks again! Chiserc (talk) 01:58, 5 September 2026 (UTC)

Thanks for your quick reponse. I hope you'll understand the difficulty sometimes in determining AI content. Your point about second languages is well taken. However, I'm unconvinced, particularly as to why you would research and write a ~7500 word article and then immediately remove a third of it as unverified rather than adding sources to it. I don't mean this to be rude, but the articles also display a noticably higher level of English proficency than your comments here, which, granted, could at least in part be explained by additional diligence and editing. We'll see what other editors have to say. If I've acted in error the articles will be restored to main and I'll apologize. —BrechtBro (talk) 17:03, 6 September 2026 (UTC)

Hello. Regarding all my latest, mostly folklore, articles, I mention again that I made some big edits in order to make B-class articles finally (explained before in detail hopefully, while you can also see my history for that). The one case you mentioned again, the White kuker of Strandzha, was a massive edit I made after meticulous review of months, and I removed significant parts after a second reading, that this should be ultra-excessive details; I don't mention unverified like it's not supported by sources, but mostly dubious by someone who cannot interpret them correctly will all the details, and some technical knowledge that is necessary. I mention that, because all articles are made within a layperson-level, so unverified means more sources needed to establish extremely detailed information that I removed (I said to myself don't think this is obvious, it may be unverified for others). Actually, the unverified tag is mostly excessive details tag, but I added unverified like there might be a problem of dubious (original) research with all these details. Also, there is the main Kukeri article for more details, but yes it might be an error when I mentioned unverified; most suitable is excessive details (no big problem on this aspect I guess).

Also, regarding English proficiency, of course, when you write an article, you spend a lot of time, in my case around 2 months, to edit it like a small academic article, especially as a non-native speaker. It's obvious that a comment shouldn't be like writing an article, hopefully it's informal; because I had some experience, trying to be in the academic world, I tried to use this knowledge in writing these articles finally. Similar comments I see they are developed within the recent discussion for User:Stupor26, and much more can be said on how faulty AI detectors are, especially for non-native English people and even harder if you have some academic writing exposure on being more strict and formulistic as may need sometimes (see also: https://mitsloanedtech.mit.edu/ai/teach/ai-detectors-dont-work/). You mentioned before errors like wrong templates, but I have made close to 40 articles, and I made one-two errors in total that I copied from the Draft:Kalogeros (folk custom) a few days before. Actually, you also made 2 errors on wrong templates with much fewer articles (see your filter edits as well; you will see two unknown templates, and your first is about AI-assisted sources). My edits are as careful as possible, with formatting issues for sure, but I trigger very rarely filters. I made very big edits and I believe some errors, like all you mentioned before, are obvious and much less than expected actually. I can also say much more (like how I chose these topics, why now, and many many more), but I will wait a little bit in order to avoid over-explanations beforehand. One last for your reference is that I made specifically Draft:White kuker of Strandzha (and probably the Kalogeros draft) a little earlier than expected, and with some obvious errors as said, because 2 days before Burgas was announced as host city of Eurovision, and I rushed to publish it, instead of make it final 2-3 days later as I developed it months before, since it's a local folk custom from Burgas, and I seized the opportunity!

Please just don't minimize all my work so far, because you already labelled many articles quite unfairly and harshly (I believe), and it has been a very tough task to create all these, with no support at all. I understand your concerns for sure, though, and thanks for your care on that! Chiserc (talk) 18:23, 6 September 2026 (UTC)

Just a small note on something, and as general advice: you said initially for Draft:White kuker of Strandzha that the links of a specific website are dead, tsarevo.info, as major AI sign. However, for me, they are still live, I can access them, and I remember correctly they were live a month ago - https://www.tsarevo.info/za-obshtinata/strandzhanski-obichai/obichayat-palikosh-v-strandzha/ and https://www.tsarevo.info/za-obshtinata/strandzhanski-obichai/kukerovden-v-strandzha-i-negovoto-obredno-bogatstvo/. Please just be a little more careful, since these accusations are serious, and more diligence may be necessary to label this and so many other articles beforehand. I hope my comments are helpful finally, thanks again! Chiserc (talk) 19:16, 6 September 2026 (UTC)

Ah, thanks for pointing that out. They didn't work at the time of the post. —BrechtBro (talk) 19:01, 7 September 2026 (UTC)

Noting that Chiserv has removed the LLM tag from Mario Jaén following a copyedit. They have also tagged the articles I draftified for G7 deletion with the note The article was accepted via AfC, but draftified and AI tagged by the same user, among 6 others; although I strongly disagree with detailed reasoning explained in AI noticeboard, I request to delete it to avoid any AI suspicion despite my heavy work here. The LLM concerns go beyond the draftified articles and this odd behavior reinforces my opinion here. —BrechtBro (talk) 17:49, 25 September 2026 (UTC)

Hello. I have requested the deletion of all drafts you tagged, since if you think are problematic, then I remove them, although I disagree. I have to move on beyond something that starts to become quite disrespectful for myself. If you mean odd behavior my clarifications to explain my editing and respect myself is okay, but again I have the right to delete drafts that I made personally, accepted once, reviewed twice, then you draftified on your own, without proper notice; you didn't add just a tag, which is always okay, but you draftify them all in once together. A friendly reminder on something: more diligence when you make such serious comments. Also, there is no further comment by anyone for almost 3 weeks now; the discussion was stalled and near to be archived. Any further comment by anyone else is more than welcome, thanks again! Chiserc (talk) 19:08, 25 September 2026 (UTC)

User:Voicecalls

 Cleanup has been requested and is being tracked at /2026-09-07_Voicecalls, 49 pages need review.

Looks like starting on Aug 27, Voicecalls has made a series of edits (including edit summaries!) that looks AI generated extensive source summaries on contentious BLPs, court cases, allegations with some information reading like a news diary - a lil like the whole news cycle got dropped into the pot! Voicecalls said on their user talk that they used AI "to draft the wording of the edits" and that "I will also slow down. The volume was more than I could check properly." In a talk discussion, @Dekimasu suggested bring this here. NicoR8 (talk) 12:06, 5 September 2026 (UTC)

I am seeing this now. I had replied at the Eye-needed post my reply at the "Eye needed" discussion. I can help you guys to categorize the posts - cosmetic changes versus substantive prose. The start date is August 27 with 200 mainspace edits, 84 pages roughly 81 cosmetic.

Only 2 new pages: one of them Perversion of Justice is a redirect. The article 2026 Maldives speedboat capsizing is a new page. All sources and citations on the new page independently verified again by me. The drafting has stopped. Voicecalls (talk) 20:14, 5 September 2026 (UTC)

Thank you for your cooperation! I have made a tracking page so we can check your work, because most people (including me) underestimate how difficult LLM cleanup is the first time we try. (I got handed an LLM article by Newcomer Tasks; I think it took me a month.) I omitted undos and redirects, and edits less than +/- 35 bytes. It's attached up above. M kuhner (talk) 01:41, 7 September 2026 (UTC)

Thank you for setting up the tracking page.

I am here to help to show which edits are the prose vs mechanical. Let me know if you need that list by category.

2026 Maldives speedboat capsizing - all sources are verified and each ones serve what it is cited for. Voicecalls (talk) 10:42, 7 September 2026 (UTC)

In the tracking page there is a button to the left of each entry. We normally mark "Unnecessary" for edits which are clearly fine (for example small copyedits), and "Ongoing" for articles which are in the process of being cleaned up or deleted. I will ask you to mark "Ongoing" for articles you have cleaned up, and I'll do a check and mark "Completed" if things look good. Thank you again! M kuhner (talk) 14:29, 7 September 2026 (UTC)

Two things to particularly watch for: Recent LLMs often say things that are true and have a source in the article, but then they cite the wrong source. This is hard for a domain expert to spot because the claims look fine. They also often say that many people, sources, critics, etc. said something when really they could only find one. I haven't looked at your articles specifically for these issues but they are extremely common. M kuhner (talk) 14:32, 7 September 2026 (UTC)

Hey @M kuhner thanks for setting this up - the AI certainly kept busy! And thanks to Voicecalls for responding to this. I think there is another issue to watch for - Wikipedia:WEIGHT and Wikipedia:BALASP. For example, see a few diffs: Tiger, Tate, Castro. These are not simply citation issues, but individual news reports, many with detailed, blow-by-blow details. In Tiger Woods, for example, the edits includes detailed toxicology findings and plea commentary. The same pattern is seen in Tate, Castro (even charges prosecutors were still considering) edits and others. So the edits may be individually verifiable, but may still be undue and disproportionate.

I think might be easier to presumptively revert the AI drafted edits first per Wikipedia:LLMPRV rather than waiting for cleanup editors to review each one and figure out what to keep while the live article remains available for the world to read. NicoR8 (talk) 19:48, 7 September 2026 (UTC)

I have fixed the BLP issue on Operation Metro Surge - charges prosecutors were considering against a living person - removed at 20:41.

I will trim Tiger Woods and Andrew Tate - procedural detail in Woods and hearing-by-hearing in the Tate case. Woods diff spans nine revisions and only five are mine.

The BLP fixes took under an hour. Voicecalls (talk) 20:50, 7 September 2026 (UTC)

Hey Voicecalls, Tiger's 9 revisions were included to show your changes collectively. The other 4 were minor edits, unrelated to your section. For Operation Metro Surge, removing that one sentence does not resolve overall issue at hand. NPOV still requires due weight and balance in context of the specific article, not simply that the sentences are sourced. That kind of article level DUE/BALASP judgment is what is missing from these AI drafted edits. They should not remain live while cleanup editors review them one by one and decide what, if anything, can be kept. It may be helpful to read the relevant policies and each article first, then decide what is actually due, and not simply whether there are allegations, court cases, or other developments that can be added because they appear in sources. NicoR8 (talk) 21:39, 7 September 2026 (UTC)

User:Stupor26

I believe the recent spate of aricle creations and major revisions of articles by Stupor26 are done by LLM. Every check I have done with GPTZero and Pangram confirms it. Indeed, I wrote the original War of the Keys article and so am personally confident that it is 100% human written. The tools confirm it, but the new text is at best a 50/50 mix of AI and human text (i.e., mine). This issue was raised on that talk page and on the user's talk page. S/he denies it. Srnec (talk) 17:33, 5 September 2026 (UTC)

“Confirmed” using the infinitely credible and unimpeachable AI checker tools. I’m quite confident in my history contributions, every single statement and expansive edit of which is made solidly from respected and reputable sourcing, and I’m equally confident my past history of contribution to previous pages speaks loudly enough for good faith. The only thing that’s ‘confirmed’ here is injured pride of authorship by one user against another over a major edit that demonstrably expands and colors an important article using firm sources every single step of the way. My edit verifiable built on aforementioned user’s foundations, incorporating more nuance and contextualizations from sources and scholarship used. Frankly, distaste and discouragement for this sort of bonafides expansion, especially when (again) every single point of expansion is done concretely buttressed in scholarship and respected voices throughout, makes such an article pointless otherwise. Stupor26 (talk) 17:56, 5 September 2026 (UTC)

Are you using an LLM to write your article edits and discussion comments? Simple answer, please. M kuhner (talk) 18:11, 5 September 2026 (UTC)

Nope. Never have, and won’t. Also never been accused of such in the history of any of my edits on previous pages I’ve contributed too. I explained why I’ve making contributing large edits over the last few days after relative account silence for a while. As mentioned there, and here, I think my bonafides have spoken for a while. I hope that’s simple enough. Stupor26 (talk) 18:18, 5 September 2026 (UTC)

I was just about to bring the article here. I think its been generated by LLM. scope_creep^Talk 18:25, 5 September 2026 (UTC)

I’m sorry you think that, but I should have hoped AI would have at least made less of the formatting, grammatical and link errors than I consistently have and have to constantly go back and forth fixing and amending, as is evident in every single one of the pages in question—or frankly any contributory edits I ever make; I have no claim to being a streamlined editor in that regard. Stupor26 (talk) 18:29, 5 September 2026 (UTC)

Not sure these are AI. They've been contributing articles since 2023, and this 2026 article and this 2024 article chunk read similarly and use similar phrasing, where if someone was using AI you'd expect them to read very differently. (i.e. the 2024 stuff would sound like 2024 AI, which this doesn't).

Some anti-tells in the prose too -- specifically, running the dataset of post-5/5/26 AI text versus the dataset of pre-2022 articles tagged essay-style, this hits several of the bottom n-grams for 1, 2, 3, and 4-word phrases. Gnomingstuff (talk) 20:48, 5 September 2026 (UTC)

they've also not triggered any edit filters relevant to AI use. But idk, Special:Permalink/1373032675 (8 kb) was created in 6 minutes, Special:Permalink/1373029994 (8 kb) in 13 mins. All of their 11 creations were over the course of two days. OTOH there were no deprecated parameters in them (assuming Zack would have got to them). It'd be the first false positive I've seen from Pangram IIRC Kowal2701 (talk, contribs) 21:26, 5 September 2026 (UTC)

As I’ve explained when asked on the relatively quite fast submissions, I’ve tend to do the lion share of work and *then* upload as one comprehensive edit or page creation. Also as explained, I tend to constantly go over and reedit because of mistakes or errors from the start. I’m grateful for the understanding, but I’ve explained myself and shown (I hope) the bonafides (and much more importantly why the expansions and new articles matter, and add value) at every single point. Stupor26 (talk) 21:54, 5 September 2026 (UTC)

It's just very odd for Pangram to give 100% human-written for your comments here, but 100% AI for every passage from your creations I try (except one AI-assisted) . What software are you using? Kowal2701 (talk, contribs) 22:33, 5 September 2026 (UTC)

I looked at the initial created version of Battle of Florence (1248). All ISBNs check out; I can't otherwise source-check due to lack of access. Article is confusing as it uses multiple names for each faction and calls two different people "Frederick", but these can be human-expert errors.

However, the prose does show WP:AISIGNS (note that I am not using detection software). These labels incorporated older familial, political and economic rivalries rather than representing simple obedience to either emperor or pope. is rule of three and negative parallelism, and doesn't quite click in context. Stürner likewise records the January 1248 expulsion of the rebellious papal faction from Florence by Frederick of Antioch, placing the action among the contemporary successes of the imperial party in Italy. has the typical vague connection words and also--why say this again, it was just said? And the entire Aftermath section resembles typical LLM summing-up and straining for greater significance. I would welcome a second opinion, and/or a pointer to an article with online sources--I find it much easier to confirm or deny LLM if I have access to the sources. M kuhner (talk) 22:39, 5 September 2026 (UTC)

What? So now "the rule of three" and "negative parallelism" are considered signs of AI use? I use both all the time. It seems to me that the AI paranoia here has gotten out of hand. At this rate I expect to be accused of being an AI agent myself. Frederick of Antioch was an illegitimate son of Frederick II, so that is not an error. The text looks fine, and the refs look fine. I've written a bit on this topic, and Stupor26's work looks excellent to me. I just added urls to the three sources cited in "Battle of Florence (1248)", and I've quoted a bit of the German text with a translation. Carlstak (talk) 01:21, 6 September 2026 (UTC)

The Fredericks are obviously not an error, but using plain "Frederick" for one of them, which the creation version does, is unhelpful to the reader when both have just been mentioned.

Most of the WP:AISIGNS are things that humans do too. LLMs just do them more.

I would be grateful if you could point me to an article with sources that are online, so I could do a source check; personally I don't like to rely on AISIGNS for more than screening. M kuhner (talk) 01:55, 6 September 2026 (UTC)

As I said, I just added urls to the three sources cited in "Battle of Florence (1248)". You can click on them, for starters. Personally, I don't rely on AISIGNS for anything, they cause trouble. I use Claude, Perplexity, and Gemini everyday as tools to find sources and assist my research—they help my WP work immensely. Carlstak (talk) 02:12, 6 September 2026 (UTC)

Stürner, p. 574 is cited for both troops numbers in the infobox (alongside Grillo). I cannot see any troop numbers pertaining to Florence on the cited page. Indeed, I see only a brief sentence, but my German is poor. Srnec (talk) 04:50, 6 September 2026 (UTC)

I've removed the Stürner cites given for the battle because they aren't present on p. 574, and as far as I can see, anywhere in the book, with an explanation in the edit summary. As I'm sure you know, not citing specific page numbers in a paginated source is often a red flag that the source doesn't support the WP text. It's possible that the editor just lazily appended that cite to the battle info without checking it, indulging in some editorial hand-waving. Some editors deliberately try to deceive, thinking no one will bother to check, but I wouldn't care to speculate about their motives. It's possible that they used a chatbot to find sources, as I do, but didn't bother to check that everything lines up. I still don't think the whole article was AI-generated, for what it's worth. Carlstak (talk) 19:40, 6 September 2026 (UTC)

User:Boutboul

 Cleanup has been requested.

Boutboul (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Blocked for LLM use, see arbitration enforcement discussion. I've PRODed his most recent articles.—BrechtBro (talk) 18:11, 6 September 2026 (UTC)

Some more analysis: I see Special:Permalink/1369233808 (13 Aug) had WP:AIREDCATs but Pangram gives human-written (tbf it has high false positives) and there aren't any more obvious AI signs. Special:Permalink/1362516754 (July) also has AIREDCAT and a red-linked See also, as well as WP:AITABLE, WP:AIDASH, but no clear prose signs. Special:Permalink/1354651478 (May) has WP:AITREND (The texts form part of the wider corpus of royal inscriptions from his reign), WP:AIVOCAB/WP:SUPERFICIAL (reflects, emphasizes).

Special:AbuseLog/39536895 (Dec 2024) had a chatgpt-derived link, but idk how far back AI-generated content goes or whether it was used outside of creations Kowal2701 (talk, contribs) 21:37, 6 September 2026 (UTC)

I'm going to LLMPROD back to May for now, having difficulty determining a start date. Special:Permalink/1281791888 (March 2025) is potentially an LLM translation Kowal2701 (talk, contribs) 18:40, 14 September 2026 (UTC)

User:Artist&Architect

 Cleanup has been requested.

Artist&Architect (talk · contribs · deleted contribs · logs · filter log · block user · block log)

User admitted to using LLMs to polish their work but denies using it to "generate." This article on a Philadelphia theater award has WLs to the wrong targets, including one with a parenthetical disambiguator to a game designer in Oregon. This article contains some oddly piped redlinks. User's articles on plays are extremely formulaic and a bit puffed up, particularly with how they introduce (excessive) quotaitons., along with some strange capitalization and headings. What stands out to me about the critical reception sections is that they don't logically follow the production history, they all lack the context of what production is being reviewed when they introduce the qoutes, or the section will have an introductory line about a regional production and then have three qoutes from NYC critics. This one has a blockqoute where half of it is off-topic. Signs of AITALK.

At John Molloy, a brief graf following an unsourced blockqoute is cited to this to an SF Gate obit which only supports a fraction of the claims in that graf. —BrechtBro (talk) 20:21, 6 September 2026 (UTC)

User has been warning and seems cooperative. I asked them to clarify affected articles or give a start date. M kuhner (talk) 01:35, 7 September 2026 (UTC)

Hi BrechtBro,

Some clarifying points here:

Well, yes, I absolutely admit to the (conservative) use of LLMs as an editorial aid, primarily to review the syntax and grammar of text I had already written, specifically during some more recent larger-scale revisions. I have never used AI generated text, nor published AI generated claims.

214: AI was not used when creating either the “Barrymore Award for Outstanding Overall Production of a Play” or the “Barrymore Award for Outstanding Direction of a Play” pages. I apologize for the incorrect WLs you identified. I spent many hours researching, writing, and properly sourcing these two pages, and I can assure you that these are human errors.

215: As for the “oddly piped redlinks” on the A. J. Shively page, there were no redlinks when I created that page on July 14, 2026. The linked pages appear to have been subsequently deleted as part of Wikipedia's ongoing efforts to address alleged AI-generated content. I therefore do not see how those redlinks can be attributed to any alleged AI use on my part.

216: I have spent many hours creating pages for noteworthy American, British, and Irish plays that I love and have encountered over the years. The characterization of these pages as “formulaic” is somewhat fair, as I have deliberately tried to model their structure on pre-existing Wikipedia entries that I regarded as good examples of how articles about plays should be organized. If I was in the wrong here, I apologize.

However, the characterization of them as “puffed up” feels both vague and mean-spirited. I'm not sure what specific passages you believe demonstrate this. In regard to the claim that the “critical reception sections don't logically follow the production history,” I was operating under the assumption that “critical reception” referred to the reception of the plays themselves rather than to individual productions. If that is contrary to Wikipedia's preferred practice, I'm more than happy to adjust my approach going forward.

220: For me, this is your most preposterous accusation. The conversations you reference were heated and, admittedly, somewhat emotional for me. It is my opinion that the particular editor involved has exhibited a pattern of hyper-policing and undervaluing the contributions of Black women. There is substantial evidence and discourse concerning racial and gender biases within Wikipedia's editor community and the ways in which those demographics continue to shape the site’s content. These comments were not “AISPEAK.” They were carefully considered responses that I wrote myself, in the context of a disagreement that I found personally quite frustrating. I was particularly diligent in formulating my responses, and discussed some of the more contentious points with trusted friends before posting them.

Frankly, I also find it noteworthy that you began scrutinizing my contributions immediately following this particular dispute with another long-serving Wikipedia editor, despite not having participated in the discussion yourself. I don't want to make any unsubstantiated accusations about your motivations or relationship to said editor, but the timing is sufficiently unusual, and therefore worth addressing.

222: This (once again) appears to be an instance of human error on my part. The biographical details articulated in that section are substantiated across the article's other sources. My use of a single citation from the SFGate was definitely an error on my part, and I take full responsibility for that.

In many ways, I am still finding my footing as an editor on this platform, and I have greatly appreciated the interventions of numerous senior editors who have identified some of my past errors and helped educate me about Wikipedia's best practices. Time and time again, I have tried to take that criticism in good faith and incorporate it into my contributions.

Your intervention, however, does not feel like that to me. It feels punitive, if not vaguely targeted. Respectfully, this feels a bit like a witch hunt. Artist&Architect (talk) 16:24, 7 September 2026 (UTC)

I'm surprised to see this given the polite interaction we had at your talk page, particularly in the light of the comment above by the other editor, and I'm sorry you feel hunted. This is routine AI cleanup work, the discussion here is to determine what, if any, cleanup is necessary. I haven't accused you of anything here, you're not being punished, some of these mistakes are certainly within the realm of human error. This wall of LLM text (whether polished or generated), however, does not inspire confidence and I suspect the comment on your talk page was the same. You are correct regarding the redlinks at the Barrymores page, I didn't look closely enough there, that's my bad. Thanks for pointing it out. I have no knowledge or interest in the dispute you mention other than from looking at your edit history. —BrechtBro (talk) 16:59, 7 September 2026 (UTC)

Yes, I was very receptive to your initial warning on my talk page. I was genuinely unaware that my editing was in violation of Wikipedia’s guidelines, and appreciated you brining this to my attention.

However, you never responded! Next I hear from you is via a post on this thread. Forgive me if I interpreted those actions as an escalation.

Furthermore, my above response is not the work of a LLM! I wrote it myself. What is so alarming to you? Its structure? Length? Attention to detail? Artist&Architect (talk) 21:59, 7 September 2026 (UTC)

Primarily WP:AIADHERE in the comments I linked to, plus the use of em dashes and curly quotaiton marks. I do not find it alarming, but, yes, the way it is structured, both how it's broken down into sections and your sentence structure, word and punctuation choice, WP:RO3, the way AITALK in my comment became AISPEAK in yours. See WP:AITALKSIGNS. —BrechtBro (talk) 00:40, 8 September 2026 (UTC)

At Special:Diff/1368042903 neither source 9 nor 10 mention awards nor contain the quote (though the quote is sourced down in the body where it's repeated), nor the number 5,000, which is not mentioned again in the body. Besides the WP:AIPUFFERY et al, it's got a pattern where the citations are not in the article in a logical way, and not used in the way you might expect a person to. New sources are brought in rather than reusing other sources, even when some of those sources are better. The best sources, like the NYT article, end up serving no real purpose in the article and isn't mined for content. —BrechtBro (talk) 01:25, 8 September 2026 (UTC)

Source 9 is also cited as a Ben Brantley review with the headline "Hunger for Love, and Other Forms of Nourishment" but the article linked is by Gwen Orel titled "Theater Festival Presents New Voices From a New Ireland" —BrechtBro (talk) 16:37, 8 September 2026 (UTC)

User:Archiwee

 Archiwee (talk · contribs · logs · block log) (7 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Created the current iteration of Jayanta Bose in Diff/1371784644 (as User:Archiwee/sandbox), which doesn't look very AI-generated at that point. Some references are added in Diff/1373208139, and nr. 1 returns a 404 error. On Talk:Jayanta Bose the user has blamed the 404 not found on geo-blocking, bot-mitigation firewalls, or dynamic URL rerouting, despite the URL never having been archived. The level of English stands in contrast to their first talk page messages in Diff/1371674119 and Diff/1371933787, which is perfectly fine but does suggest some use of LLMs to communicate.

They've also created the page as Acharya Jayanta Bose in AbuseLog/45043764, which has UTM parameters showing that it is ChatGPT that is being used. That was likely a copypaste move from draft space, there are AfC headers. This seems to have been a problem because they tried removing a G6 tag in AbuseLog/45043944 and there are multiple notices on their talk page about copypaste moves. In User talk:Archiwee § Please..., they were asked to stop copypaste moving pages and to not LOUTSOCK to remove CSD tags. In Diff/1373339217, a TA was used to remove an {{ai-generated}} tag. There are more UTM parameters from ChatGPT in AbuseLog/45044000, which looks like another copypaste move from draft space. The current iteration of the page was moved to draft space in Diff/1373315025 and then moved back shortly afterwards in Diff/1373337055. After the draftificaton, the user wanted to know which portion of the text seems machine written in § Jayanta Bose moved to draftspace on their talk page. It is a bit hard to trace the full history of this page through the copypaste moves. ‑‑gurkubondinn 10:59, 7 September 2026 (UTC)

Hi Gurkubondinn. I see your concerns about me using AI models to write the article.

I would like to be totally transparent; being a new editor, i did take help from AI models. I used the deep-research feature of chatgpt to search for newspaper articles on the subject.I manually opened and verified the source articles before citing them, but when I copied the web links over, I didn't realize they contained ⁠utm⁠ tracking tags from the tool. Also, in my earlier drafts, I took help from AIs to polish my language, but i have stopped using AI to write any part of the article, after my article got flagged for being AI written once.

And, i would like to address your concern about the earlier copy-pasting issue. I was actually not familiar with the move function, hence i copy-pasted my draft to the mainspace. I am extremely sorry if i have caused any nuisances or problems. I am still learning, and i only want to contribute to wikipedia. The updated article that is live now was fully written by me. I have refrained from using AIs for polishing my language after the page was flagged due to AI usage. I truly thank you for your constructive criticism. Archiwee (talk) 11:22, 7 September 2026 (UTC)

Did you copy this response from somewhere? ‑‑gurkubondinn 11:39, 7 September 2026 (UTC)

I did not! And just letting you know, i have cleared up the utm tag from the 1-2 citation urls that had them. Again, thank you for your feedback. Archiwee (talk) 11:42, 7 September 2026 (UTC)

Are you absolutely sure that you didn't paste in anything? ‑‑gurkubondinn 11:52, 7 September 2026 (UTC)

I really am positive, i did not use any large ai models to write my articles. As i said, i used the deep reseaech feature of chatgpt to fetch me some data and citings, but i assure you, no part of the article is AI written. Archiwee (talk) 11:55, 7 September 2026 (UTC)

So you are saying that you typed everything in the sentence I didn't realize they contained ⁠utm⁠ tracking tags from the tool yourself? It was not copied from anywhere? ‑‑gurkubondinn 11:59, 7 September 2026 (UTC)

Youre asking about the reply? I was actually not sure of what utm meant, so i did search it up, and copied the part, which resulted in the emojis appearing. However, the overall reply was written by me! And about the article, i really did not use any AI language. I did use grammer tools, to fix any obvious verb, preposition or tense error. Archiwee (talk) 12:06, 7 September 2026 (UTC)

No, those emojis showing up were because of a mistake that I made in Diff/1373697873 (and fixed in Diff/1373699670), so that doesn't make sense. I'm also not sure why you would have gotten confused because I asked if you did copy this response. But regardless, Diff/1373522771 and Diff/1373556030 both triggered a paste check, so something isn't adding up here. ‑‑gurkubondinn 12:14, 7 September 2026 (UTC)

Oh, i see, i was also confused by the emojis. Im sorry for any confusion. I would totally avoid pasting anything from now on. I did not know pasting something triggers a paste check. And i am not doing it out of malice, i sometimes do include a sentence or two from what i search online into my responses. However, I did not use any AI model in the latest article that is currently live. I did get flagged for AI usage once in my earlier draft submissions, so i have stopped using anything but my own language, and a tool that checks for any obvious grammatical error. Archiwee (talk) 12:21, 7 September 2026 (UTC)

I wanted to address your concern about my usage of the terms “geo blocking”, et cetera. I searched online what might have caused the editors to not being able to access the citations, and i learnt that many links use geo blockers, firewalls, etc. Thats why i used the terms. Thank you! Archiwee (talk) 11:50, 7 September 2026 (UTC)

User:Aakib.0000

 Aakib.0000 (talk · contribs · logs · block log) (7 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

New user, but prolific in some sports articles with multiple large and rapid-fire edits. There are markdown code blocks in AbuseLog/45061175 and AbuseLog/45061319 as well as the byte-sequence described in WP:AISIGNS § turn0search0. Created the 2026 Saudi Super Cup article, added a reference with a ... placeholder for the URL in Diff/1371064909 and a reference that leads to a 404 in Diff/1371065296 (produced in one minute). There's possibly more AI-editing in the same article, because a different user showed up and added a red category in Diff/1371190772 (two of them are red now, but one seems to have existed at the time). Also created Al-Ittihad Club (Saudi Arabia)Al-Ittihad Club (Saudi Arabia) in Diff/1371060246 (AbuseLog/44951560), before then doubling it in size in Diff/1371061227. The page was turned into a redirect by @Reconrabbit in Diff/1371262713 the next day. I've looked into this user before, but I don't remember it was that I saw that caused me to do that. ‑‑gurkubondinn 11:27, 7 September 2026 (UTC)

I don't think any AI was involved in the creation of Al-Ittihad Club, its content was copied from the current redirect target. -- Reconrabbit (talk) 13:19, 7 September 2026 (UTC)

You're right, the text in Diff/1371061227 also seems to be from the redirect target, which Aakib does not seem to have edited. ‑‑gurkubondinn 13:28, 7 September 2026 (UTC)

User:Kent Wang

 Kent Wang (talk · contribs · logs · block log) (7 September 2026)
This issue is currently under review. 

Pasted a chatbot block in Diff/1373672821. Says in User talk:Kent Wang § September 2026 on their talk page that they use LLMs to generate the wikitext, and now wants a third opinion on if they are exempt from the NOLLM guideline. ‑‑gurkubondinn 12:25, 7 September 2026 (UTC)

The block I pasted is all photos and captions. I look forward to other opinions on this. Kent Wang (talk) 14:21, 7 September 2026 (UTC)

WP:NOLLM does not make an exception for captions. Have you used LLMs on other articles? M kuhner (talk) 15:07, 7 September 2026 (UTC)

Can you look at the Diff/1373672821 and tell me if any of the captions are bad? Or any of it is bad?

I personally human-reviewed all captions. This is covered in the WP:NOLLM exception "Editors are permitted to use LLMs to suggest corrections to their own writing, and to incorporate them after human review." Kent Wang (talk) 15:51, 7 September 2026 (UTC)

No, again, that is limited to spelling, punctuation, capitalisation, grammar. I've already told you this, and it plainly says so on the guideline's page. There are no carveouts for captions or to generate wikitext. Have you been using an LLM for anything else? ‑‑gurkubondinn 16:05, 7 September 2026 (UTC)

Would you be happy if I made the change again but without the captions?

> Have you been using an LLM for anything else?

I have replied to that below. Kent Wang (talk) 16:48, 7 September 2026 (UTC)

Yes, I used LLM to generate the wikitext and proof my edits on another page Igreja da Madalena Special:Diff/1373712248/1373712902. Please review the diff and tell me what you think. Kent Wang (talk) 15:57, 7 September 2026 (UTC)

That is an edit I personally would mark "unnecessary" in a tracker as there is no substantive text in it. My concern is that you may have used LLMs, perhaps not deliberately, in cases where they would introduce substantive text. They tend to go beyond the limits of what they are asked for quite frequently; anything that touches a citation is dangerous as they do thinks like "helpfully" add ISBNs, DOIs, etc that are not correct. M kuhner (talk) 16:12, 7 September 2026 (UTC)

Yes, I'm well aware of that and never use LLM for substantive text. Kent Wang (talk) 16:46, 7 September 2026 (UTC)

User: FeedMeGrapes

Cleanup efforts are ongoing.

User:FeedMeGrapes seems to be an autonomous AI agent, which has edited and rewritten multiple articles: Special:Contributions/FeedMeGrapes. All of the edits seem to be written by AI, because of the distinct style and short periods of time for huge changes

Some examples:

Special:Diff/1373451348 — +9,980 bytes to Atomicity (database systems) in a single edit, described as a comprehensive rewrite covering history, recovery mechanisms, distributed atomicity and implementations.

Special:Diff/1373489889 — +9,462 bytes to Data mapping in a single edit at 06:40 on 6 September.

Special:Diff/1373491390 — only 20 minutes later, at 07:00, creation of Draft:R2RML with +11,827 bytes. This means over 21,000 bytes of substantial technical prose across two different topics in 20 minutes.

The submission R2RML was also independently declined by a reviewer who stated: "This draft appears to contain text generated by a large language model" Keschler (talk) 14:00, 7 September 2026 (UTC)

Do you have evidence that they are an autonomous bot? This seems quite rare; usually it's a human cutting and pasting text to and from the LLM. (Not that that's better, but next steps are different if we think it's an actual bot.) M kuhner (talk) 14:38, 7 September 2026 (UTC)

Don't have concrete evidence. For me it was the generic name + These two contributions. I could definitely be wrong. Keschler (talk) 15:03, 7 September 2026 (UTC)

Hi guys, I'm definatly not an autonomous agent, I don't know how one goes about doing human verification here, although I'm based in USA, so I can always jump on facetime with anyone and prove that lol

Those two edits across 20 minutes were because I make the edits locally up ahead while I have spare time at work (I'm a DBA and data engineer for over 15 years and tend to have a lot of down time) and then I update the articles on wikipedia when I'm at home. You can see my pattern of uploads pretty much falls on weekends, when I'm bored at home.

I've been donating to Wikipedia for years and I read recently that the amount of contributors has gone down in the last year, and I checked and the Database wikiproject is inactive, so I decided I would help out in my own field that I'm passionate about.

I have definatly been using LLMs to help clean up some of my writing, as well as to help find additional sources that I can't find on my own..if that's not allowed I will stop. But even in the cases where I have AI assistance, I'm checking all sources and all the underlying writing is my own and its all double/triple checked. I just wanted to help out in my particular field of expertise.

Anyway, happy to continue this discussion and learn best practices from anyone who is willing to guide me here :) FeedMeGrapes (talk) 19:36, 7 September 2026 (UTC)

I'm reviewing Atomicity. Review is far from complete, but so far I haven't found the typical LLM source/text inconsistencies or other issues.

I suggest being extremely wary of LLM clean-up of your text. It can easily introduce LLM "smells" and, worse, nasty little inaccuracies like ISBNs, DOIs, or wikilinks that point to the wrong place. Vagueness and puffery are also issues. Your plain English will be just fine.

I don't see any evidence to suppose you are a bot. That's actually very rare. A person with a bot, alas, is behind probably 90% of all big single-edit article creations we see recently, so we have to verify by looking in detail. I'll post back here when done. (In the meantime I'll say, I'm enjoying the article, which is a rare treat in cleanup.) M kuhner (talk) 20:25, 7 September 2026 (UTC)

"Your plain English will be just fine." - Noted. I'm pretty self-conscious in that regard, but I'll do that going forward.

"In the meantime I'll say, I'm enjoying the article, which is a rare treat in cleanup." - Thank you, I try to make an effort to add as much history, examples and popular implementations as possible to make topics more interesting, as I find that's what helps me when learning or reading up on a topic. FeedMeGrapes (talk) 20:47, 7 September 2026 (UTC)

@FeedMeGrapes, thanks for understanding and adjusting. Dw31415 (talk) 02:30, 8 September 2026 (UTC)

I just reverted that article, the previous edit to it (Fabrice31) was also one in a series of AI edits Gnomingstuff (talk) 04:52, 8 September 2026 (UTC)

cleanup is done, except for Automatic parallelization, ACID, and Data mapping which are tagged (Prediction by partial matching I've inline-tagged) Kowal2701 (talk, contribs) 15:09, 15 September 2026 (UTC)

User:AkaneNel

 Cleanup has been requested.

AkaneNel (talk · contribs · deleted contribs · logs · filter log · block user · block log)

This article has AILEGACY, REDCATS, and a AI 1.0 from gptzero. The hip injury is a thing that happened, but it isn't mentioned in the cited source for that, and there's that a legacy claim about the NYT, LAT, and Variety, while only citing two of the three. Extrapolated from a photogragh of the subject is the claim that, His photographic records are held within the New York Public Library's Digital Collections.

Perhaps an AITABLE Special:diff/1356965252 here. AILEGACY, more redcats. AILEGACY and a rather impressive amount of offline sources, almost all of them used only a single time. There's a change in the pattern of edits when creating articles in November 2025, going from multi-step to single edit. —BrechtBro (talk) 05:07, 8 September 2026 (UTC)

They've been adding lots of deprecated infobox parameters as well. Haven't been warned for LLM use until now

BeauSuzanne (talk · contribs) is potentially another account according to Special:Diff/1303209109, and what d'ya know, they appear to be using AI: Special:AbuseLog/42661223. The EIA seems to be above-board , but courtesy ping @S0091 anyway Kowal2701 (talk, contribs) 11:17, 8 September 2026 (UTC)

Thanks for the ping. The reason I left that note was because they were confirmed to each other several years ago and the CU left a sock note but neither responded. BeauSuzanne was mentioned by a filer in a later SPI so I left the note again to see if I could get a response. They both denied having other accounts and as you can see in the EIA, they rarely cross each other. S0091 (talk) 14:38, 8 September 2026 (UTC)

I tagged the unreviewed articles in the NPP backlog by AkaneNel with the AI template, with links to this discussion. AkaneNel has begun editing the articles and removing the templates but have yet to respond about LLM usage at their talk page (or here). —BrechtBro (talk) 01:59, 13 September 2026 (UTC)

@AkaneNel, please could you respond here? I see you've been rewriting some of the articles that were tagged, if you want we can make a tracking page (like Wikipedia:AI noticeboard/2026-09-11 PrettyPony1234) to make things easier? Just some advice about AI cleanup: The biggest issue with LLM use is hallucinations (i.e. WP:V failures), though there are other important issues like WP:NPOV/WP:WTW and WP:COPYVIO. Usually we recommend that people stubbify or rewrite LLM-generated articles from scratch since it's often very time-consuming to attempt to preserve such content, but it's up to you. Of course deletion is also an option (WP:LLMPROD). Thanks, Kowal2701 (talk, contribs) 14:22, 13 September 2026 (UTC)

Hello Kowal2701. I apologize for the late response. I haven’t used AI but only for translations. And it’s alright I will rewrite it. I will be careful not to do LLM. I had seen some articles you like Clark Gable and Bette Davis so I was trying it in that style but I understand and I will stop with the LLM issue. Once again I apologize for it.(AkaneNel (talk) 16:50, 13 September 2026 (UTC))

@AkaneNel, none of the articles discussed above appear to be translations? Kowal2701 (talk, contribs) 18:28, 13 September 2026 (UTC)

Hello Kowal2701. I didn't meant the English I mean the French and Korean for it I use translation. I also go on different languages wikipedias find some information. No the articles are not from translations.(AkaneNel (talk) 18:37, 13 September 2026 (UTC))

noting Talk:Mickey Squires#WP:V Kowal2701 (talk, contribs) 17:19, 16 September 2026 (UTC)

The Sergei

The Sergei (talk · contribs), XTools

Warned for AI use August 2025, copyvio Feb 2026, warned May 2026. Special:AbuseLog/40962963 (June 2025) had a chatgpt-derived link (lots more after this), Special:AbuseLog/43638743 (March 2026) triggered Special:AbuseFilter/1325. Special:Permalink/1360752330 (June 2026) had AIREDCAT but it's a list so no other signs (they're mostly lists). Notified them of this thread. Kowal2701 (talk, contribs) 11:43, 8 September 2026 (UTC)

Special:Diff/1371653300 was translated from hewiki, not sure they speak Hebrew. Special:Permalink/1365632763 (July 2026) used a non-existent cite template, Special:Diff/1365434113 (same month) was made in 6 minutes on mobile Kowal2701 (talk, contribs) 11:50, 8 September 2026 (UTC)

1) June 2025: If there were links from chatgpt.com at the end, I was told in August User talk:The Sergei#Ways to improve List of international prime ministerial trips made by Kristrún Frostadóttir that they should be removed to make sure that the site is not created by Chatgpt itself.

2) August 2025: About Dick Schoof's trips to the Caribbean parts of the Netherlands, I didn't know that sources might not be permitted, but government sources, for example, might be permitted, I don't remember which sources I used. StevoLaker told me in August 2025 that "You can put just about any information on a Wikipedia page so long as it is anbout the topic and is referenced by a reliable source (almost any news or government source)."

3) February 2026: About text from Petteri Orpo visits, the sources belong to the Finnish government (I might be wrong, and not all of them are like that), and as far as I know, you can use government ones, + their images can be uploaded from Flickr (Not all, but some images uploaded about his trips), although this does not apply to Wikipedia.

4) May 2026: Regarding Japan's attitude toward the ICC arrest warrant against Israel, I can say that I could have asked for AI somewhere. Well, if it does not match the title nor the quote, then I should have checked what was written and copied the existing text.

5) July 2026: About "List of international prime ministerial trips made by Janez Janša," I translated his page from Slovenian to English. If you're talking about the source that explains why there are few visits due to the pandemic, I see that there was no source at first, then I added one, although I don't remember what it was. The Sergei (talk) 14:46, 8 September 2026 (UTC)

You can put just about any information on a Wikipedia page so long as it is anbout the topic and is referenced by a reliable source (almost any news or government source). - Sadly, the advice you received is incorrect (and significantly so). The policies WP:V, WP:RS, WP:NPOV, and WP:DUE dictate what can and cannot be added. You cannot in fact just add anything that's backed up by any news or government source, never mind that some such sources can be considered generally unreliable or marginally reliable. We don't add minutiae just because it was mentioned in the news, for example, per WP:NOTNEWS. Accessedgrant (Epicgenius mobile alt) (talk) 03:18, 9 September 2026 (UTC)

It looks like you were actually told in User talk:The Sergei#c-Lova Falk-20250817124500-Thank you and... to use Citer in order to format citations properly, and also to find out whether ChatGPT had invented websites that didn’t exist. Removing the utm_source part of the url doesn’t change whether the url points to a valid page or not. Suðurhafsljósæta (talk) 10:01, 9 September 2026 (UTC)

It remains unclear to me whether your claim is that you have used LLMs or not in the course of your editing. Fermiboson (talk) 10:15, 9 September 2026 (UTC)

I translated pages, some from their language, or from another. This concerns relations between countries and the list of trips made by someone. The Sergei (talk) 10:45, 9 September 2026 (UTC)

Do you speak all the languages you translated from? Kowal2701 (talk, contribs) 11:58, 9 September 2026 (UTC)

No, I don't The Sergei (talk) 19:28, 9 September 2026 (UTC)

@The Sergei, sorry but that isn't allowed, translations have to strictly follow WP:LLMTRANSLATE Kowal2701 (talk, contribs) 00:32, 12 September 2026 (UTC)

I looked through the editor's prolific history back to 2024. There is admitted LLM use June 4 2025 (only for searching, editor says). There are unacknowledged translations, for example one in October 2025 where infobox parameters are in Spanish. Editor occasionally responds to complaints but always sounds bewildered; there is little sign that they change their very fixed editing style in response. Sourcing is a major issue: I found a claim sourced to a user-supplied caption on a photo marketing site, several to Twitter, and a great many not sourced at all. This is true throughout the entire time period. Copyvio has been a recurring issue. There is also a bad habit of putting articles in mainspace which are clearly not ready, for example one Aug 1 2025 which has "TBA" in most of its table fields.

On the other hand, that edit was presumably not full LLM. I suspect the text is a mixture of LLM and human, and has been since mid-2025 at least.

Early on their edits were challenged more by other editors, but the articles on which they came to focus have few watchers: they are lists of government-head visits to different countries. There was an AfD on one list article where it was clear they felt very strongly about this work, but not necessarily in an encyclopediac way (i.e. "why does the reader need this?" was not in their priority list). They got a fair amount of support on this view--it's the Great List Debate--and the article was not deleted.

I am honestly not sure what to do here. The Talk page does not suggest that warnings and discussion are very helpful. M kuhner (talk) 18:06, 18 September 2026 (UTC)

User:Wolf july

 Wolf july (talk · contribs · logs · block log) (8 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Newer user, but has created a few articles. The latest article is Titanic (video game), which was created after having been pasted with a chatbot block in AbuseLog/45064544. Despite meeting the G15 criteria, the CSD was declined so I've sent it to AfD instead, where I believe the user has commented with some WP:AITALK in Diff/1373890037 claiming that Wikipedia does not establish a clear, mechanical threshold whereby such formatting elements alone prove that an entire article was AI-generated. I also find it odd to say that sources and content of this article can be independently verified without mentioning that another user has said in User talk:Wolf july § September 2026 that they did verify the sources, because I find that a strange thing for a person to just gloss over and not mention at all. The user has also confirmed some use of AI on Wikipedia, but I think that their edits need to be looked at closer. Haven't seen any other AITALK from them. ‑‑gurkubondinn 15:48, 8 September 2026 (UTC)

For what it is worth, Pangram classifies the AfD comment as being AI-generated (which should of course be taken with the usual grain of salt). ‑‑gurkubondinn 15:51, 8 September 2026 (UTC)

And Corneliu Drăgan-Târgoviște was created in Diff/1367002209 (AbuseLog/44773422) with a red category (the infobox image was deleted from commons later). ‑‑gurkubondinn 16:02, 8 September 2026 (UTC)

Here i don t understand what happened. I uploaded the picture of the artist and i forward the email with the required form...but the picture disaperead..it was erased.And it was pretty hard for me to obtain it since the artist is very hard to reach. Maybe you can help me to underastand..thank you..) Wolf july (talk) 16:48, 8 September 2026 (UTC)

I am finding that it is more likely than not that most of this editor's contributions are either partially or totally concieved using AI. I marked one section of Hysteria (video game) as "failed verification" because Crash does not call the gameplay "repetitive" (also, there are 3 named reviewers in the linked article). I'd appreciate if Wolf july was forthcoming about whether they've used AI for anything besides "correcting grammar". Regarding Titanic (video game), I've already said what I have to say about that on the AfD page. -- Reconrabbit (talk) 16:10, 8 September 2026 (UTC)

I read the same review and there's one more source verification issue there, only one of the reviewers "praised the graphics", but the article implies that the review as a whole did. ‑‑gurkubondinn 16:21, 8 September 2026 (UTC)

i used the crash source for the overall score. it is written down 78%. i can move at the begining of the frase, after the score - but the source is valid . :) Wolf july (talk) 16:35, 8 September 2026 (UTC)

I don t intend to comment too much on this topic...my answer come after 4 minutes... I did not had time to check that was an answer...if you check you ll see are 4 minutes between the comments...The rules adopted in March 2026 as i understood forbides the using of LLM to generate or rewrite the texts! There are exceptions for minor copyedits on the texts without introducing new content! Me - i research and write my articles myself. I already explain my case. I think my editing history and the sources used in my articles can be independently reviewed. Anyway...i must admit, however, that there is a certain irony in suspecting someone of using AI and then using an AI detector as additional evidence for that suspicion. :)...something like...aa..you are using ai...i checked on y with ai...:))...i found it very funny...:)..also on the other hand i am open to any advice! Wolf july (talk) 16:39, 8 September 2026 (UTC)

Well, your replies here basically confirm the AITALK on their own. The English idiom "take it with a grain of salt" means that something may be incorrect and should be viewed with a degree of skepticism. ‑‑gurkubondinn 16:45, 8 September 2026 (UTC)

the content in Special:Permalink/1374056558 cited to ref #1 appears human-written and checks out, but appears to be WP:CLOP, I'll leave a comment on their talk Kowal2701 (talk, contribs) 20:40, 16 September 2026 (UTC)

How many ways can you write that a game was published by a company in a given year? -- Reconrabbit (talk) 21:01, 16 September 2026 (UTC)

that falls under WP:LIMITED Kowal2701 (talk, contribs) 21:03, 16 September 2026 (UTC)

User:Jbaird FEA

LLMPRODs are done, no drafts to tag

Cleanup efforts are ongoing and being tracked at /2026-09-12 Jbaird FEA, 15 pages need review.

Jbaird FEA (talk · contribs)

Enormous amount of edits with various AI indicators, including at least one Markdown instance in filter log, and marked difference between their general writing style and their grammatically mangled comment of False and unbased acquisition of AI. The bulk of it probably should just be reverted.

Suspect something like Claude here as opposed to ChatGPT Gnomingstuff (talk) 17:49, 8 September 2026 (UTC)

Yeah that's pretty blatant. Reverting, will get down to source text discreptancies in an hour or two. Fermiboson (talk) 09:43, 9 September 2026 (UTC)

jeez, see Special:Permalink/1316013361 (Oct 2025). LLMPRODing Kowal2701 (talk, contribs) 00:39, 12 September 2026 (UTC)

User:Garudindflamewing

Cleanup efforts are ongoing and being tracked at /2026-09-10 Garudindflamewing, 36 pages need review.

 Garudindflamewing (talk · contribs · logs · block log) (10 September 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. Editor's autopatrolled permission has been removed.

User with 20k+ edits and autopatrolled. I noticed Raghuttama Tirtha was full of sentences such as The diverted bridge stands just outside the temple complex to this day as an enduring testament to the saint's living presence. Satyadhyana Tirtha is similar, with sentences such as This period provided him with valuable experience in management, finance, and human relations. His sharp judgment of men and events earned him respect as a capable leader. At the same time, his humour, warmth, and accessibility made him popular among students and scholars. and Through rigorous logic and scriptural authority, he decisively refuted their arguments, a victory that entered the annals of Vedāntic history. Both articles are reckoned to be mixed by GPTZero. They were created before 2022 but were significantly expanded recently with the AI-sounding prose.

Prasanna Venkata Dasa was created recently, and GPTZero judges it to be AI with certainty, with sentences like His life and works exemplify the fusion of scholarship, devotion, and musical creativity in service of spreading Madhva philosophy and devotion to Lord Vishnu. Beside the prose, the editing patterns of large edits, sometimes in rapid succession, mostly on mobile, with conspicuously vague edit summaries like "added sourced content" don't give much confidence.

Compare pre-2023 pages such as Satyanatha Tirtha, which GPTZero confirms to be human. Kisseran (talk) 15:47, 9 September 2026 (UTC)

WP:GPTZERO, but in this case I think you're right, though it is a weird case. I mean the tone is certainly promotional, but it is too overtly promotional to be typical LLM use (of course these things vary with prompts a lot). Edit pacing is also atypical. Will come back with a more detailed analysis in a bit. Fermiboson (talk) 02:12, 10 September 2026 (UTC)

Blatant AITALK on User talk:Narayan Dikshit. @Admins willing to patrol AINB: can we get an autopatrol pull real quick while we look at this? Fermiboson (talk) 02:38, 10 September 2026 (UTC)

@Fermiboson: You are correct that AI text is present there (User talk:Narayan Dikshit), but I want to clarify the extent. I only used an AI assistant to help me draft and format the specific section regarding Wikipedia guidelines. The rest of the message and my actual comments were written entirely by me. However, I now understand that pasting any AI generated text onto a talk page is strictly prohibited and undermines community trust. That was a big mistake on my part. I will strictly avoid using AI for drafting articles or talk page messages going forward. - Garudindflamewing ^(talk) 07:13, 10 September 2026 (UTC)

Removed autopatrol, noted that the removal is temporary and can be restored by request. I considered removing the reviewer right also, because reviewers nowadays need to be able to recognize LLM-generated content, but I left that alone. ~Anachronist (who / me) (talk) 14:40, 10 September 2026 (UTC)

Filler sfn at . Fermiboson (talk) 03:05, 10 September 2026 (UTC)

Tracker page created. I ticked only the pages with substantial non-restorations, though my gut feeling is that there are significant WP:CT/CASTE violations all the way back past 2023. Fermiboson (talk) 03:29, 10 September 2026 (UTC)

Thank you for the ping and for reviewing these articles. I want to be completely transparent about my workflow here, as it differed between the pages mentioned. For Satyadhyana Tirtha page i used GPT to some extant for drafting and copyediting. But I forgot to neutralize the article to meet NPOV. I will fix this article give me some time. Regarding Raghuttama Tirtha article it is entirely based on books, recently I expanded it. This article expansion is based on Kannada book Sri Raghuttama Tirtharu by a prominent Scholar Vyasanakere Prabhanjanacharya. Since I adhered closely to the source material, the Kannada to English translation ended up retaining the author's reverential and devotional tone. Coming to Prasanna Venkata Dasa article as noted this is a very brief article but I will review it as well to ensure the tone remains neutral. I will not use GPT for drafting going forward. Thank you! - Garudindflamewing ^(talk) 05:36, 10 September 2026 (UTC)

It would be helpful if you could indicate all of your previous creations and edits that were AI-generated. (Note: Please include the phrase "per my previous point" in the first sentence)Regardless, I believe that your autopatrol right is not compatible with the pattern of edits that have been exhibited. Aside from cleanup, it would also be helpful for you to provide the prompt you have used. Note for the future that WP:LLMT is a thing, and also that overly close translations of copyrighted material may itself constitute a copyright issue. Fermiboson (talk) 09:42, 10 September 2026 (UTC)

@Fermiboson:, thank you for the feedback and for linking to WP:LLMT. I want to firmly clarify the situation regarding Raghuttama Tirtha. I did not do an overly close translation of copyrighted material. I manually researched the Kannada book Sri Raghuttama Tirtharu and wrote everything entirely in my own words. My error was strictly an NPOV one. I was talking about the devotional tone of the original author unintentionally carrying over into my writing. I will go back and neutralize those adjectives, but there is no copyright or close-paraphrasing issue there.

Except for these two instances, I have not used AI for any other articles or edits. For Satyadhyana Tirtha article my prompt was "Expand the article with the text I'm giving" and I provided my own manual inputs. When ChatGPT generated the content, I only took partial content and headings from it. For the User talk:Narayan Dikshit talk page message I provided the AI with my own manual message and asked it to help me add relevant Wikipedia guidelines to draft the reply. I understand the removal of my autopatrol rights during this review. My immediate focus will be manually cleaning up the NPOV tone in the articles mentioned. - Garudindflamewing ^(talk) 12:15, 10 September 2026 (UTC)

Thank you for your reply. I look forward to your cleaning up of the relevant articles. I would be grateful if you could also label the cleaned-up articles in the tracking page appropriately. Fermiboson (talk) 12:20, 10 September 2026 (UTC)

Special:Permalink/1309146234 (Sept 2025) had markdown btw Kowal2701 (talk, contribs) 15:46, 25 September 2026 (UTC)

User:Kagerou47

creations draftified and tagged

Would like to get a second opinion. Looking at Kagerou47 page creation log they are creating drafts within minutes of each other. For example, on August 31st, they created 6 drafts. @Guninvalid declined Draft:Daigaku Tsushin in part due to AI and I declined another but they have several pending review and some have been accepted. I do not want decline more if I am wrong but I did leave them a warning. S0091 (talk) 18:02, 9 September 2026 (UTC)

Unsure about complete AI generation. It's definitely strange for someone to create so amyn drafts at once but it is possible they had been working on them for some time before finally submitting. I also notice that almost all the topics are Japanese, and at least one of them has a translation on Japanese Wikipedia. So it's likely they are translating from Japanese Wikipedia, though I suspect undeclared WP:LLMT. guninvalid (talk) 23:31, 9 September 2026 (UTC)

I left them a reminder about unattributed translations and LLMT - it seems reasonable that these would be translations, but either way we'll hopefully be able to get some information on the editor's workflow. ^purring_maggot (talk) 01:18, 10 September 2026 (UTC)

Thanks, this is helpful. Will keep an eye out to see if they respond, even passively by adding attribution. S0091 (talk) 18:16, 10 September 2026 (UTC)

Hi there. Looks like a lot of things happened all of a sudden since my previous login. I just finished going through on what happened those draft that I have submitted before, and also went through the cases that appears here in this noticeboard.

I must admit that I did use AI tools to help me find references, prepare the draft structure, and adjust things a bit here and there.

I also admit that I am not aware the Wikipedia AI policy before this (and ironically become aware of this when me myself ended up in this noticeboard haha).

Sorry to trouble the contributors here, to undo my previous edits and submissions, and marking most of my submission as AI generated content accordingly. I do appreciate these efforts to keep the integrity of Wikipedia. I hope that similar standard also applied not only to English Wikipedia, but also other languages / versions of Wikipedia as well.

As a safeguard, I don't mind if I ended up permanently blocked from making future submission.

To wrap up, thanks again to the contributors and all these efforts. Cheers. Kagerou47 (talk) 15:53, 12 September 2026 (UTC)

Thank you @Kagerou47 for the response and your honesty, you won't be blocked as such measures are only to prevent disruption to the encyclopedia and never to punish (WP:NOTPUNITIVE). This isn't really your fault, Wikipedia needs to do a better job at informing newcomers of its AI policy. In case you wish to work on your drafts (listed here), some advice I can give: The biggest issue with LLM use is hallucinations (i.e. WP:V failures), though there are other important issues like WP:NPOV/WP:WTW and WP:COPYVIO. Usually we recommend that people stubbify or rewrite LLM-generated articles from scratch since it's often very time-consuming to attempt to preserve such content, but it's up to you. Please let me know if you have any questions or need help with anything. Best, Kowal2701 (talk, contribs) 16:00, 12 September 2026 (UTC)

Thanks @Kowal2701. I think I will step aside from those drafts for a while now (thanks as well for putting them in some sort of a quarantine btw).

Maybe for now, I simply want to see whether other contributors with similar interest will continue working on those as well or not haha.

Those are a bit niche topics I can say, especially the first one I submitted about Yoichi Aoyagi, one of the Japanese photographer from the 1970s. Then the subsequent submission, is kind of branching out from there, which I got carried away to use AI tools again to submit those as well.

About the hallucinations issue, I did try to check the provided links and references and did a bit of editing before the submission. I try my best to assure that the draft did not contain wrong information.

My main challenge for now, I would say is the source material, especially for a figure (such as Yoichi Aoyagi) or reading materials (such as those submission related to Japanese old magazines) that kind of famous at a particular period of time and to a particular group of people, which did not get enough coverage online or even at physical books at the recent times.

Now that you mentioned about stub, maybe in future if I did consider to continue the contribution again, I will consider to stubbify it first.

Thanks a lot for the guidance. Kagerou47 (talk) 16:25, 12 September 2026 (UTC)

No worries, creating articles is often the hardest thing for a newcomer to do, there's some advice at Help:Your first article in case you do decide to return to editing (WP:TEAHOUSE can also help with any problems you may have). There is also Help:Find sources which may be useful. Good luck! Kowal2701 (talk, contribs) 16:50, 12 September 2026 (UTC)

Special:Permalink/1371449374 is very different to the Japanese article, same deal with Special:Permalink/1372263711 and ja:GORO. Draftifying the rest and tagging, will subscribe to the convo as well. Oddly they don't edit jawiki Kowal2701 (talk, contribs) 00:54, 12 September 2026 (UTC)

can someone decline their AfC submissions? Kowal2701 (talk, contribs) 00:59, 12 September 2026 (UTC)

This article Draft:Sunday Mainichi does not have any obvious AI content and most of its refs are real and related to the content cited (there were one or two odd refs but that could be due to the source changing). I would inclined to move this particular article back to mainspace (haven't looked at the others. thanks. Aszx5000 (talk) 14:39, 12 September 2026 (UTC)

it's very different to the Japanese article, however doesn't really have any AI signs like the others (Pangram also gives it human-written, but has high false negatives). Unfortunately I think it'd still need spotchecks for WP:V before returning to mainspace, as the articles before it have AI signs + Pangram positives. Idk but I expect they are writing these themselves in Japanese, and then using an LLM to translate that (which'd still cause hallucinations and isn't covered by WP:LLMTRANSLATE) Kowal2701 (talk, contribs) 15:30, 12 September 2026 (UTC)

Some things bother me in Draft:Sunday Mainichi. During WWII [...] publication continued under the English-free title Weekly Mainichi. Given that the whole point was that the magazine wasn't using English at that time, it seems extremely odd to translate its name!

In 2008, the Imperial Household Agency lodged a protest over a February 3 article on Empress Michiko, saying the piece contained claims that were not factual. Machine translation of the source makes it sound like very small potatoes--more a correction than a rebuke--though I may not be able to interpret tone in Japanese affairs. But it sure does raise the question, what happened next? The article apparently has no interest in this.

The paragraph on Aum Shinrikyo is too-close paraphrase of its source. A few earlier sentences are also very close to this source (which is in English).

On the other hand, that source discusses two magazines, and as far as I can tell only material on the correct one has been used in this article. I would not expect pure LLM text to get that right; it's a very hard task for them. M kuhner (talk) 06:22, 22 September 2026 (UTC)

User:Capankajsmilyo

 Cleanup has been requested.

 Capankajsmilyo (talk · contribs · logs · block log) (13 September 2026)
Editor has acknowledged AI/LLM use, committed to stop, and will follow WP:NOLLM. Cleanup may still be needed. Autopatrolled permission has been removed from editor's account.

Autopatrolled user with 47k+ edits. Extremely suspicious editing pattern. e.g. the sequence of revisions starting at Special:Permalink/1373528950 in which 2000 bytes of prose are added in a span of 8 minutes. The sequence of revisions starting at Special:Permalink/1372157922 where nearly 3000 bytes of prose are added in a span of 7 minutes. Two separate unsourced stubs at Special:Permalink/1371558608 and Special:Permalink/1371559265 within 4 minutes of each other. Half an hour later (with continuous edits in between) two more unsourced stubs Special:Permalink/1371564229 and Special:Permalink/1371565136 6 minutes apart. Despite editing since 2015 the user only started tripping edit filters for broken sfns in late 2025. The added content itself is obviously full of AISIGNS as well. Kisseran (talk) 05:26, 10 September 2026 (UTC)

FWIW I put Special:Diff/1373528950/1373530000 through Pangram as a sanity check: . Notified them, also courtesy ping to @Jishara who's reviewing a GAN Kowal2701 (talk, contribs) 09:27, 10 September 2026 (UTC)

I was not aware of these new guidelines. I used LLM for english drafting and avoiding copyvio. I'll restraint use of LLM now. Thanks for pointing out @Kowal2701 Capankajsmilyo (talk) 10:05, 10 September 2026 (UTC)

Capankajsmilyo, can you let us know when you started using LLM? It would really help with cleanup efforts. M kuhner (talk) 23:17, 18 September 2026 (UTC)

Thanks for the courtesy ping, that explains a lot. Jishara (please ping upon response) (talk) 13:17, 10 September 2026 (UTC)

@Jishara For what it's worth, you could have quickfailed the GAN on criterion 3 since it needs a LLM cleanup banner. Kisseran (talk) 13:35, 10 September 2026 (UTC)

Good point - I didn't feel I should quickfail though because the GAN's already been going for a bit, so I regular-failed under Criterion 2 instead, since it's at least partially LLM-generated per Capankajsmilyo, many of the sources are unavailable online, and even before a spotcheck some of the facts didn't line up. Jishara (please ping upon response) (talk) 13:46, 10 September 2026 (UTC)

@Admins willing to patrol AINB: Can you revoke autopatrolled here as well? Kisseran (talk) 09:50, 13 September 2026 (UTC)

I have removed the autopatrolled permission from Capankajsmilyo's account. Capankajsmilyo is welcome to reapply for the permission at WP:PERM/A after requalifying for the guidelines for granting with 25 valid article creations (in addition to the articles he had previously created). — Newslinger talk 14:24, 13 September 2026 (UTC)

User:Baronmilone

creation draftified and drafts tagged

Cleanup efforts are ongoing and being tracked at /2026-09-12 Baronmilone, 278 pages need review.

Baronmilone (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Markdown block in . Many, many, many content additions to be tracked. Likely AITALK. Fermiboson (talk) 13:29, 10 September 2026 (UTC)

I have responded to the notice on my user talk page and will address any specific concerns raised here. Baronmilone (talk) 13:55, 10 September 2026 (UTC)

User:RogerYg

 Cleanup has been requested.

7k+ edits. ChatGPT tracking parameters in filter log on talk page messages. Other likely AITALK examples include Special:Diff/1356868700. Markdown in mainspace prose in Special:Diff/1356002131, accompanied by glaring AI prose like Special:Diff/1356096567. Rather conspicuous edit summaries that seem human-written but with typical AI emphasis on adherence to policy. Kisseran (talk) 13:54, 10 September 2026 (UTC)

final warning as they denied AITALK Kowal2701 (talk, contribs) 01:16, 12 September 2026 (UTC)

they denied it again (not just currently, never ). Markdown mentioned above was in May. I guess we wait for more AI edits and then ask for a block. Kowal2701 (talk, contribs) 21:20, 15 September 2026 (UTC)

I am happy to explain and clarify any AI charges. I can assure you that I am not using AI for my Wiki page edits. There may be very few isolated incidents where I searched for some WP policy on Google Search in AI mode or ChatGPT, which I may have used in a TALK page discussion by mistake.

Hi @Kisseran and @Kowal2701 I can assure you that I will not use AI LLMs for Wiki edits in the future. I have been editing long before AI LLMs were invented, and I have a very good record with 7K edits. Thanks. RogerYg (talk) 21:53, 15 September 2026 (UTC)

User:Guywelch2000 (2)

Cleanup efforts are ongoing and being tracked at /2026-08-18 Guywelch2000, 154 pages need review.

Previous AIN thread: Wikipedia:AI noticeboard/Archive 13 § User:Guywelch2000

 Guywelch2000 (talk · contribs · logs · block log) (10 September 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. 

Previous AIN guest and somewhat prolific user. Made substantial edits to Wawona Hotel today. The changes were first worked out in the user's sandbox, where there are chatbot blocks in f.ex. Diff/1374144270 and Diff/1374146823 from today. ‑‑gurkubondinn 14:04, 10 September 2026 (UTC)

Just a quick note that I have acknowledged the warning. Guywelch2000 (talk) 14:52, 10 September 2026 (UTC)

Most of the pages that were draftified the last time have been moved back into article space, and the {{ai-generated}} tags have been removed. ‑‑gurkubondinn 14:54, 10 September 2026 (UTC)

I contributed to many articles in 2024 that were moved into draft space in early August. I spent the last few weeks revisiting them, checking the original sources, and revising the articles based on the issues identified, which often came down to clumsy, verbose prose that I was very motivated to fix. Also, I found many issues typical of first draft articles that were a side effect of my research; extraneous detail, over-sectioning, and so forth.

My goal is that these articles move beyond Start class, provide knowledge, and are a foundation for others to improve. I have used AI technology to proof my grammar and target weak spots in my writing, which tends towards the promotional, stating the significance of something instead of letting the facts speak for themselves. I have also found AI technology to be helpful in spotting discrepancies between sources. An example of this is articles that cite federal documents over several decades, where I have corrected errors and updated citations to go with what seems like the best and most authoritative source. This was the case with the Wawona Hotel article.

I love this space and genuine effort in making it better through my contributions. For now, I will both take a pause and revisit the editing guidelines. Thanks for your patience and consideration. Guywelch2000 (talk) 02:33, 11 September 2026 (UTC)

User:Campaigner1942

Campaigner1942 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Warned for LLM use in June, but the latest AfC accepts show that they haven't stopped. No time to do comprehensive analysis yet; I have a list of like 40 possible open cases in a text file right now. Please take a look. Fermiboson (talk) 15:39, 10 September 2026 (UTC)

hasn't edited since June, chatgpt-derived link in Special:AbuseLog/44387834 (June). Creations are very plausibly AI but no major signs at a glance. Pangram says mixed and AI , I suspect this user was more clueful re AI detection than our usual customers.

Spotchecks of Special:Permalink/1353543670

• Working with Friends of the Earth, the Association for the Conservation of Energy and cross-party MPs, Bailey participated in campaigns supporting Diana Maddock's Home Energy Conservation Act 1995, which required energy conservation authorities to improve the energy efficiency of residential accommodation in their areas. is sourced to a primary source and is typical AI OR
• Bailey was involved in campaigns for freedom of information and local government transparency during the 1980s. He contributed to discussions surrounding access to local authority information and published work on official secrecy and municipal governance. is completely unsupported by

These were the only two I checked, I've seen enough, LLMPRODing Kowal2701 (talk, contribs) 01:30, 12 September 2026 (UTC)

LoreGatherer2 (talk · contribs) edited at least two of their creations, was created in July and has triggered Edit Check + filter 1325 . Another potential one is The Lexicographist (talk · contribs) who added an image to Campaigner's creation. Looks like a UPE farm, pinging @NicheSports in case they fancy a crack at this? (I suck at SPI) Kowal2701 (talk, contribs) 01:36, 12 September 2026 (UTC)

Yes, this is worth an SPI filing, plus I found another. I'll ping you NicheSports (talk) 02:16, 12 September 2026 (UTC)

Sorry for the delay, I am still working on this one. I will have an update in the next day. NicheSports (talk) 13:16, 17 September 2026 (UTC)

Ok, the SPI was filed at WP:Sockpuppet investigations/The Lexicographist. This identified several more accounts: Middlesbroughmight, I-squared-you-squared, Scrutinzer1978, Dartycrossing73, Greenadvocator, Knockonwood1981 that are confirmed to each other and most of the accounts discussed here. LoreGatherer2 is a newer account that requires behavioral adjudication. I have already cleaned up many of the edits from this cluster, but I'd suggest that we wait on the SPI case to be closed before doing further cleanup. NicheSports (talk) 12:19, 22 September 2026 (UTC)

User:Shivaji Mitra

 Cleanup has been requested.

Shivaji Mitra (talk · contribs · deleted contribs · logs · filter log · block user · block log)

See recent contribs. Behavior including: edit summaries like "Draft article on Rajive Raturi v. Union of India (Supreme Court accessibility case) with sources; to be moved to mainspace after review" followed by moving to mainspace 1 minute later, oneshot 15k article 13 minutes after last edit, oneshot 27k article 26 minutes after last edit. ~ A412 ^talk! 15:52, 10 September 2026 (UTC)

Special:Permalink/1178991287 (Oct 2023) is blatantly AI-generated, 2026 creations are less obvious. Have warned for the first time Kowal2701 (talk, contribs) 01:41, 12 September 2026 (UTC)

I have replied on my talk page at User talk:Shivaji Mitra#September 2026. Shivaji Mitra (talk) 03:12, 12 September 2026 (UTC)

just noting: they were very responsive and good-faith, looks like they've done the cleanup themselves, haven't time to review though Kowal2701 (talk, contribs) 19:12, 26 September 2026 (UTC)

User:Wilfredor

 Cleanup has been requested and is being tracked at /2026-09-19_Wilfredor, 61 pages need review.

Wilfredor (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Blatant AI edit summaries. At a glance the drafts look LLM and edit pacing is weird. Again, have not looked in detail yet. Fermiboson (talk) 16:53, 10 September 2026 (UTC)

Special:AbuseLog/42160183 (Sept 2025) had a chatgpt-derived link, on 31 May 2026 they created 62 articles (about a dozen more were draftified and failed AfC, one got G15'd, all appear to be loose LLM translations from eswiki). Last edited 17 August. I suggest LLMPROD 31 May creations to start with Kowal2701 (talk, contribs) 22:56, 10 September 2026 (UTC)

More info about this situation can be found here. I'm not using any LLM anymore ever for translation Wilfredor (talk) 03:37, 11 September 2026 (UTC)

Thank you for agreeing not to use LLM anymore! If you could give us the dates you started and stopped it would really help with cleanup. M kuhner (talk) 03:57, 11 September 2026 (UTC)

Consensus needed to LLMPROD the 31 May creations, I'm not sure about the few previous ones but those are the priority imo Kowal2701 (talk, contribs) 22:27, 16 September 2026 (UTC)

Support May 31 LLMPRODs. M kuhner (talk) 01:05, 17 September 2026 (UTC)

Made an (approximately) May 31 tracker. Can extend to other dates if found to be needed. M kuhner (talk) 01:31, 19 September 2026 (UTC)

Done LLMPRODs and tagged drafts for 31 May, unsure if LLM use extended beyond that Kowal2701 (talk, contribs) 19:33, 26 September 2026 (UTC)

User:Fabal77

Fabal77 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Pacing is highly suspicious; user has created 26 articles in a single day, with a gap of just two minutes between a couple of them. Some articles have redlinked categories, but not too many other obvious AISigns have stood out to me, so requesting input from all you experts here. In solidarity, nil ^nz 00:34, 11 September 2026 (UTC)

Dmitri Bogolepov in addition to the REDCATS, was created with three sources, two of which are WP:CIRCULAR (one explicitly, the other discloses on its about page). This isn't an LLMTRANSLATE situation because the Russian article doesn't use said sources, has somewhat different content and there would have either been more red categories or the editor would have checked them and had none. Similar situations with several more of these (this one actually directly cites Russian Wikipedia). I think I'm going to draftify all these. ^purring_maggot (talk) 01:06, 11 September 2026 (UTC)

The one I examined, Iosif Korenevsky, is a translation from Russian Wikipedia. I looked at the original with Google Translate and it appears to be a faithful one. It is cited to ruWiki, not a proper translation notice, but at least this gives means to locate the original.

The redcats don't match, though: the enWiki article had, at creation, redcats People from Zaslawye, Belarusian State University people, Rehabilitated Soviet people; none of these cats is on the Russian article. M kuhner (talk) 01:08, 11 September 2026 (UTC)

Were you possibly thinking of Draft:Alexei Argamakov? This was the only direct translation I noticed - Iosif Korenevsky cites academic.ru, which itself cites Wikipedia, but it doesn't cite the Russian Wikipedia article and the content and structure of it is a little different.

Further analysis:

• Draft:Anzhalika Melnikava contains WP:AIATTR. Some of the attribution may be appropriate given the events described, but it sneaks in "according to colleagues", "some Belarusian commentators", and "Belarusian media".
• The user's only previous creations are Dainius Gaižauskas and Rimas Jonas Jankūnas which are direct (unattributed) translations from dewiki.

All of the 10 September creations are tagged and draftified now, and I will leave a warning for the user. ^purring_maggot (talk) 02:34, 11 September 2026 (UTC)

No, I was looking at Iosif, but I assumed academic.ru was serving as a mirror of ruWiki due to the citation saying "Academic.ru, citing Wikipedia (Russian edition)". When I machine translated the academic.ru page it looked pretty similar to the article, many sentences nearly verbatim. M kuhner (talk) 02:59, 11 September 2026 (UTC)

Oh, I see now. The attribution on academic.ru is dated 2010, so it must have been an earlier version of the wiki article. ^purring_maggot (talk) 03:14, 11 September 2026 (UTC)

yeah, real suspicious. It's wait it out to see if they do more of this, then warn. | One Reaction was here. Got a complaint? 14:38, 11 September 2026 (UTC)

User has removed AI-generated tags from two of the drafts without acknowledging my post on their talk page. This edit comes off the most suspicious to me - shuffles around every single reference, removes the draft categories and drafts moved to mainspace templates, and for some reason rephrases my reason in the AI-generated tag (which the editor removed in the next edit). ^purring_maggot (talk) 18:05, 11 September 2026 (UTC)

User:Electragod77

Cleanup efforts are ongoing and being tracked at /2026-09-16 Electragod77, 157 pages need review.

2k+ edits. Extremely fast edit pacing e.g. Special:Permalink/1358301594 and Special:Permalink/1358304739 4k in 20 minutes. Prose full of AISIGNS and has even tripped edit filters for plain, obvious AI vocabulary. Kisseran (talk) 11:03, 11 September 2026 (UTC)

also previously warned for copyvio and deprecated infobox parameters. Just given first AI warning Kowal2701 (talk, contribs) 01:48, 12 September 2026 (UTC)

their creations were between 2022 and 2023 and aren't AI-generated, triggered filter 1325 twice in June . They effectively took a break from 2023 until April 2026, after which they made a ton of rapid edits (a lot of files created, I've no clue what to do w those). Most edits are minor (adding files etc.), quotes, or infoboxes. I'll create a tracker excluding small ones Kowal2701 (talk, contribs) 22:40, 16 September 2026 (UTC)

User:BoonDock

 Cleanup has been requested.

Impossibly fast editing pace and triggered AI filters. Text has every AISIGN under the sun. I've G15-nominated South African Army Vision 2020 but there are still many to look at. Kisseran (talk) 11:52, 11 September 2026 (UTC)

15k edits and account created in 2005, asked when they started using AI (appears to just be Sept?) Kowal2701 (talk, contribs) 01:53, 12 September 2026 (UTC)

they responded w Kowal2701 (talk, contribs) 15:45, 12 September 2026 (UTC)

On September 9 they created a new module and then put it into dozens of articles at a pace which must represent some form of automation (multiple pages per minute); they then went back and fixed pages it broke (at least, according to edit summaries). On September 10 there is more of that type of activity, but also large new pages start appearing at a rapid rate. I don't know what process produced the Sept. 9 edits and don't feel confident it's LLM as opposed to some other semi-automated process. They are difficult to check as it's all tables. I will spot-check some Sept. 10 ones, but that may be the start date. M kuhner (talk) 00:14, 28 September 2026 (UTC)

Creation version of Draft:South African Defence Review 1998 on Sept. 10 has WP:AIATTR and WP:AIBOLD (as well as a weirdly misleading edit statement that says it's creating a mainspace article, which it wasn't).

I have concerns about the large edit to 2nd South African Infantry Division on Aug. 23, particularly The full text of this operation instruction was not reproduced in the published intelligence bulletin. Not sure though; would appreciate someone else looking at this edit.

Editor did not seem to make enormous, rapid edits prior to that, at least not in the last 500 or so. Very dramatic change in editing pace and style around September 2026. M kuhner (talk) 00:33, 28 September 2026 (UTC)

User:金色黎明

Cleanup efforts are ongoing and being tracked at /2026-09-19 金色黎明, 66 pages need review.

LLMPRODs are done, no drafts to tag

 金色黎明 (talk · contribs · logs · block log) (14 September 2026)
Editor has been blocked site-wide for AI/LLM policy violations. WikiCleaner1013 (talk · contribs) has also been blocked as a sockpuppet of 金色黎明.

金色黎明 (talk · contribs · deleted contribs · logs · filter log · block user · block log) This user has been blocked for persistent addition of AI slop on Chinese Wikipedia. Just out of curiosity, I roughly checked their contributions, and found they also used a LLM/AI to make edits here.

• 43045940 Blatant ChatGPT track parameter.
• Cytoplasmic-nuclear male sterility (GPTZero AI score = 1.0) It appears that this article contains a total translation of their creations, zh:核质互作雄性不育, from Chinese Wikipedia. The latter is obviously generated by AI and tagged with G15 by an admin (zh:special:diff/94274371). The former has many signs of WP:AITREND. (e.g. The term emphasizes that sterility is not determined by cytoplasm alone, but by disharmonious interaction between cytoplasmic sterility factors and the nuclear genome. and Cytoplasmic-nuclear male sterility is important in plant breeding because male-sterile plants can be used as female parents without manual emasculation. It has been used in hybrid seed production in crops such as rice, maize, sorghum, pearl millet, rapeseed, sunflower, sugar beet, onion and pepper. --椿 (talk🎀) 12:15, 11 September 2026 (UTC)

Thank you @椿, several of their zhwiki articles are tagged for speedy deletion, the earliest appears to be 17 Jan 2026. The zhwiki cleanup page is rather long. I see several of their GAs are being reassessed (zh:维基百科:2026年AI调查), they also have 2 GAs here that are potentially AI: 2024 Taiwanese President of Legislative Yuan election (Dec 2025 - June 2026) and 1984 Indiana's 8th congressional district election (Jan - Feb 2026). The chatgpt-derived link was Dec 2025, triggered filter 1325 in Jan 2026, looking at their creations:

• Nova Roma (video game) (Feb 2026) has WP:AIDASH, WP:AIVOCAB ("highlighted", "emphasized"), WP:AICONNECT ("combining"), overeliance on primary sources
• 1984 Indiana's 8th congressional district election (Jan 2026, now a GA) has WP:AIDASH, WP:AITREND ("reflecting"), extrapolated WP:AIATTR re "contemporary reporting", Pangram says AI

I'll WP:LLMPROD back to Jan 2026 for now, need to wait for Pangram credits to go further back lol (just for sanity checks ofc) Kowal2701 (talk, contribs) 19:22, 12 September 2026 (UTC)

On to the two GAs (analysing early revisions):

• 1984 Indiana's 8th congressional district election (Jan 2026) had AIDASH, AIVOCAB ("emphasizing"), SUPERFICIAL (...reflecting the extent to which national party organizations treated the seat as a proxy battle over the legitimacy of the 1985 House decision) + Pangram positive , I've tagged it, courtesy ping @Thebiguglyalien as reviewer: idk whether this ought to go to GAR or just LLMPRODed, tbf your spotchecks didn't bring up any hallucinations apart from maybe one
• 2024 Taiwanese President of Legislative Yuan election (Dec 2025) had AIDASH, AITABLE + Pangram positive , I've tagged it, courtesy ping to @ZKang123 as the reviewer: same above, similarly your spotchecks didn't turn any major hallucinations

Kowal2701 (talk, contribs) 14:59, 14 September 2026 (UTC)

I believe the AI-ness might be due to them translating from Chinese sources to English. Much of the coverage is in Chinese and my spot checks generally can tell that they support the prose. Also, I recommend using GPTZero. Disappointed to learn that said user has been using AI to pollute the Chinese Wikipedia.--ZKang123 (talk · contribs) 00:30, 15 September 2026 (UTC)

They were translating zhwiki pages they created which AFAICT are determined to have been AI-generated (I can't read/detect AI wrt Mandarin). Personally I've found GPTZero to be pretty unreliable, Pangram has high false negatives but a positive from it is usually a strong sign. That very few WP:V failures were found is odd, the articles are pretty short (~1000 words) so maybe they did just review it all for hallucinations before submitting to GAN Kowal2701 (talk, contribs) 11:32, 15 September 2026 (UTC)

Some more analysis:

• Alexander McIver (Sept 2025) had AIDASH, AIPUFFERY (A noted mathematician...), SUPERFICIAL (His career illustrates the intersections of education, governance, and rural reform in nineteenth-century North Carolina.), some AICURLY, AITREND (Alexander McIver played a central role in reshaping North Carolina's education administration), AITITLECASE, no need to waste Pangram credits on this
• Tom Arceneaux (Feb 2025) was the one before that and had SUPERFICIAL (His considerable engagement in civic activities encompasses participation in various organizations...), AIVOCAB ("emphasized"), but no other signs, Pangram says AI-assisted

Need to wait for more Pangram credits again :/ Kowal2701 (talk, contribs) 15:12, 14 September 2026 (UTC)

• United States presidential straw polls in Puerto Rico (Dec 2024) has AIREDCAT, AILEGACY ("marking a significant milestone") + Pangram AI-assisted .

Start date appears to be late-2024 (prior creations look fine), looking at ordinary edits I can't tell whether LLM use went beyond Dec 2024 (@Fermiboson, could you tell zhwiki this?). @M kuhner, is there a way to extend the tracker that idk of (back to Dec 2024) Kowal2701 (talk, contribs) 13:43, 26 September 2026 (UTC)

Yes. You proceed (carefully--there is an easy way to make a bad mistake here!) as if making a new tracker with the required dates, but don't actually create the tracker, instead using "copy tracker code to clipboard". Then you edit it onto the end of the existing one. I'll give it a try, though I'm not the brightest right now. M kuhner (talk) 14:55, 26 September 2026 (UTC)

 Done M kuhner (talk) 15:03, 26 September 2026 (UTC)

(I notified zhwiki of this thread at zh:维基百科讨论:2026年AI调查/User:金色黎明#英文维基百科 (English Wikipedia)) Kowal2701 (talk, contribs) 19:37, 12 September 2026 (UTC)

Thanks for pointing that out. I've notified @金色黎明 myself since they didn't join the discussion, although we will get no response from them, I think. 椿 (talk🎀) 10:18, 13 September 2026 (UTC)

@Kowal2701 @椿 User:金色黎明 is now gone offline in both zhwiki and enwiki to evade this issue, which is a definite nonsense. Please consider giving the user a full indefinite block as that as in zhwiki. Пусть от победы☆к победе ведёт! 10:54, 14 September 2026 (UTC)

Neither of us are admins, but I can ping @Anachronist. Usually we don't block someone who goes inactive, but I've seen in cases of WP:ANIFLU where a mainspace block is given to make them address concerns before returning. Will ofc defer to Anachronist Kowal2701 (talk, contribs) 11:57, 14 September 2026 (UTC)

@Admins willing to patrol AINB: Now we've got a DUCK @WikiCleaner1013 who has just registered and tried to remove this thread and the LLM tags. Blocks are now very appropriate. Kisseran (talk) 14:20, 14 September 2026 (UTC)

Done. Both blocked. ~Anachronist (who / me) (talk) 14:29, 14 September 2026 (UTC)

Quite a lot of (presumably socks) blanking this section.. ‑‑gurkubondinn 12:51, 16 September 2026 (UTC)

yeah they're socks, feel free to report any you see to m:Steward requests/Global for lock evasion Kowal2701 (talk, contribs) 13:18, 16 September 2026 (UTC)

Created WP:Sockpuppet investigations/金色黎明. Srapoj (talk) 13:44, 17 September 2026 (UTC)

Made a tracker only to Jan 1 2026 covering both accounts. Can extend if deemed necessary. M kuhner (talk) 02:00, 19 September 2026 (UTC)

University of Sydney course mass AI use across at least two years

Needs reviewSoroban

Needs reviewTokimeki Memorial 4

KeptInca Kola  The most significant error (adding "s" after specific years instead of decades) has been fixed already.

Needs reviewList of One-Punch Man characters

Needs reviewLingda

Kept1770s  Fermiboson  The original is way too bad to revert to, and the rewrite is at least factually correct by common sense, so might as well.

Needs reviewCage the Elephant

Needs reviewFeature (machine learning)

Needs reviewC. W. Smith (writer)

Needs reviewOverachievement

Needs reviewInternational sanctions against North Korea

Needs reviewJulius Caesar's invasions of Britain

Needs reviewCasper (octopus species)

Needs reviewCivil Lines (magazine)

See Wikipedia:Education noticeboard#University of Sydney undisclosed course with mass AI use for detailed analysis. We have here an unknown number of student accounts, thus far 100% of which have used LLMs, and which are still not all cleaned up from 2025, and this year's course is happening right now. Eyes needed. Fermiboson (talk) 16:40, 11 September 2026 (UTC)

Special:Diff/1374354626 isn't good, but it doesn't seem AI, so the 100% figure can be revised a little bit. Not enough, but it's not all AI slop out there. Kisseran (talk) 16:59, 11 September 2026 (UTC)

I don't know, The first-hand records at that time showed that this kind of friendship had common hobbies, emotional support and intimate physical interaction is suspect and I am quite confident the summary is AI. Relatively moot in either case. Fermiboson (talk) 17:02, 11 September 2026 (UTC)

Records left over from history record a large number of emotional words in letters and diaries in men's romantic friendships stood out to me as being not AI, but looking at the rest of the edit and the edit summary does indeed not spark any confidence. Kisseran (talk) 17:09, 11 September 2026 (UTC)

(any updates?) Kowal2701 (talk, contribs) 23:02, 16 September 2026 (UTC)

There was more discussion at EDUN where Fermiboson reported yesterday I am happy to report that the three latest revisions by students do not read obviously as AI generated. As to the cleanup work that remains I'll report on AINB. —ClaudineChionh (she/her · talk · email) 23:52, 16 September 2026 (UTC)

thanks, maybe a list of users whose contribs need cleaning? Kowal2701 (talk, contribs) 23:57, 16 September 2026 (UTC)

Check everything in , maybe. Fermiboson (talk) 21:12, 17 September 2026 (UTC)

Forked to cover just this year's course if we want to prioritise: quarry:query/109392. —ClaudineChionh (she/her · talk · email) 00:21, 18 September 2026 (UTC)

I've started a small list of what looks like high priorities for investigation. Don't know why Dr vulpes's script placed the list at the bottom – maybe because there isn't a single user being tracked? Can DVRTed's script handle a situation like tnis? —ClaudineChionh (she/her · talk · email) 14:38, 18 September 2026 (UTC)

Kind of. There's a fresh new update enabling the creation of a unified tracking table for multiple users. It's very fresh and... "unified" (as in everything is crammed into a single table!). See User talk:DVRTed/AINB-helper § Multiple users (sockfarms). — DVRTed (Talk) 14:51, 18 September 2026 (UTC)

User:EZPopoluv

 Cleanup has been requested.

• EZPopoluv (talk · contribs · deleted contribs · logs · filter log · block user · block log)

 EZPopoluv (talk · contribs · logs · block log) (30 September 2026)
Editor has been blocked site-wide for AI/LLM policy violations. 

This is a referral from an arbitration enforcement request (permalink) within WP:CT/A-I that I recently closed. Per my comment in the discussion:

...it appears that EZPopoluv's editing process is:

1. Use an LLM to generate a proposed edit to an article
2. Use a private script to format the LLM-generated edit as a "staging" page and create it under User:EZPopoluv/staging/
3. Use the script User:EZPopoluv/common.js to load the LLM-generated changes from the "staging" page into the wikitext editor for the target article
4. Review the changes
5. Submit the changes

As convoluted as the "staging" process is, this workflow is functionally the same as having an AI chatbot generate an edit, then copying and pasting it into the wikitext editor without creating a "staging" page. EZPopoluv said, "After receiving feedback about making too many quick consecutive edits, I set up a local script to publish directly via the Wikipedia API to streamline my workflow", which indicates that their edits prior to creating the "staging" process were also LLM-generated.

— Newslinger talk 19:00, 11 September 2026 (UTC)

triggered Paste Check multiple times early on, but their article and draft creations don't really have strong AI signs (ofc they're all very plausibly AI). Needs a closer look Kowal2701 (talk, contribs) 22:34, 12 September 2026 (UTC)

currently editing, spotchecks needed here Kowal2701 (talk, contribs) 18:58, 23 September 2026 (UTC)

they're still doing this, zero change in behavior or indication that they care what anyone else thinks; @Admins willing to patrol AINB: may need to step in Gnomingstuff (talk) 08:01, 30 September 2026 (UTC)

Blocked. Special:Diff/1377462264 has information not supported by the cited source, as well as outright unsourced information. Normally this might result in another warning, but with clear evidence of LLM use and a failure to communicate, I see no reason for further leniency. Toadspike [Talk] 09:34, 30 September 2026 (UTC)

Draft:Historical accounts of homosexuality in African societies

Hi.@Fifteen thousand two hundred twenty four@Gurkubondinn@ChrysGalley@Kowal2701@Anachronist@M kuhner@AndyTheGrump

I've improved this article in the areas it was lacking in the previous version []. I've fixed paraphrasing problems. Each line is stated in my own words, but still loyal to what the source says. I've restructured how the section should look. I took in the advice of some users to change the title name to better capture the content inside of the article. It's highly attributed. I've fixed the citation number issue and used secondary sources in some sections and primary in others. The new title name resolves the issue of WP:due that was invoked, and the lead section.

There was no llm use here. It was drafted by me. Broadly I did explain I did use llm to ask questions about societies that had homosexuality in Africa. That was it. Its the only use. Llm can't read books. No source online that is reliable exists that can be as specific as this article. Llm needs the direct material to read what's inside. It can't read a book. They are behind paywalls and some are not even recorded in txt. You can see there is no hallucination from any llm. The names, places are all names used in the books. I do not believe Ai can do that, based on my personal experience with Ai, especially on African topics with barely any coverage.

So I'm here to notify the users previously involved in the other discussion and a vote took place I knew I couldn't just send to mainspace. Sending it to mainspace is my goal here, after I've properly requested for the mainspace title unbanned. So I'm curious to hear your thoughts and what you consider to be still problematic on this updated version. Omniking2100 (talk) 20:21, 11 September 2026 (UTC)

I think sending it to Articles For Creation would be appropriate. If it is accepted, any issues with the title can easily be dealt with at acceptance.

I will read it now and try to put comments on the Talk page. M kuhner (talk) 00:41, 28 September 2026 (UTC)

User:Rednaxela1962

 Cleanup has been requested.

We in the german AI/Cleanup have a report about this user. According to Pangram, this user is adding ai-generated and ai-translated text to articles and create articles with ai, but not always so we don't know when this user started it to use. I found an edit of this user from March 17th 2025 that according to Pangram (90%) added ai-generated text to the article. The suspicion of ai involvement is there and it could affect even more articles. Skranon (talk) 20:51, 11 September 2026 (UTC)

There are UTM parameters showing ChatGPT use in Diff/1285617441 (AbuseLog/40517587) and Diff/1285619749 (AbuseLog/40517732). ‑‑gurkubondinn 21:48, 11 September 2026 (UTC)

Special:Permalink/1280840675 on 16 March 2025 appears to be AI-assisted . Some creations even as late as 6 September were done with the mw:Content translation tool, but I'm struggling with detecting signs. In the edit you linked above, they use straight quotes, while they usually use curly quotes (including in the edit I've linked). I see you've already asked them , that looks like our best bet unfortunately, idk what to think Kowal2701 (talk, contribs) 23:00, 12 September 2026 (UTC)

On the german AI/Cleanup the user said he is using AI for one year. But the article I linked here in my report is from may 2025. Skranon (talk) 06:14, 15 September 2026 (UTC)

I'm not all that bright, their creations were all translations from dewiki. 5 sizeable creations on 23 March 2025, some within 20 mins, which seems implausible without LLMs or machine translation (@Skranon, this might be helpful). For enwiki: really the only way to tell whether these comply with LLMTRANSATE is spotchecks unfortunately Kowal2701 (talk, contribs) 23:13, 16 September 2026 (UTC)

The user clarified that he uses AI since March/May 2025. The english article Jean Violet created in May 2025 is a translation from the german article. According to Pangram both german and english Article have ai-generated text. Even if they are translations, the german article can have ai-generated text and can be transferred with a translation to the english article. --Skranon (talk) 04:36, 17 September 2026 (UTC)

According to Pangram, these articles have ai-generated text:

Jean Violet, Kuwait–United Kingdom relations, Niger–Russia relations, Luxembourg–Netherlands relations, Austria–Bosnia and Herzegovina relations, Jaroslav Haščák Skranon (talk) 12:19, 17 September 2026 (UTC)

thanks, those are May to October 2025. Pangram does give a lot of false negatives (I assume especially wrt translations), but it's positives are usually reliable. Oddly one of the 23 March translations are coming back as human-written. I haven't got the time to do the spotchecks for some of these to see if they comply with LLMTRANSLATE, it may have to go to the backlog Kowal2701 (talk, contribs) 14:11, 17 September 2026 (UTC)

User:Martinpri

Martinpri (talk · contribs · deleted contribs · logs · filter log · block user · block log)

 Martinpri (talk · contribs · logs · block log) (13 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. User was blocked from article space and then unblocked in User talk:Martinpri § September 2026.

Previously warned for LLM use, they have clearly not stopped. is blatant AITALK (in my opinion, Claude). Oren Dobronsky is not blatant but has some amount of signs, mostly vagaries. Edit summaries are LLM. Fermiboson (talk) 03:59, 12 September 2026 (UTC)

Definitely did not stop, look at Diff/1357486051.

• Reference nr. 10 has placeholders ([REVIEWER LAST NAME FROM BMCR PAGE], [FIRST NAME]).
• Reference nr. 13 has the wrong title, which has not changed since the page was archived in 2024.
• Reference nr. 5 does not exist, but has been saved on archive.org.

The user ignored the first warning in February 2026, but then acknowledged the second warning in April 2026 and claimed to have since stopped using it.

@Admins willing to patrol AINB: the warning in April 2026 made it clear that they would be blocked if they continued. They have continued, so I think that they need to be blocked from at least article space. ‑‑gurkubondinn 15:02, 12 September 2026 (UTC)

pblocked. ARandomName123 (talk)^{Ping me!} 00:34, 13 September 2026 (UTC)

Hi everyone I saw the discussion here and on the other page. I believe the block I received is unwarranted. Following the warning several months ago, there is not a single edit done by me that has not been written, reviewed and links verified by me. I saw the few links someone flagged as evidence of LLM or AI use. One of the links I believe is an article that has simply been taken offline after I used it . One has a wrong title probably my mistake.

Anyway, I took the warning seriously. As mentioned, all information is doublechecked by me and verified. I don't believe you'll find any false or inaccurate information on my edits (and if so, it does not represent the majority of the hundreds of edits I have done), and I believe they contribute to the project. I use AI agents from time to time to look for links or information in cases when Google or other online sources yield very little information. Or to consult on certain issues. But as mentioned, I review all links and edits entirely to ensure the sources exist and are valid (And relevant for the edit).

Since the warning several months ago, I've done several hundred edits with probably many hundreds or over a thousand links and citations. I believe a look at them will show they are all or mostly all valid and accurate.

I would like to request you reconsider and that the block be lifted. I am very happy and proud to be part of the community and contribute, and I would like to continue doing so. Again, given the above, I will be even more vigilant in verifying all links and content, but as mentioned the vast majority is my work and not AI, and I'm pretty sure you can verify that. Thanks for the consideration.

Martinpri

Martinpri (talk) 22:59, 14 September 2026 (UTC)

@Martinpri: How did you end up with placeholder names for the reference above? In addition, was the talk page comment linked above AI-generated? ARandomName123 (talk)^{Ping me!} 15:22, 16 September 2026 (UTC)

Hi.

I'm not sure about the placeholder - I honestly don't remember but as you can see it's a fluke and a one-off. I did a spot check on my work going several months back and haven't found any similar issues. As you can see, also from the articles, there are almost no mistakes or issues on my links other than this. Again, I honestly don't remember why this link is like this. It's not typical of my work, a can be seen). Martinpri (talk) 08:10, 17 September 2026 (UTC)

Regarding Oren Dobronsky, the article was translated from Hebrew (with the use of AI, which I understand is allowed), and then edited by me, including addition of info and sources not in the original page in Hebrew. And I rechecked the article - all links accurate, working, properly formatted. Martinpri (talk) 05:42, 18 September 2026 (UTC)

Do you speak Hebrew? Fermiboson (talk) 05:58, 18 September 2026 (UTC)

Yes, I do. Martinpri (talk) 10:25, 18 September 2026 (UTC)

Martinpri, the procedure for asking for a block to be lifted is given in the block message on your Talk page. I suggest you post your reasoning there, not here. Unblock requests on AINB won't be seen by admins, and non-admins like the volunteers here can't do anything about them. M kuhner (talk) 02:14, 19 September 2026 (UTC)

Thanks, @M kuhner. The block message in my Talk Page said to use the Unblock Wizard. I did. Is there anything else I need to do? Martinpri (talk) 05:14, 22 September 2026 (UTC)

I don't think there's anything else you need to do. There is a backlog at the unblock requests noticeboard, so it may take a little while for an admin to get to your request. M kuhner (talk) 05:18, 22 September 2026 (UTC)

The user is now unblocked. ‑‑gurkubondinn 15:33, 23 September 2026 (UTC)

Can we find a start date for tracking? M kuhner (talk) 02:17, 19 September 2026 (UTC)

@Martinpri: can you tell us when your AI use began? ‑‑gurkubondinn 15:35, 23 September 2026 (UTC)

User:Kskhh

 Kskhh (talk · contribs · logs · block log) (13 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Created PakID with a chatbot block and AIREDCAT in Diff/1374667063 and then partially removed the chatbot block in Diff/1374667134, I've requested G15 speedy deletion. Has also created Muvi Cinemas in Diff/1372996767 (AbuseLog/45032343) with four AIREDCATS and non-existing templates. Created Dhurandar Part 2 with protection templates in AbuseLog/42937161, deleted after Wikipedia:Articles for deletion/Dhurandhar: Part 2. Has edited before the public release of free-access chatbots, so it is unclear when AI use began. ‑‑gurkubondinn 12:03, 13 September 2026 (UTC)

Also created Hunger Station in Diff/1374665489 with two AIREDCATS and a pretty irrelevant link to Jahez (a redirect to Dowry system in India) in § See also. The article reads very promotionally and was created c. 13 minutes before the PakID article. ‑‑gurkubondinn

Almost the same with ARY Plus, submitted less than 20 minutes before Hunger Station in Diff/1374649592. There's two AIREDCATS, duplicated references, redlink in § See also, and it is very promotional. ‑‑gurkubondinn 12:46, 13 September 2026 (UTC)

In Feb, March, and Sept they added non-existent templates, which they never did before. 2026 PIMS hospital fire (August 2026) rings as human-written to me, though there's The incident was among the deadliest hospital fires in Pakistan in recent years, highlighting concerns over the country's under-resourced and overburdened public healthcare facilities. (AITREND/AIVOCAB), Pangram gives a partial positive (tagged this one). Muvi Cinemas (3 Sept) had AIREDCATs (have draftified that one). That's all there is wrt creations, finding a start date for ordinary edits looks like it's going to be tricky Kowal2701 (talk, contribs) 15:10, 17 September 2026 (UTC)

User:Benigo36

 Benigo36 (talk · contribs · logs · block log) (13 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Seemed to have created Mick Garvey, Madge Harvey, Gavin Ramsbottom, Tiger Dyke and Geoff Maltby, large articles over several days. There could more. scope_creep^Talk 14:42, 13 September 2026 (UTC)

Between March and May 2026 they triggered Special:AbuseFilter/1325 a few times and had chatgpt-derived links , earliest creation is April. Warned, will wait for a response Kowal2701 (talk, contribs) 16:25, 13 September 2026 (UTC)

All articles mentioned have been created after comprehensive research into sources to make them suitable for Wikipedia. I have used templates (Infobox character). Found images that can easily be uploaded for the infobox to demonstrate the character involved in the article. I have followed a simple though effective formatting, which includes casting, appearances, development and reception. They took hours to create and were not AI generated creations. Benigo36 (talk) 21:26, 23 September 2026 (UTC)

User:2screeny2

 2screeny2 (talk · contribs · logs · block log) (18 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

2screeny2 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

The usual. High-volume AI editing since 2025. Warned in June 2026, said they would "be vigilant" (which is questionable on its own because what's there to be "vigilant" about? Are you not in control of your own actions?), does not appear to have actually stopped. Gnomingstuff (talk) 21:41, 13 September 2026 (UTC)

 Unlikely. The spread of edits and pace of edits seem more "interested editor writing about their hyperfixation" than AI user. It's not impossible they're using AI, but I'm not seeing a smoking gun. Special:Diff/1374197013 in particular seems like the kind of extremely human mistake that every editor makes multiple times in their editing career, but which an AI user would either not make or not notice and correct. It's impossible to prove a negative, but they're at least being careful with the content they're adding, enough to go back and fix their mistakes. guninvalid (talk) 09:55, 14 September 2026 (UTC)

I'm afraid I'm more skeptical than you about this case, @Guninvalid. The rate of recent edits doesn't strike me as necessarily inhuman, either, but looking at their most recent contributions, I see the following:

• Special:Diff/1374200412 gets the name of the list wrong (it's just 'Most Powerful Women', not 'Most Powerful Women in Business', and it's right there in the previous item and the citation title/url).
• Special:Diff/1374201539 mentions the retrospective being shown at three specific European museums. This actually is supported in the references—but neither reference mentions it happening in '2025–2026' (with en-dash). They only say it's being shown at Museum Ludwig in August, the Stedelijk Museum in September, and was previously at Fondation Beyeler.
• Special:Diff/1374198065 adds a Prizes section with three items. The first is cited to kids.kiddle.co, which gets its material from Wikipedia (and come on, does it really look like a reliable source for a grown-up encyclopaedia?). The second refers to the 'First Prize in Sculpture, Gabrovo, Bulgaria' when the Spanish-language source gives a more specific name: 'Kiter Peter Sculpture Grand Prize at the VI Biennale of Humor and Satire, Gabrovo, Bulgaria' (machine translation). The third gets both the prize title and the year it was given wrong, even though there's hardly any more information at the cited website.

This is all very suspicious to me. Let's not write it off yet. Suðurhafsljósæta (talk) 10:59, 14 September 2026 (UTC)

Special:Diff/1364074360 includes two references: the first says nothing at all about Orlopp resisting the takeover, and the second mentions a negative statement she made about it but does not support the very LLM-typical reaffirming the bank's strategy of remaining independent amid ongoing negotiations. Additionally, the edit includes a red link to 'Bettina Orlopp', when there's already an interlanguage link to that page two paragraphs above. Suðurhafsljósæta (talk) 09:50, 18 September 2026 (UTC)

I have posted a link to this discussion on @2screeny2's talk page. Suðurhafsljósæta (talk) 10:19, 18 September 2026 (UTC)

Asked them to reply, either here or on their talk page. Suðurhafsljósæta (talk) 11:08, 24 September 2026 (UTC)

Special:Diff/1363927774 adds one dubiously-worded claim (it was Starmer who announced the funding, though Reeves is mentioned in the article) and one absolutely wild made-up one (no mention of financial regulation reform or "£80 billion" in either source!). Suðurhafsljósæta (talk) 10:41, 18 September 2026 (UTC)

Special:Diff/1370649752 links to the 2026 version of a list as a citation that Friedman was on the list in 2023 and 2025. Actually, it seems she's been included every year from 2020 to 2026, so this was faint praise!

Okay, I'll stop for now, I really do have other things to do today. Suðurhafsljósæta (talk) 11:03, 18 September 2026 (UTC)

There are chatgpt-derived links between Feb and June 2025 , only one stubby creation so large edits need analysis Kowal2701 (talk, contribs) 12:52, 15 September 2026 (UTC)

User:SerChevalerie

Cleanup efforts are ongoing and being tracked at /2026-09-14 SerChevalerie, 0 pages need review.

SerChevalerie (talk · contribs · deleted contribs · logs · filter log · block user · block log)

• Xtools (pages created)

SerChevalerie has been proactive about disclosing LLM use in edit summaries . I used these edit summaries to compile two lists of article creations at the case subpage: 1) articles created via LLM (n=89) and 2) articles partially composed of LLM-translated text from Konkani Vishwakosh, a free-use Konkani language encyclopedia which was scanned and uploaded on wikisource (n=98). My thanks to SerChevalerie for their consistent and proactive disclosure. Cleanup may be required however, based on an analysis of two of their LLM-translated articles I will add below. Also pointing out the recent thread User talk:SerChevalerie § Use of chatbots, where SerChevalerie helpfully answered questions about their use. Here , however, they mention they continue to use LLMs to "improve the tone of an article", which is unlikely to be covered by the exceptions at WP:NOLLM. NicheSports (talk) 05:47, 14 September 2026 (UTC)

I took a close look at two of the Konkani-translated articles after having GPT Sol translate the sources (I don't speak or read Konkani at all). I had Sol translate both the scans and the transcriptions to ensure nothing was lost in transcription, which happened in one case. Of course there could still be an issue with these translations.

• Venkatamadhava sourced only to . This sparse article is a faithful translation of the Konkani encyclopedia. However, the subject is an 11th century writer mentioned in 2-3 short paragraphs in the source. The article does not come close to meeting notability guidelines sourced as it is
• Vishnushastri Pandit sourced only to . An interesting subject treated in some depth by the Konkani encyclopedia, I do not have the same notability concerns. However, this article is not just a translation. Some sections, such as the lead and the bullets at the end of Social reforms, are "original" writing and I suspect may contain de-novo AI-generated text. The article contains many NPOV issues and verification failures. Sample: Known for his progressive stances and powerful oratory bolded is non-NPOV and fails verification, he dedicated much of his life to advocacy for women's rights and economic reforms all of this is questionable given source, and is SYNTH at best, the editor of Induprakash, a prominent newspaper the translation I reviewed did not mention the paper's prominence (also, AI word), where he presented highly structured, evidence-based arguments my translation did not mention this, While women's liberation and widow remarriage remained his primary focus never mentioned, Pandit was a skilled writer, translator, and editor NPOV issue, fails verification.

NicheSports (talk) 06:32, 14 September 2026 (UTC)

Hi. I have read WP:NOLLM and understand it completely. I also agree to any cleanup that may be required from the AI translations, as pointed out. I have myself done some cleanup for Vishnushastri Pandit. SerChevalerie (talk) 09:59, 14 September 2026 (UTC)

That article still has several unresolved WP:V failures, unfortunately. I thought about this over the night. There are about 190 heavily AI-assisted articles in mainspace. The two articles I randomly selected showed significant issues. It took me several hours to even investigate this issue, let alone clean it up. Assuming that each article takes ~2 hours to resolve notability, NPOV, and WP:V issues, we are looking at about 400 hours of community time (and this is very optimistic). The best path forward is LLMPRODs, but maybe with a twist. WP:LLMPROD#Requirements allows for use of the procedure if the article creator agrees to cleanup. Would you be willing to agree to LLMPRODs of these articles? The twist I was thinking of earlier is – if you want to "save" a few dozen to your userspace (ideally, ones that clearly meet notability guidelines), you can move them and work on carefully verifying all content, then moving them back to mainspace when ready. We will LLMPROD the rest. However, if there are any articles in the future that you want undeleted so you can "fix" them, you can made an undeletion request to one of the {{@AINBA}} admins, who I believe would be happy to do so. This approach is most considerate of community time, while also getting the problematic content out of mainspace and giving you a chance to preserve the articles you feel most strongly about. Lmk - but I'm hoping this can work out! NicheSports (talk) 20:33, 14 September 2026 (UTC)

I guess that works. Can you please share the list of articles and I will work on this over the next 1 week? Thank you. After that we can decide with one can be LLMPRODed. SerChevalerie (talk) 05:51, 15 September 2026 (UTC)

The lists are at the case subpage: Wikipedia:AI noticeboard/2026-09-14 SerChevalerie. Take your time – let me know when you're ready. If you need more than a week no problem. NicheSports (talk) 06:29, 15 September 2026 (UTC)

Also, you should know that LLMPROD can only be used on articles where the creator is its only significant contributor. So any article that has had a few sentences worth of content added by another user should rather be tagged as {{ai-generated}} and left in place for cleanup. Ok, that's all for now. Thank you for the cooperation! NicheSports (talk) 06:47, 15 September 2026 (UTC)

@SerChevalerie, you need to slow way down while fixing these articles. As I noted above, your clean up to Vishnushastri Pandit was incomplete and now you are "verifying" articles at what looks like a speed of one every few minutes, which means you are missing things. There are no shortcuts here. If you can't/won't take the time to go through every single line of these articles to ensure compliance with core content policies, I will pursue presumptive deletion of all of them under option #1.D of WP:LLMPROD#Requirements. NicheSports (talk) 10:58, 15 September 2026 (UTC)

As mentioned, I have already verified the AI output line by line before this. I'm only tagging those which I am confident about. SerChevalerie (talk) 11:01, 15 September 2026 (UTC)

I have demonstrated in this thread that As mentioned, I have already verified the AI output line by line before this. is incorrect, and re: I'm only tagging those which I am confident about. I immediately found a sentence that failed verification in the first of your "cleaned up" articles that I checked. NicheSports (talk) 11:06, 15 September 2026 (UTC)

Accepted after our recent discussion. Give me some more time, I will do this more slowly. (This will take longer than a week.) In the meantime, I am okay with tagging the pages as "AI-generated". SerChevalerie (talk) 14:19, 15 September 2026 (UTC)

I'm currently going through the articles. For the most part, the translated articles have very minor issues so I don't see why they should be WP:LLMPRODed. However, I am okay with temporary draftification of those articles I am not able to review line-by-line in the immediate future (next couple of weeks). I am currently busy IRL so will need a few weeks for the cleanup, which I am doing line-by-line. SerChevalerie (talk) 04:18, 22 September 2026 (UTC)

User:Curt1us

creations draftified and tagged

 Cleanup has been requested.

This user seems to be expanding & writing new articles using AI with obvious WP:AISUMM, and content tagged as AI-generated by several users and as AI-assisted by Pangram. At least one of their articles (Shell Technology Centre Bangalore) has already been tagged as having LLM-generated text. They've also deleted at least three users' talk page messages asking about AI use without any real attempt to address the issues raised [1] [2] [3], so some further action might be necessary here. A few of their edits (e.g. on Chemelot) might be compliant with WP:LLMTRANSLATE however, depending on their fluency in Dutch which I'm not 100% sure about. Thank you all in advance for your help! Knypster (talk) 11:51, 14 September 2026 (UTC)

Fermiboson reverted one of their large edits a couple days ago, which the editor reinstated: . They can somehow make multiple large rewrites/expansions within just a couple of hours, with and being made within 15 minutes of each other. And I agree AISUMM is on full display. ^purring_maggot (talk) 18:56, 14 September 2026 (UTC)

For that specific edit, I ran an automated check on it and did not find any implausibly human-made mistakes, so I let it go as a borderline case. It might be useful to get an admin in here to talk about LLMTALK first, as if it is AI, then the model/pipeline/amount of human involvement might be good/large enough that we won't get anywhere without the human participating. It is also not implausible that the summary itself is generated with AI but the edit is not, because I have seen a few cases of that happening, for God knows why. Fermiboson (talk) 23:25, 14 September 2026 (UTC)

not triggered any relevant edit filters, though they did add a non-existent template, and remove AI and COI tags . First creation was a 15 August translation from nlwiki which they effectively rewrote in . There second creation was an unattributed translation of which they edited and added a markdown codeblock. Chemelot (13 Sept) has several signs. I'll draftify the creations and give a final warning. Kowal2701 (talk, contribs) 15:27, 17 September 2026 (UTC)

A practical point: the affected articles are now materially less complete, less current, less well sourced, and/or less logically structured than the versions that were removed.

I worked on these subjects because I know them intimately through my professional background, local knowledge, longstanding OpenStreetMap contributions, travel, field familiarity, &c. I would therefore appreciate it if other editors could review the previous revisions and consider restoring them, or reincorporating the material that improved the articles.

The relevant drafts and revision histories remain available, and the articles are generally rather short, so their accuracy should be reasonably straightforward to assess. Curt1us (talk) 21:13, 18 September 2026 (UTC)

AI-generated content is not allowed on Wikipedia, which is the only "practical point" that is necessary. Adding AI-generated content does not make an article more "complete" or "improve" an article, in the same sense that you do not improve a cake by pumping it full of sewage, and in the process swindling by proxy everyone who donated their money because they believe they are getting sewage-free cake. Gnomingstuff (talk) 16:20, 26 September 2026 (UTC)

User:Tipkorir

 Tipkorir (talk · contribs · logs · block log) (27 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

 Cleanup has been requested.

I came across this prolific editor on Kenyan football and found several articles that were G15-eligible due to multiple non-existent citations. (Others have just one non-existent citation and are thus not G15-eligible: Vincent Wasambo and Boniface Akenga, for example.) WP:AISIGNS are also present. I appear to be the first to have warned the user on AI use, but the user's creations are so prolific I think cleanup is in order. Dclemens1971 (talk) 23:48, 14 September 2026 (UTC)

User likely continues to use LLMs; this expansion included two dead link sources with no archived version (, ). Dclemens1971 (talk) 16:05, 16 September 2026 (UTC)

That's incredibly disappointing, I've seen this user around a lot. I wonder how long they have been using LLMs for. Spiderone^{(Talk to Spider)} 16:30, 16 September 2026 (UTC)

the earliest chatgpt-derived link was in April 2025, they've not triggered other relevant filters. On their talk they promised to stop using LLMs, but appeared to continue using AI (?). Creation speeds are pretty normal AFAICT. Austin Oduor (Nov 2025) and several in August 2026 all had red cats, but vast majority of creations don't AFAICT (I found the August ones through doing an EIA with @Bearcat lol , but for the Nov 2025 one they corrected the red cat themselves). Could check through the ~150 articles at time of creation for red cats but don't want to. These are too stubby for signs to appear, spotchecks needed for WP:V unfortunately Kowal2701 (talk, contribs) 19:29, 17 September 2026 (UTC)

Special:Permalink/1336349212 (3 Feb) seems to have markdown italics for the external links, same w Special:Permalink/1333056732 (15 Jan), Special:Permalink/1322441285 (Nov 2025), Special:Permalink/1297968135 (June 2025), and Special:Permalink/1248930258 (Oct 2024), but LLM-generated articles are basically never that stubby w one-sentence-long paragraphs. A lot of their pre-Nov 2022 creations had red cats. Tbf Special:Permalink/1298572222 (July 2025) is plausibly AI-generated and has that thing w the external links, Special:Permalink/1247410761 (24 Sept 2024) is the earliest one I can find w the link thing. But not even sure these are AI. I'll ask them Kowal2701 (talk, contribs) 19:35, 27 September 2026 (UTC)

I just notified Tipkorir of this thread. ‑‑gurkubondinn 19:38, 27 September 2026 (UTC)

did it at the same time lol Kowal2701 (talk, contribs) 19:41, 27 September 2026 (UTC)

spotchecks for some articles with weird link markdown thing:

• Special:Permalink/1376490467 (24 Sept): ref #2 doesn't appear to exist or to have ever existed (neither the url or article), other refs are doubled up and I cbf
• Special:Permalink/1247410761 (Sept 2024): ref #3 doesn't mention a timeframe
• Special:Permalink/1298572222 (July 2025): ref #3 doesn't verify where he made 42 league appearances and scored 2 goals between 2018 and 2023, ref #4 verifies fine, ref #5 doesn't support to reunite with his former Nzoia Sugar coach Salim Babu
• Special:Permalink/1322441285 (Nov 2025): ref #1 404s but only because of the dash at the end (?), otherwise it verifies fine
• Special:Permalink/1297968135 (June 2025): has hyphens instead of dashes, ref #3 doesn't say it was organised by the Tanzanian FA, but verifies the rest

I don't think these are AI-generated tbh, but still baffled nonetheless. Maybe using chatgpt as an uncited source, idk. Would appreciate a second opinion Kowal2701 (talk, contribs) 23:22, 28 September 2026 (UTC)

I had a look through Tipkorir's edit history. Cleanup seems the only choice! Some looks needing source verification too. Not everything in the basket may be bad, but given the pattern going back to at least April when edit got tagged as possible AI generated citations, a presumptive cleanup sounds like the way to go. NicoR8 (talk) 18:48, 18 September 2026 (UTC)

Support community consensus for cleanup. M kuhner (talk) 04:53, 28 September 2026 (UTC)

I'll do some spotchecks later Kowal2701 (talk, contribs) 12:26, 28 September 2026 (UTC)

Support cleanup via LLMPRV by community consensus as well. ‑‑gurkubondinn 15:03, 28 September 2026 (UTC)

Support cleanup by LLMPRV. Suðurhafsljósæta (talk) 11:22, 29 September 2026 (UTC)

User:CockroachHunter

Cleanup efforts are ongoing.

LLMPROD is not needed in this case

High rate of short content edits to a very wide range of articles. For example this edit just 2 minutes after a pervious edit from an editor that doesn't normally edit about football. Source selection is also all over the place. Dodged the question when challenged. Only started editing 31 August 2026 so should be possible to clean up with a simple mass revert.©Geni (talk) 10:24, 15 September 2026 (UTC)

Ridiculous rate of edits on a preposterously wide range of topics given the speed. I chose an edit at random, in the [[cement] article: This edit, which repeats info in the previous paragraph, contains a claim about cement being used in the El Tajin acropolis which the source does not support, in what to me is a very characteristic AI slip. OsFish (talk) 14:35, 15 September 2026 (UTC)

The source directly supports what I added to the text. Here are the quotes from the text of that source:

1. "...El Tajin displays advanced construction techniques as many structures have concrete slab roofs..."

2. "In the 8th century CE, the Pyramid of the Niches was completed and the huge raised acropolis platform of Tajin Chico was constructed. El Tajin was destroyed by fire and abandoned around 1100 CE or even earlier."

What I wrote on the cement article is standard synthesis. Nothing points to LLM or not editing in good faith. Also, there are no "AI slips", and nothing is "characteristic". If there is something "characteristic" or "slipped", please @Geni @OsFish show evidence, not just "gut feeling". Thank you. CockroachHunter (talk) 18:03, 15 September 2026 (UTC)

Your edit here seems to be added by LLM: WP:AICURLY and the extra spacing in the reference. Masterhatch (talk) 18:31, 15 September 2026 (UTC)

Their very first edit is conclusive. It inserts a citation to The Conversation titled "The myth of the soulmate – how the word came into common use" but that article title only appears in the wiki article and the link in their edit actually goes to a totally unrelated article. -- LWG ^{talk (VOPOV)} 18:41, 15 September 2026 (UTC)

Over at Two-party system, his 1st 2 edits were good. His 3rd reeked of AI. I reverted that 3rd edit and a few other recent edits from other articles that have clear LLM signs. Masterhatch (talk) 19:06, 15 September 2026 (UTC)

Thank you for providing the quotes that show your source does not actually match the content it's cited for. OsFish (talk) 05:44, 16 September 2026 (UTC)

I reviewed all of their edits up through the end of Sep 3 and reverted/refactored obvious problems. Not all edits appears to be LLM, though all were bot-like. Some edits were constructive. Many consistent of inserting redundant and unnecessary prose, or adding citations that did not verify the content to which they were attached. -- LWG ^{talk (VOPOV)} 20:19, 15 September 2026 (UTC)

I looked at some of the Sept 13 edits since that was the date of a muddled edit that cropped up on my watchlist. I would not recommend a mass revert because quite a few of these edits are valid typo fixes. LLM use is obvious - over 100 edits that day, all on different obscure topics, often only one or two minutes apart. Paul Matthews (talk) 03:53, 16 September 2026 (UTC)

edits on 11 and 13 September had markdown . I'd create a tracking page if there weren't the risk of more AI edits Kowal2701 (talk, contribs) 22:45, 17 September 2026 (UTC)

Yucalyptus

 Cleanup has been requested.

User is mass dumping 5-10,000+ byte single edit article creations/expansions in rapid succession e.g. , which would be extremely difficult to do without AI. All of them tend to have a characteristic "AI-ish" writing style.Hemiauchenia (talk) 16:05, 15 September 2026 (UTC)

Yucalyptus (talk · contribs) chatgpt-derived links in Dec 2025 , Special:AbuseLog/42851871 (Nov 2025) triggered filter 1325 for a strong AI sign, creations need analysis Kowal2701 (talk, contribs) 16:44, 15 September 2026 (UTC)

The Odyssey (1987 film) (July 2026) had a silly amount of red cats but no other signs really (quite stubby). Cowpasture Road (1 Sept) had AIATTR (The Roads and Traffic Authority described it as a regional freight link between these industrial areas.), all quite promotional for an article on a road. Pangram says (drumroll) AI-assisted . Special:Permalink/1306688629 in August 2025 has a spacing error in one of its cats. Euphorbia succulenta (Nov 2025) had AIDASH, plausibly AI + Pangram positive for AI-assisted (AFAIK "AI-assisted" here doesn't mean exactly that, just that the certainty that it's AI-generated is middling) Kowal2701 (talk, contribs) 23:14, 17 September 2026 (UTC)

Euphorbia succulenta also gives USDA zones that are not mentioned in the given source. M kuhner (talk) 04:15, 19 September 2026 (UTC)

User hadn't received any warnings for AI use so warned and notified. DraconicDark (talk) 18:00, 16 September 2026 (UTC)

Thanks for the warning. As I said, I have used AI for correction of my grammar. However, the overall content I've written is in my words, backed up by sources I have found and saved. Before creating an article, I gather sources and save them in my documents (i.e. my personal 'sandbox'), hence the 'rapid succession'. If grammar correction by an AI source is prohibited, I will stop of course. Yucalyptus (talk) 00:08, 17 September 2026 (UTC)

Grammar correction is formally permitted. However, if you push your text into an LLM and ask for grammar correction, it will very often do more than that. A non-LLM grammar checker, which can't hallucinate new text, is much safer.

Euphorbia succulenta has USDA zone information that I can't find in the source you cite: how did that happen? M kuhner (talk) 04:16, 19 September 2026 (UTC)

Thanks for the advice. I'll look for non-LLM grammar checkers online (some are better than the others). Regarding Euphorbia succulenta, I have no idea honestly -- I may have missed adding its source on its hardiness zone. I rarely forget sources and I always try to add them to backup the material I included. This source clearly states the USDA zone information. I may add that one if it is deemed reliable. Yucalyptus (talk) 01:17, 21 September 2026 (UTC)

I don't know this site and can't find any information on its policies in a quick search. There's a risk in such cases that the information is circular, pointing back at Wikipedia, or is user-supplied without fact-checking. You might investigate a bit more. M kuhner (talk) 01:24, 21 September 2026 (UTC)

Unfortunately a lot of sites like this are now AI generated, generating info on lots of plants without checking the accuracy of the information to maximise advertising revenue. You have to be very careful to verify the quality of the sites. Regardless as to whether it's AI generated or not, worldofsucculents.com looks like a content farm optimized for SEO rather than quality. I would avoid it as a source. Hemiauchenia (talk) 22:38, 22 September 2026 (UTC)

Fair point. The website/source does look "off". As such, I'll remove the material regarding the plant's hardiness zone since it is, after all, unsourced (or no reliable source backs it up). Yucalyptus (talk) 13:44, 24 September 2026 (UTC)

User:Wikieual

 Cleanup has been requested.

creations draftified and tagged

User engaged in suspiciously rapid article creation yesterday some of the article show some obvious AI signs:

• WP:AIREDCAT: Hippodrome d'Enghien-Soisy
• WP:AIDASH: Pari Mutuel Urbain, Fédération nationale des courses hippiques

And a lot of these creations also haver high ai scores on GPTZero.

Pari Mutuel Urbain also has a page on fr-wiki, but this page does not appear to be a straight-forward machine translation.

I left them a message on their talk page about a day ago, but they haven't been online since. I'll draftify the obvious ones from yesterday, but their other work might also need a look

Giulio 22:32, 15 September 2026 (UTC)

Took a 5-year break in article creation from 2021 to August 2026. EUCS (cloud computing) (25 August) had AIREDCATs, they've triggered no relevant edit filters. Kowal2701 (talk, contribs) 23:24, 17 September 2026 (UTC)

that creation also gets a Pangram positive , draftifying the rest Kowal2701 (talk, contribs) 20:02, 27 September 2026 (UTC)

Thanks for raising these points, and apologies for the slow reply.

On the pace: each of these articles was prepared offline over several weeks and then pasted in section by section, which is why the saves cluster into a few minutes. The reading, the sourcing and the drafting did not happen in those minutes. Should I do it differently, please let me know.

On tools: I used an LLM to ensure proper English spelling and grammar. Some spaced em dashes (WP:AIDASH) were introduced in the same process, therefore I will be cautious in the future to avoid such replacements.

On the red category (WP:AIREDCAT), it is the name of the commune where the racecourse stands: Soisy-sous-Montmorency (as present in the article). I was thinking of creating such a category in the near future to reference all content related to this city. It was not an invented title, but I have removed it to take your comment into account.

On the specific defects, I took care to make the modifications in the related drafts. Please let me know if they are OK.

On sourcing, I would welcome spot-checks. The material comes from various sources, including Légifrance, the Cour des comptes, Base Mérimée and the racing bodies' own publications. If any statement fails verification, please let me know.

I have several further articles in preparation. I will hold them until this is resolved.

Thanks for your feedback. Wikieual (talk) 20:44, 21 September 2026 (UTC)

Did you generate this talk page comment with an LLM? Fermiboson (talk) 14:31, 23 September 2026 (UTC)

User:ManoiCMU

 Cleanup has been requested.

User engaged in suspiciously rapid article creation and subsequent editing;

For example Kad Suan Kaew was written yesterday at 15:09 and then completely re-written at 15:12, the article was further edited until 15:26 then Wat Si Rong Muang was published at 15:40

Other AI signs are also present: second reference on Kad Suan Kaew appears to be a broken link, WP:AIREDCAT in Wat Ban Kao and Suphrom.

Giulio 20:13, 16 September 2026 (UTC)

First of all, I will defend myself. I am one of the main editors driving the Thai English Wikipedia project. As a very experienced editor, I have full knowledge of Hallucination (artificial intelligence) and Wikipedia:Signs of AI writing.In the past, I have had no major problems with any Wikipedia policies, such as notability, writing errors, or translating from Thai Wikipedia articles. I have also never been banned for any reason. I have been a fighter in AfD discussions on Thai articles, and I have saved all the articles that I was involved in those AfDs. All the articles I have created fall under the notability guidelines, and none of them have been deleted. I have had no problems at all before.

During this time, I have created several articles because I have had more free time due to my school break. If I use a large language model (LLM), my articles would be automatically expanded beyond what is supported by the references. My articles do not show signs of containing both WP:AIATTR and WP:SUPERFICIAL. Moreover, my sentences in the articles are directly supported by the references given, and you can easily verify them one by one. If you cannot read Thai, you can use Google Translate or ask one of the Thai native editors to verify the information in each article. If you find any content that cannot be verified, we can open a case on the article's talk page.

You accused me of using AI here because I placed broken references that never existed or were hallucinated by AI, correct? Actually, reference 2 from Kad Suan Kaew is directly copied from the sources cited in th:อุทยานการค้ากาดสวนแก้ว. You can check the Thai Wikipedia article as well and determine whether it was added by AI or by editors. I used this reference in English Wikipedia for the retail floor area claim.

Thank you for pointing out the reference. 5. I have checked the references and found that reference 5 was misplaced and added as backup. Reference 5 is also used for general history and facilities. I added reference 5 to another sentence as a "backup" even though that sentence already had another reference. You can verify the sources one by one. So, I may have forgotten to add a reference to this sentence, and I did not want to leave it blank beside the sentence. However, this does not mean that the information is fake or AI-generated. That information came directly from the Thai Wikipedia article. I use some content from Thai Wikipedia articles by translating it and mixing it with my own writing. I rewrote these articles because I translated content from Thai Wikipedia some of them. Sometimes, I did not want to use or add the page's edit history for attribution under WP:CWW, and sometimes the rough translation sounded like storytelling, so I rewrote it.

You can also check the Thai Wikipedia article for this claim. To be sure, you can check this article (this is a source for the claim in addition to the Thai Wikipedia article) and for the theatre-related facts.

I also have a source that directly supports the South Pacific opening-performance claim: . It says:

โดยเฉพาะอย่างยิ่งโรงละคร “กาดเธียเตอร์” ที่เปิดตัวด้วยการนำคณะละครบอร์ดเวย์จากต่างประเทศเรื่อง South Pacific มาแสดง

A close translation is: “Especially the Kad Theatre, which opened by bringing a Broadway theatre company from abroad to perform South Pacific.” So the source directly supports the claim that Kad Theatre opened with South Pacific and that the performance was presented by a foreign Broadway theatre company. Therefore, this is not AI-hallucinated information.

If my content fails verification, you should add {{cn}} or mark it as "failed verification" and question me on the article's talk page first so that we can discuss it. Bringing me to the AI noticeboard is too strong and has reduced my motivation to create more articles. If I use fake information, then of course I should be warned.

Regarding WP:AIREDCAT, this is very normal for me. I used test categories like this before AI was born (at the time nobody knew what AI was), and you can check my older edit history. I am also very lazy about clicking the preview button before saving.

Please kindly review the Thai article before concluding that some of the content in my articles is unsupported, since some of my sources and content come from Thai Wikipedia. Thank you. ManoiCMU (talk) 06:56, 17 September 2026 (UTC)

Ping @DraconicDarkfor understand expalin. thank you ManoiCMU (talk) 06:57, 17 September 2026 (UTC)

Can you please explain your process for Kad Suan Kaew? There are two edits minutes apart: at 15:09 you create the article, then at 15:12 you completely re-write it . Giulio 15:47, 17 September 2026 (UTC)

The first edit was simply the initial creation of the article, using a rough translation from the sources and the Thai Wikipedia article. I had gathered all the useful information, including minor facts copied from sources, and pasted it into a Note on my MacBook. I also had many different draft sentences prepared before creating the article on Wikipedia. Copy paste from my notepad is my method. However... I tested Gemini AI to smooth my rough translation version sometimes, but AI often made it worse by adding a superficial tone like attributed or showcasing or renowned and unnecessary overanalysis, making my version worse and more horrible. After seeing that, I chose not to use AI and instead write my own research using mixed content from different sources. AI is useless even for paraphrasing.

If I felt that a sentence sounded strange or unnatural, I would replace it with another sentence from my draft notes or rewrite it. But...my browser also has Grammarly enabled for minor grammar corrections. To be honest, if you ask an AI chatbot such as Gemini or Grok etc.., “Hey, please create an article about Kad Suan Kaew for Wikipedia with refs,” and try it yourself, you will probably get a disgusting version containing fake information, unsupported expansion, unreliable claims, and possibly fake references or dead links. AI-generated versions can also have very recognizable sentence structures and wording. You should compare such a version with mine; they would definitely not be the same. Here are examples: https://gemini.google.com/share/c207c1190acb?skid=50f77144-1c03-4ea6-8da9-00e6b44399ac and the Grok version https://grok.com/share/c2hhcmQtMw_7122d800-6b2d-4075-ad6c-7aad5af437c5 and the GPT version https://chatgpt.com/share/6aacb34f-9640-83ec-ac39-9b4b11840167.

I have also reviewed many Thai Wikipedia articles created by other editors in drafts that were marked for LLM-style writing. I corrected them and successfully helped publish them in the main space. I am familiar with identifying paraphrasing produced by LLMs. For my article, there are no fake references, and you can verify the content sentence by sentence. If you fail to verify any content, you can ask me or raise the issue anywhere appropriate. And the Kad Suan Kaew article is now tagged for speedy deletion under CSD G15, with the reason, “References 2 and 5 are hallucinated.” Even after I explained that the references came from Thai Wikipedia and pointed out that the facts are not fake and can be verified through the pointed sources, the speedy-deletion nomination is still open. May I know why, and can I remove the tag now that I have fixed the issues? Thank you. ManoiCMU (talk) 03:55, 18 September 2026 (UTC)

Do you have a premium version of Grammarly? Fermiboson (talk) 05:48, 18 September 2026 (UTC)

Also, I am slightly confused by I have also reviewed many Thai Wikipedia articles created by other editors in drafts that were marked for LLM-style writing. I corrected them and successfully helped publish them in the main space. Do you mean English Wikipedia articles about Thai subjects or articles on Thai Wikipedia? Fermiboson (talk) 05:52, 18 September 2026 (UTC)

Sorry for the late response. I was busy with my private job. I mean English Wikipedia articles about Thai subjects. You can see some of my cleanup work in Wat Rachathiwat and Draft:Suki Teenoi. I often clean up AI-written content in Thai-topic articles when I find them in the draft space. If the article creator does not listen to the concerns or reverts my cleanup work, I may propose the article for deletion. ManoiCMU (talk) 14:37, 21 September 2026 (UTC)

The first edit was simply the initial creation of the article, using a rough translation from the sources and the Thai Wikipedia article.

How much of your work is translation of Thai articls? Translation would explain the rapid rate of content creation, but at no point that I could see have you declared any of your work a translation as required for proper attribution per WP:TFOLWP. Giulio 14:22, 18 September 2026 (UTC)

I understand that translating Thai Wikipedia content requires attribution under WP:CWW. I provide attribution when a large amount of my translated content comes from another Wikipedia article. You can see attribution tags on the relevant article talk pages.

However, when the content is mixed with material from other Thai sources and my own writing, or when only a normal amount of content is translated, I do not add attribution. Here is an example of my attribution under WP:TFOLWP:

Based on my experience with AI, most AI-generated content may appear to have sources, but those sources often cannot be verified because they are fake or do not support the claims. I have already stated on most of my articles that if a source does not support something I wrote, anyone can ask me for the source on the article's talk page at any time.

Thank you. ManoiCMU (talk) 14:53, 21 September 2026 (UTC)

chat-gpt-derived links in Nov 2025 , Yi Peng festival (Nov 2025) has lots of signs and a Pangram positive , it is clearly not a translation from thwiki's article. Looking at 2026 creations, Wat Ban Kao (15 Sept) has AIREDCAT, a bit of AICONNECT (The site is associated with Khruba Chaiyawongsa Phatthana...), Pangram comes back as human-written . There aren't many signs in the 2026 creations, spotchecks are needed imo Kowal2701 (talk, contribs) 16:17, 18 September 2026 (UTC)

Sure... I already admitted above that I tested AI for a while, but after I learned that using AI is easily detectable on Wikipedia, I did not use it anymore, even for paraphrasing. You said that there were lots of signs and a positive Pangram result at . Actually, I created this section as a stub; however, another editor expanded it to the current version. You can see .

For AIREDCAT, I already explained this above, but let me explain again. I added Category:Buddhist temples in Lamphun to Wat Ban Kao because the temple is located in Lamphun. Is this really such a big mistake? I did not add some unrelated category such as Category:Buddhist temples in Mars Planet. I added the category because the temple is actually in Lamphun. What is wrong with that? The fact that the category name does not include the word "Province" Category:Buddhist temples in Lamphun province do not automatically make it AIREDCAT. There is also Category:Buddhist temples in Chiang Mai (https://en.wikipedia.org/wiki/Category:Buddhist_temples_in_Chiang_Mai), so I understood Category:Buddhist temples in Lamphun in the same way. Chiang Mai and Lamphun are both provincial capitals. I don't understand why Lamphun would not have a similar category for its Buddhist temples like Category:Buddhist temples in Lamphun. You accused me of AICONNECT with this sentence: The site is associated with Khruba Chaiyawongsa Phatthana. The sentence “The site is associated with Khruba Chaiyawongsa Phatthana” was not generated from an AI inference or invented by me. Thai-language sources identify Wat Ban Kao (วัดบ้านก้อ) as the main place of Khruba Chaiyawongsa Phatthana. For example, The Nation source describes and the old Wat Ban Kao as a former spiritual center and explicitly identifies it as the origin root of Khruba Chaiyawongsa Phatthana. Well, please give me a source showing that the monk is not associated with this place.

To be honest, because of the monk associated with this place, I also created the Wikipedia article for Khruba Chaiyawongsa Phatthana after I created Wat Ban Kao, because I hate to see redlinks on articles that I created. I really feel you guys were trying to put me to a death sentence in court rather than focusing on the countless AI article creators on present-day English Wikipedia. Even if you guys take over 100 years to clean up AI editors, you can't possibly have enough time to take over 100 years and keep up your nice work, and i support it and trying to catch big fish. I already said above that I rewrite an article if my article seems to have a strange tone. Why would I use AI when I can write an article with 100% human tone? I have never used an AI paraphrasing tool like QuillBot (famous for human writing), even though these tools are easily detectable as AI writing. Btw, the Yi Peng festival article also passed the DYK process. At the time, the article was tested with Pangram and Copyvio, and every sentence in the article needed to be verifiable by the sources used. If a source did not support a single sentence, it would not pass DYK. I also faced questions about factual claims in the Preparation area , and I successfully passed all of them without any problems. If there are major signs of AI hallucinations in my articles, I am super ready and happy to accept appropriate action against me, and I welcome a warning if this is the first time. ManoiCMU (talk) 15:30, 21 September 2026 (UTC)

User:Peoplelikeyoulikeme

 Cleanup has been requested and is being tracked at /2026-09-20_Peoplelikeyoulikeme, 193 pages need review.

Mark Holgate is clearly AI-generated per the tag by Gnomingstuff, and I suspect the editor's other numerous and ongoing article creations may be AI-generated as well. There are references in the editor's articles in which the information given is not fully supported by the sources used: so, for instance, in Chris Oh the "Artnet News" article by Katie White does not name the painting by da Vinci mentioned in the text, and the other "Artnet News" article by Eileen Kinsella barely mentions any of the content in the last paragraph of the article. Other editors more experienced in AI detection may be able to find more definite evidence.

I would suggest that LLMPROD might be appropriate, if other editors agree that there is a likely AI issue here. Dionysodorus (talk) 23:42, 16 September 2026 (UTC)

tons of chatgpt-derived links in Oct and Nov 2025 , they also triggered filter 1325 for a very strong AI sign w Special:AbuseLog/40297290 in March 2025 ("underscores the importance of"). 63 creations since Jan 2023, Divided Publishing (Feb 2023) had AIREDCAT but also some weird errors that look human, also uses hyphens instead of endashes/emdashes. Zora (platform) (Jan 2024) was deleted in part for PROMO. What is going on w the square brackets in Special:Permalink/1157904631. Bob Glück (June 2023) had the unsourced Bob Glück's body of work spans several decades and encompasses a wide range of poetic styles and themes. He is known for his innovative approach to language, delving into contemporary life and experiences of queer identity, unconventional relationships, narrative discourse, and moments where language, power, and desire fornicate.. OCDChinatown (Nov 2024) had AIDASH, AICURLY, AIVOCAB, ridiculously promotional, probably SUPERFICIAL if I could stomach reading it closely, Pangram gives human-written though. Ughh Kowal2701 (talk, contribs) 16:37, 18 September 2026 (UTC)

Spot-checking the refs on the user's latest article, Travis Boyer, immediately throws up sources that do not contain the information that they are cited for: so, for instance, source 8 does not mention the deCordova Museum, Rhode Island School of Art and Design, etc., and source 17 does not mention the subject's paintings on silk velvet (and does not even discuss the subject: his only presence in the article is that he made the image illustrating it).

On the basis of all this, is there enough for consensus for WP:LLMPRV? Dionysodorus (talk) 00:40, 20 September 2026 (UTC)

They're inactive and I think there's enough evidence of LLM use for it to habitual, so meets WP:LLMPRV#Requirements and articles can be LLMPRODed, but we need to establish a rough start date which looks like it'll need WP:V checks of older creations. I should be able to do this some time next week (unless you fancy it, no worries if not) Kowal2701 (talk, contribs) 20:19, 20 September 2026 (UTC)

Well, if we look at the user's first page creation in Jan. 2023, in its original version, we can see WP:AITITLECASE in the first section, and then an unreferenced sentence at the end of the first section and at the end of the second paragraph of the second section (is it an AI thing to have occasional unsourced sentences in an otherwise sourced text like that? I think I have seen it in other cases). A look at source 13 fails verification, since it does not mention the fact that his mother Patricia's songs are haunting and soulful, nor the length or format or platform of the album mentioned.

So my suspicion here is that we need to go all the way back to the creation of the account in Dec. 2022. Dionysodorus (talk) 20:37, 20 September 2026 (UTC)

Sourcing that ends before the end of the paragraph is, in my experience, a tell, though not one mentioned in WP:AISIGNS. Especially so when the material could easily be attributed to sources already in the article.

Thanks for the date info. I'll make a tracker back to the start of the era. M kuhner (talk) 23:22, 20 September 2026 (UTC)

Pangram gives human-written, but it could be a false negative. Imo using hyphens instead of endashes/emdashes is a strong human sign, and shows up in Feb 2023. One WP:V failure is pretty normal unfortunately, though that one is somewhat typical of LLMs. Wouldn't be comfortable LLMPRODing based on this, but can do some more spotchecks. For Bob Glück (June 2023) the refs check out but a lot is unsourced. "Untitled" (Golden) (Jan 2025) reads like AI but a Pangram negative . Tbh I think the start date's closer to March 2025 when there was Special:AbuseLog/40297290, but irdk Kowal2701 (talk, contribs) 19:22, 21 September 2026 (UTC)

Possible AI usage by Whiskerpedia

 Courtesy link: User talk:Whiskerpedia

Request: Please hold this discussion on Whiskerpedia's Talk page, User talk:Whiskerpedia, until AINB is removed from semi-protected status. Relaying messages back and forth is IMO interfering with the discussion.

Whiskerpedia (talk · contribs · logs) is a brand new editor (30 edits) who started on 15 September, and has been rapidly adding content to about a dozen Quebec town articles, and appears to be editing in good faith. They have created >100kb of polished content on 9/16 in about eleven articles, making edits of up to 20kb creating polished-looking artice or Draft content without requiring a follow-up edit. I removed one 12kb edit to Brome, Quebec but it was reverted. They blanked the first message I left on their UTP but have engaged constructively with the second message I left. Collectively, Whiskerpedia's edits sure look like AI to me, as nobody can create content that fast, but they have a counterargument. I told them I'd bring this here so we could discuss further. Plenty of diffs in the Talk page discussion. Thanks, Mathglot (talk) 05:52, 17 September 2026 (UTC)

The counterargument looks like another AI edit. CMD (talk) 06:03, 17 September 2026 (UTC)

Housekeeping note: Whiskerpedia reports that because this page is semi-protected (news to me!), they cannot respond here. As they are here in good-faith, it would be nice if lurkers or participants here could check Whiskerpedia's Talk page periodically, to see if they have a comment they would like to have ported over. Thanks, Mathglot (talk) 07:07, 17 September 2026 (UTC)

Material copied from User talk:Whiskerpedia as they are unable to reply here:

Thanks, I appreciate you taking it to the noticeboard instead of just reverting everything.

Unfortunately it’s semi-protected right now, so I can’t actually respond there myself. I did see the comment saying my last response looked AI-generated. It wasn’t. I wrote that response myself.

I also want to explain the volume a little more clearly. I’ve been home sick all week, with a lot of time on my hands, and this is quite literally what I’ve been doing to keep myself occupied. I got interested in these local articles and just kept going from one to the next. So I understand why the amount of editing looks unusual, but there really isn’t much more behind it than that.

It’s a bit discouraging that now that I’m trying to engage and explain myself, even that is being treated as further evidence of AI use. For now I’ll just follow the discussion and see what comes of it. Whiskerpedia (talk) 06:29, 17 September 2026 (UTC)

(copied by M kuhner (talk) 11:58, 17 September 2026 (UTC)) ➤ ported W's 06:29 signature by Mathglot (talk) 20:36, 17 September 2026 (UTC)

ChatGPT UTM param in Special:Diff/1375077880, this is not really ambiguous Gnomingstuff (talk) 15:07, 17 September 2026 (UTC)

"Yes, that part is true, and it’s what I was trying to explain earlier. I did use AI during the process, especially for research and finding sources. I wasn’t trying to claim otherwise.

I also did remove the ChatGPT tracking part from most of the links because I knew that seeing "ChatGPT" in the URL would probably turn the whole thing into a debate about AI rather than the source itself. Obviously I missed one. In hindsight, I can see how removing it probably looks worse now that someone has found one, but I’m not going to pretend I didn’t do it.

What i’ve been trying to say from the beginning is that using AI as part of my research process isn’t the same thing as asking it to write an article and then pasting that article into Wikipedia. That’s the distinction I’ve been trying, maybe not very successfully, to explain." Whiskerpedia (talk) 15:25, 17 September 2026 (UTC) ➤ ported from their UTP by Mathglot (talk) 20:36, 17 September 2026 (UTC)

Mont Écho has bucketloads of AIATTR at a start. The utm diff has AISUMM. Wait a bit for verification analysis, but FWIW I don't believe them at the moment. Fermiboson (talk) 21:39, 17 September 2026 (UTC)

Contemporary newspaper coverage referred to Mont Écho when the ski centre was being developed in 1963.[6] Later accounts of the ski area state that the property on which it was developed had been known as Echo Valley, although the Commission de toponymie does not establish this as the origin of the mountain's name. I mean... really?

Shockingly, however, I cannot seem to find any source-text discreptancies after combing through it with the machine and with spot-checks. There are a few really minor points, like the reserve covering "more than 78 sq km" -> source "nearly 78 sq km" in 2009, and the 2025 forest fire sources not mentioning fire remaining in the soil, but this is far below the usual discreptancies we see in LLM-generated text. I suspect some combination of source-finding and copyediting. But then that wouldn't explain the AISUMM... Fermiboson (talk) 01:01, 18 September 2026 (UTC)

"Thanks for taking the time to actually check it that closely. What you found is pretty much what I’ve been trying to explain. I do use ChatGPT a lot on the research side, especially for finding sources, pulling out the parts that look relevant and keeping track of things I still need to look into.

On the Mont Écho/Echo Valley sentence, the reason it’s so cautious is that the sources don’t really establish the connection. One source already has Mont Écho being used in 1963, another says the ski property had been known as Echo Valley, but the Commission de toponymie says the origin of the mountain’s name is unknown, so I didn’t think it would be right to turn those things into a definite connection when the sources themselves don’t.

The 78 km² point is fair... it should be "nearly 78 km²."

I did check the fire point again too. The source attached to that sentence doesn’t support the soil part, so you’re right about that. There is an earlier Town of Sutton update that says the fire was very deep in the soil, so the underlying point wasn’t something I just made up, but I clearly attached the wrong update to it. That’s the kind of thing I can see happening when I’m moving through these articles as quickly as I have been, and I’ll fix it.

As for AISUMM, if you mean the NAS Oceana edit summary itself, I was using ChatGPT around that edit for research and source-finding too, as I’ve already said. If there’s something specific in the summary that you think is particularly telling, point it out and I’ll answer it. I’d rather be specific about what I did than just guess at what you mean." Whiskerpedia (talk) 01:39, 18 September 2026 (UTC) copied from user talk by M kuhner (talk) 01:55, 18 September 2026 (UTC)

This is so confusing. That comment reads like Claude, but the source parameter is ChatGPT. And quite frankly, if it is a human, the human has clearly not read WP:AISUMM and also did not address the stylistic/copyediting concerns. Either we're being played professionally here, or Whiskerpedia should get a job in LLM ghostwriting.

That said, I don't see anything blatant yet in terms of other core policy violations, so my vote for this one is rope. And in case this is one of those statistical rarities and we're talking to an actual human - the original quoted sentence isn't really encyclopaedic because it infers things from the lack of information in a given source, which is WP:Synthesis. Instead of attributing to vague "later accounts", etc. you should also either state explicitly what the statement is attributed to, if it requires in-text attribution, or just state it as a fact in wikivoice if not. Fermiboson (talk) 05:04, 18 September 2026 (UTC)

"Well I went back and read WP:AISUMM properly. I understand what you meant now. I had taken your earlier comment as referring to something specific in that one edit summary, rather than the broader pattern the page is describing. I can see why some of my summaries stood out in that context. I've already explained how I've been using AI on the research and source-finding side, so I don't think there's much more I can usefully add to the comparisons with Claude or ChatGPT beyond that.

On the Mont Écho sentence, I do see the issue with the way I put those facts together. I wasn't trying to establish that Echo Valley was the origin of the name Mont Écho. I was actually trying to avoid making that connection because the sources don't establish it. But putting the 1963 use of Mont Écho, the later Echo Valley references and the Commission de toponymie's unknown origin together in the same passage can still make the connection look more meaningful than the sources support. That can be separated more cleanly. I also agree that 'later accounts' is vague and should either identify the source or be stated directly where attribution isn't needed.

I also appreciated that you actually went through Mont Écho against the sources in detail. You said you expected to find the kind of source-text discrepancies you normally see in LLM-generated material, but after checking it closely you didn't find that pattern apart from the couple of relatively minor issues you identified. Now I don't think that settles every concern people have raised, but it was useful because it gave us something concrete to look at rather than relying entirely on how the writing appears stylistically.

As I've mentioned before, this account is new, but I'm not new to Wikipedia. I've edited under an older account before and no longer have access to it. I've also explained why the volume was unusually high this week. I've been home sick with a lot of time on my hands, and most of these articles are about closely related places where the research overlaps quite a bit.

I'm not planning to stop making substantial edits or creating articles, but I may space out some of the very large expansions for the moment so they're easier for other editors to review. If there are specific content or sourcing problems like the wording on Mont Écho, those are useful to identify and I can deal with them directly." Whiskerpedia (talk) 06:17, 18 September 2026 (UTC) copied from user talk by M kuhner (talk) 06:19, 18 September 2026 (UTC)

And in case this is one of those statistical rarities and we're talking to an actual human

and we're talking to an actual human who is using ChatGPT for at least one thing, and whose text produced is also like ChatGPT, to the letter

do people think we're stupid Gnomingstuff (talk) 18:28, 18 September 2026 (UTC)

I think you're jumping from “ChatGPT was used somewhere in the process” to “therefore these edits were generated by ChatGPT,” and I don't think the evidence gets you there.

I've never denied using ChatGPT during the process. The UTM parameter shows that. But even WP:AISIGNS says that a ChatGPT UTM shows involvement, not that the writing itself was generated by ChatGPT. It specifically mentions things like finding citations. Wikipedia's own guidance also allows LLMs to be used for research and finding sources. So the fact that I used ChatGPT in one part of the process doesn't automatically tell you how every sentence in an article was written.

The speed of my edits, the summaries and the style can obviously make you suspicious. Fine. But suspicion is still not the same thing as establishing that the article text was AI-generated. WP:LLM itself says some human writing can resemble LLM output and that you need more than basic linguistic or stylistic signs.

And when Fermiboson actually checked Mont Écho against the sources, they said they weren't finding the source-text discrepancy pattern they normally see in LLM-generated material. They found a couple of specific problems, I acknowledged them, and those can be fixed. That's an actual check of the article And its sources. Saying the text sounds like Claude or ChatGPT is a much more subjective judgment in my opinion.

The “do people think we're stupid” also comment bothered me for the same reason. It doesn't add anything to the evidence. It just makes anyone who doesn't immediately agree with your conclusion sound gullible. I'd rather keep this about the edits themselves.

And, what I really don't understand is why you then started reverting the material while the AIN discussion was still unresolved. Fermiboson's position was “rope.” I wasn't blocked, I hadn't agreed to a presumptive cleanup, and I don't see where there was a consensus at AIN to apply WP:LLMPRV to all of these edits.

Now, If you found a specific sourcing, synthesis, neutrality or other content problem in one of them, point to it then. That's something I think I can actually answer. But, if these were being removed presumptively as AI-generated content, then I think it's reasonable to ask what part of WP:LLMPRV you thought had been met before you started reverting them. Whiskerpedia (talk) 19:42, 18 September 2026 (UTC) copied from Talk page by M kuhner (talk) 21:46, 18 September 2026 (UTC))

Request: Please hold this discussion on Whiskerpedia's Talk page, User talk:Whiskerpedia, until AINB is removed from semi-protected status. Relaying messages back and forth is IMO interfering with the discussion. — Preceding unsigned comment added by M kuhner (talk • contribs) 21:59, 18 September 2026 (UTC)

Conversation is ongoing at User talk:Whiskerpedia and continues to need feedback. Thanks, Mathglot (talk) 18:21, 22 September 2026 (UTC)

I am inclined to think that the preponderance of evidence is against NOLLM-violating AI use, based on the fact that the last TP comment reads a lot more human, I could not find anything major and the things that did show up in the check look more like human mistakes than AI ones, and that their description of the editing process matches all technical indications. Fermiboson (talk) 06:57, 26 September 2026 (UTC)

Do we have a resolution here yet a whole bunch of people have been reverted and we're trying to clean this up......just today we have a draft article that was published full of AI content.

Mont Écho https://www.pangram.com gets 100% AI for the text below in the newly published draft

}}

Mont Écho was developed as an alpine ski area in 1963. Five businessmen from Waterloo formed a company to develop land on the mountain, and the centre opened for the 1963–1964 ski season.

The new ski area had two double chairlifts and a Poma surface lift, with about fifteen marked trails. The longer chairlift climbed approximately 1,500 ft (457 m) over a line more than 7,000 ft (2,134 m) long. A La Presse article published shortly before the opening described it as the longest double chairlift in Canada.

A three-storey base lodge was subsequently constructed at the foot of the slopes. The ski centre was marketed largely through nearby Knowlton, since many visitors approached the mountain by way of Knowlton and Chemin du Mont-Écho, although the ski area itself was within Sutton.

The centre had some successful seasons during the 1960s, but its finances were unstable. By 1968 the property had been put up for sale, and financial institutions became involved in its operation. The ski area nevertheless remained open through most of the 1970s.

Its final season was the winter of 1977–1978.[3] After the closure, neighbouring Mont Sutton acquired much of Mont Écho's chairlift infrastructure. Twenty-five of the former area's 29 lift towers were moved by helicopter in the autumn of 1978 and reused at Sutton.

The former ski runs were gradually reclaimed by forest after the centre closed. The base lodge remained in use for a number of years by a hunting and fishing club before being destroyed by fire in the mid-2000s.

Organized hiking on Mont Écho developed as the ski operation was ending. Sentiers de l'Estrie, an organization created to establish a long-distance trail network through the Eastern Townships, completed its first trail section in 1977. In 1979, with assistance from a federal employment program, a new section was opened between Bolton-Centre and Mont Écho.

The trail system continued to expand southward through the Sutton Mountains during the 1980s and 1990s. Mont Écho became one of the areas crossed by the Sentiers de l'Estrie network, with access from the Chemin de la Falaise sector.

Conservation of the surrounding forest became a local issue in the 1990s. In 1994, residents mobilized over two forest properties owned by Domtar that were being offered for sale in the Nine Holes and Fullerton Pond sectors. The Mont Echo Conservation Association was established during this period, and the Nature Conservancy of Canada began considering the acquisition of lands in the Sutton Mountains

Moxy🍁 05:25, 28 September 2026 (UTC)

@Sulfurboy: Did you review that draft for AI content before moving it to mainspace? ~Anachronist (who / me) (talk) 05:38, 28 September 2026 (UTC)

Yes. Looks AI polished, but not fully machine generated and not seeing any indication of hallucinated info in the article. This is an issue for correction in main space, not the AfC process, considering notability is clearly met. Sulfurboy (talk) 07:37, 28 September 2026 (UTC)

User:Troy26Castillo

 Cleanup has been requested and is being tracked at /2026-09-28_Troy26Castillo, 65 pages need review.

Troy26Castillo (talk · contribs · deleted contribs · logs · filter log · block user · block log)

I rejected a draft with obvious AIATTR. They Immediately edited the draft with an AI edit summary Special:Permalink/1375350945 and resubmitted it, then replied with obvious AITALK Special:Diff/1375351879. They were warned of AI-use earlier this month, which they emphatically denied. I have just tagged that relevant article, List of awards and nominations received by BGYO, for G15. Kisseran (talk) 08:22, 17 September 2026 (UTC)

I was able to find the sources for all three claims with a quick search. Is that suffice to then remove the G15, or are there still underlying problems?

PeepeeDino (talk) 08:37, 17 September 2026 (UTC)

Unless you followed the links given on the article, then no, because the links being broken has no reasonable explanation other than being LLM-generated, and those links have been on the page since the first revision. Kisseran (talk) 08:40, 17 September 2026 (UTC)

Well, after "fixing" (properly sourcing it I assume?) the links, would the article then be fine? Or is the entirety of the article just a mess then?

PeepeeDino (talk) 09:04, 17 September 2026 (UTC)

From experience, the best way to deal with an LLM-generated article is to delete the LLM text, read the sources (and add new ones as needed) and write a human article.

Trying to fix the bad sourcing is more work than writing a new, clean article--speaking as someone who's done it a couple times and won't be doing it again. M kuhner (talk) 12:03, 17 September 2026 (UTC)

Update: The G15 tag has been removed presumably by a LOUTSOCK Special:Diff/1375360217 after they rewrote the article again with obvious AI summaries: Reviewed and corrected the cited references, replaced or removed problematic sourcing, verified award claims against the cited material, and copyedited the article for accuracy and consistency and Restored concise group and career context to the lead, including the members, former name, debut releases and Billboard Next Big Sound achievement; retained the existing awards and sourcing revisions.. They also posted on the talk page Talk:List of awards and nominations received by BGYO with obvious LLMTALK. On their talk page, they replied to my LLMTALK warning with Please note that I have checked the references mentioned in the CSD myself and fixed the ones that had problems. The edits are visible in the article history. I would like to understand why the whole deletion contest was collapsed instead of the sourcing points being reviewed. Could you please point me to the specific guideline being applied here? which completely ignores the fact that the relevant guideline is linked both in the collapse template and in my edit summary. Kisseran (talk) 16:39, 17 September 2026 (UTC)

there's a 4 year break between creations, starting again in June 2026 with 29 done since. All music articles which I'm hopeless at. There have been 3 creations since this was filed, Fresh (BGYO song) (22 Sept) had some AIVOCAB, AITABLE, AILEGACY ("marked" in lead), Pangram positive , BE US Live Album Showcase (22 Sept) had the gem The members described the album as encompassing different moods and themes and said its material reflected their development since their debut, some AICONNECT, AIVOCAB, + Pangram positive , @Admins willing to patrol AINB: a block please, possibly sitewide? Kowal2701 (talk, contribs) 16:30, 28 September 2026 (UTC)

User:Eliezer1987

There is a dispute at Talk:Jabalia_refugee_camp#Discussion_regarding_LLM_use about possible use of LLM on a talk page. Pinging involved editors @Katzrockso and @Eliezer1987.

Further input from additional editors here would be appreciated. IOHANNVSVERVS (talk) 16:55, 17 September 2026 (UTC)

Thanks for the ping. English is not my first language. I write my points in Hebrew first. Sometimes I asked a tool to help turn them into English for talk pages, then I posted. I understand the concern about whole posts looking like llm written. From now on I will write talk replies myself in shorter English, without that step. Happy to answer questions here. Eliezer1987 (talk) 17:18, 17 September 2026 (UTC)

Thank you for this commitment. Your English seems very good so I don't think translation tools are necessary Katzrockso (talk) 18:06, 17 September 2026 (UTC)

Agreed, your English is perfectly understandable. We'd rather talk to you directly with a bit imperfect English than going through a chatbot. ‑‑gurkubondinn 18:18, 17 September 2026 (UTC)

Thanks a lot, but is it possible to know what is allowed? Or is it better to avoid using them on talk pages altogether? Eliezer1987 (talk) 08:05, 18 September 2026 (UTC)

It is better to avoid them altogether. Fermiboson (talk) 08:57, 18 September 2026 (UTC)

This user has previously confirmed using LLMs in User talk:Eliezer1987/Archive 1 § LLM use. ‑‑gurkubondinn 18:03, 17 September 2026 (UTC)

Special:AbuseLog/40025360 (Feb 2025) had a chatgpt-derived link (which is allowed). Creations all appear to be translations from hewiki, the time in between is normal AFAICS. Spotchecks probably needed, but they appear to have been reviewed (relevant guideline is WP:LLMTRANSLATE). Kowal2701 (talk, contribs) 13:04, 20 September 2026 (UTC)

User:DrMatzahBall

LLMPROD is not needed in this case

DrMatzahBall (talk · contribs · deleted contribs · logs · filter log · block user · block log)

AITALK in Talk:Fuad I of Egypt and AISUMM all around. On mobile and will investigate further. Fermiboson (talk) 11:26, 18 September 2026 (UTC)

Userpage also obviously written by AI too. Sarsenet•he/they•(talk) 12:19, 18 September 2026 (UTC)

not triggered relevant edit filters apart from citing Wikipedia and adding unsourced content , no Paste Check, newcomer tasks though. Looked at large edits back to 15 August, nothing jumped out but haven't done spotchecks, uses semi-colons, Special:Diff/1369594679 does a weird thing w the percentage Kowal2701 (talk, contribs) 22:40, 20 September 2026 (UTC)

User:RETR0kiT

LLMPROD is not needed in this case

RETR0kiT (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Edits make mistakes that no reasonable human would make and other edits are mostly solely minor grammar edits. I don't believe the user themselves is an agent but I believe they are very heavily relying on AI to make their edits, which is fine for translations' sake, like his clearly AI replies, however Retr0 has already had an AfC draft denied because of suspected AI use.

If you look at his recent contribs, you will see three reverted edits that I flagged. The edit summaries were very clearly AI, and the edits themselves were simple grammatical edits except for the fact that they removed nearly all other page content. I can excuse that once as human error, but three times and two warnings? No human in my opinion would make that same mistake over and over again and respond an hour later to the warnings (with a clearly AI response) while in the meantime actively editing very minor grammar issues (that seemed served to the editor back-to-back and were not marked as newcomer tasks).

In my opinion, RETR0kiT is misusing AI and may damage the project if he is allowed to continue. Shield ^{\\ |T| - |C| - Solidarity \\} 18:30, 18 September 2026 (UTC)

I apologize for these errors and for the inconvenience they caused. I intended to make minor corrections, but I accidentally worked from an incomplete copy of the article and failed to notice that a substantial section was missing before publishing. The deletions were unintentional. I understand the concern, and I will carefully review the preview and diff before submitting any future edit. Thank you for bringing this to my attention. RETR0kiT (talk) 18:43, 18 September 2026 (UTC)

@RETR0kiT: you're not in any trouble, it's just better for you to be honest: are you using AI tools for purposes other than translation (i.e. queueing copyedits and executing them automatically, generating edit summaries, etc.)? I understand that you are an English-second speaker (do the Maltese speak a common language, or is it named Maltese? I'm genuinely curious), and using AI for translation purposes is perfectly fine, but undisclosed AI use for other things is generally discouraged.

By the way, sorry for not notifying you of this noticeboard listing; I received other notifications immediately after posting it and it slipped my mind. Shield ^{\\ |T| - |C| - Solidarity \\} 18:49, 18 September 2026 (UTC)

Thanks, and no worries about the noticeboard notification.

The language question is a bit complicated, but English is actually my first language. Malta has two official languages, English and Maltese.

On the AI point, the answer is also a bit more complicated than just yes or no. I use AI a lot generally, but for Wikipedia I am mostly using my own tools rather than having AI generate and post edits for me.

I have built, and am still building, my own local stack. My spell and grammar checker uses a self-hosted Qwen model through Manifest LLM Router, so I am not using a paid LLM for that. I tend to be fairly obsessive about grammar anyway, so I normally make the edits myself and then use Qwen as a second check.

I am also building a small tool for finding dead links. It is basically an automation script made in UBot Studio, which is pretty ancient technology now. Codex has helped me with parts of the code, but it is my own tool. It is still manual at the moment. I open a page, click "find", and it shows me the dead links. I also use Scrapebox for checking links.

When I am on my Debian laptop, a lot of this is easier because I can use more of my own scripts and command-line tools.

So the short version is that I am basically a human editor with a fairly advanced spell and grammar checker, plus some home-built tools for finding dead links. I am not queueing AI-generated edits and automatically posting them to Wikipedia.

If I get the dead-link tool into a decent state, I will probably share it publicly here as well. RETR0kiT (talk) 19:03, 18 September 2026 (UTC)

...so yes, you are using AI to generate at least part of your edits. For various reasons, AI usage is prohibited on Wikipedia. A dead-link tool is probably a good idea, but generating front-facing text is prohibited. guninvalid (talk) 21:37, 18 September 2026 (UTC)

I am now rewriting all tools so that it uses no AI at all. It is not a big change. I was mainly using AI because I had not bothered to finish the fully mechanical version.

I am going to pause editing for now, finish the non-AI web app, and then return to manual editing using a browser tool that has no AI component.

The tool will basically work like an assisted find-and-replace. It will identify the dead link, look for the corresponding Archive.org snapshot, and present the replacement for me to review manually.

Before I start using it, would you be willing to have a look at it and confirm that you are comfortable with how it works? I should have it live tomorrow, and I can give you a guest login so you can see exactly what it is doing. RETR0kiT (talk) 20:10, 19 September 2026 (UTC)

...there already exists a tool to do that, you don't need to reinvent the wheel. See InternetArchiveBot. msk 22:49, 20 September 2026 (UTC)

Yup, saw that, I have a spare server for the next 11 months so built one that can run 24/7 using the api finding all the repairs first, then exactly where the new link should be, then builds it into a webui queue for me to manually work through.

I get a list to work through each day, its been a good project for me. RETR0kiT (talk) 14:48, 22 September 2026 (UTC)

GenAI college course from Spring 2025

Cleanup efforts are ongoing.

LLMPRODs are done, no drafts to tag

In Spring 2025, an English course from the LaGuardia Community College had students learning how to incorporate generative AI in their academic research. A series of assignments from that course had students using generative AI to contribute to Wikipedia.

Although the course was over year ago and I don't think the participants should be punished for contributions they made a year before Wikipedia banned WP:LLMs, I am concerned about AI-generated content potentially still on Wikipedia as a result of that course. Many of the students created AI-generated articles that are still up, those being African American film score composers, Food Freedom Acts, Football economics, Transnational cinema in India, Wellness guru (the only one currently tagged for AI content), and Witches in popular culture. Should we delete these articles? Lazman321 (talk) 01:07, 19 September 2026 (UTC)

CCs: instructor @Doctorxgc and Wiki Ed liaison @Brianda. I would note that this is before the 2026 guideline change that banned LLM use in article writing. That being said, do the articles identified as problematic qualify for LLMPROD? Accessedgrant (Epicgenius mobile alt) (talk) 02:54, 19 September 2026 (UTC)

Yes, they can (and should be) deleted via LLMPROD. No LLM-generated text is "grandfathered" in, and the NOLLM guideline does not have a start date. ‑‑gurkubondinn 10:47, 19 September 2026 (UTC)

Thanks. I know that LLM text isn't grandfathered in, just wanted to clarify that the editors themselves should not be penalized based on the guideline. Accessedgrant (Epicgenius mobile alt) (talk) 12:51, 19 September 2026 (UTC)

LLMPRODs done, cleanup is done except for African American film score composers which is tagged. Content farm was edited as well and now clean. There was also Adaptive fashion which was BLAR'd Kowal2701 (talk, contribs) 23:08, 20 September 2026 (UTC)

nvm, the only other edits to that article were by the instructors ... Kowal2701 (talk, contribs) 23:11, 20 September 2026 (UTC)

I and the other instructors are significant contributors to these articles, have assumed responsibility for them, and already worked to review and rework them last year. As such, I have contested and removed the the PRODs. (Since when do instructors not count as significant contributors?)--Pharos (talk) 00:10, 21 September 2026 (UTC)

Hi, (just reasoning wrt the LLMPROD) IIRC that article was the only one that had lots of edits by you guys, but they were still pretty minimal and it wasn't at all clear that the articles had been reviewed (ik WP:LLMPRV says "significant contributors", but it should really say "significant contributions from others" to be inline w WP:BANREVERT). Does that mean there are little-to-no WP:V failures or other PAG vios in them?

Tbh I've been meaning to contact you about Talk:Artwork title#Use of ChatGPT, I wanted to do it in the most collegial way I could and avoid any drama/illwill, but we're here now ig. While experimenting did you use LLMs for any other edits/articles, and could you please give a list of them? (ofc WP:CHOICE applies and you don't have to, it would just save us a lot of time and reduce the possibility that edits you wrote yourself get caught up in cleanup efforts) Kowal2701 (talk, contribs) 00:45, 21 September 2026 (UTC)

I've restored the LLMPRODs, we don't keep AI-generated articles. Don't think that there is all that much leeway here when this was some kind of a course that was having students farm slop into the encyclopaedia. Not cool. ‑‑gurkubondinn 06:46, 21 September 2026 (UTC)

Yes, the whole point of the course was an experiment to work in-depth with students to review and correct the content they generated. The instructors' contributions were not purely in the edit history - we spent many hours with the students, and a lot of their subsequent edits came directly in response to our classroom feedback. It was not necessarily the most efficient use of time, but we wanted to do it right, and indeed to ensure that there were little-to-no WP:V failures. So yes, I think we certainly had multiple "significant contributors" involved, which is what is in the actual guideline.

No, I don't think I used LLMs for other edits/articles that I've written myself in the last 4 years since Artwork title, certainly not without disclosing it at the time on the talkpage. The goal of the initial article 4 years ago was to try something new in a transparent way, not to go on a mass editing campaign. Pharos (talk) 17:00, 21 September 2026 (UTC)

Well, none of that sounds consistent with the WP:NOLLM guideline, which I'm sure you've read but I'll repeat anyway: The use of large language models (LLMs) to generate or rewrite article content is prohibited. LLMPROD does not provide a loophole that allows you to generate content with an LLM so long as you rework it later. In fact, it is the opposite: LLMPROD requires you to affirm that the content you are trying to salvage complies with guidelines, which definitionally includes the guideline that content not be LLM-generated. Are you going to affirm that this content was not LLM-generated? Einsof (talk) 21:34, 22 September 2026 (UTC)

The LLMPROD tag has been removed again. ‑‑gurkubondinn 09:35, 23 September 2026 (UTC)

I sent the witches article to AfD. The RfCs underpinning NOLLM are clear that LLM-generated text is just not allowed, period, save for two narrow exceptions that don't apply here. Possibly we need to revisit LLMPROD to clarify that it is not a loophole to generate LLM text and then try to clean it up at the 11th hour once it has attracted scrutiny. Einsof (talk) 04:00, 27 September 2026 (UTC)

Are you going to affirm that this content was not LLM-generated?

I would also like to see this answered, since that is what removing the LLMPROD tags meant. ‑‑gurkubondinn 09:26, 29 September 2026 (UTC)

That isn't what removing the tag means, the tag quite explicitly says: To object to proposed deletion, an editor may take responsibility for the content and assume the onus to achieve consensus for its inclusion. Assuming responsibility requires that all content and its sources be closely reviewed and reworked as needed to ensure that they comply with polices and guidelines. This may involve checking that the cited sources support the content and are reliable, determining whether the content is due for inclusion, and rewriting the content to revise any tone issues and resolve possible copyright violations. Once this has been done, this template may be removed. That was the process we tried to go through more than a year ago, and that's why I removed the tags being applied scattershot to all of the articles. Now if there are actual objections to a specific article's content that is something quite different, and I have no objections to the AfD for Witches in popular culture under the present circumstances. Pharos (talk) 03:57, 30 September 2026 (UTC)

I think your stronger argument would have been that while the edit history may not have made this evident, they were collaborative, multi-editor articles (and thus not necessarily in scope for LLMPROD). Whereas having a review held before the implementation of the WP:NOLLM guideline simply isn't good enough. ChrysGalley (talk) 04:22, 30 September 2026 (UTC)

Indeed, I think I made that case earlier, at some length. But here I was replying to a very specific question about why the prod tag could be removed. Pharos (talk) 05:15, 30 September 2026 (UTC)

Again, for all content [to] comply with polices and guidelines, it must not be LLM generated, per the NOLLM guideline. It's not clear from this discussion or from the AfD that you are willing to acknowledge that fact even though it's been pointed out to you repeatedly. Einsof (talk) 05:28, 30 September 2026 (UTC)

@Pharos - if you are going to take on a LLMPROD case then the process is shown in WP:LLMPRVOBJ. And it's clearly has not been done here (a) because you simply removed the tags without doing anything more and (b) because problems clearly remain. So the very first thing I checked was Food Freedom Acts where under Purpose there is the sentence runs: Proponents argue that expanding the freedom for entrepreneurs to operate home-based food businesses fosters small businesses, innovation, and economic growth by providing opportunities for self-employment and flexible work., This is then sourced to Alec, a caucus of like minded free-market legislators, who presumably would argue exactly as you have suggested, however they have not done so on the source provided. Instead, that source is to the wording of the legislation itself, effectively a primary source, which includes wording similar to the green quote. It's pretty obviously unchecked. Had the quote said "the legislation argues that..." then it would have been fine. At the moment the source says one thing, Claude (it looks like) has said something else. One can get Claude to repeat this mistake quite easily. Now, challenge me on this, but there were 17 other discrepancies of this sort, the wording that particular article is very LLM style, and since WP:NOLLM was implemented it really should not be in Wikipedia in my view. And disclosing LLM usage on the Talk page is not covered by the guideline, it's just banned with two narrow exceptions. ChrysGalley (talk) 10:22, 23 September 2026 (UTC)

As I explained, we already did this review more than a year ago, both through giving extensive feedback to the participating students so they could correct their own work, and also through direct editing by multiple instructors. Check out the edit histories! (Also, if you had read the talk page, you would see it wasn't Claude used.) From years of experience with Wikipedia course assignments, nothing ever turns out perfect, but as I replied regarding Kowal2701's standard of "little-to-no WP:V failures", I believe we have met that. Yes, this took many hours of manual work, very likely more than if no GenAI was used at all. But the point of what we did last year was to try something innovative in as responsible a manner as possible. As to your example from Food Freedom Acts, you seem to misunderstand what the cited page from ALEC is; it is not a piece of law passed by any legislature, but a model act directly from proponents. Pharos (talk) 05:55, 24 September 2026 (UTC)

@Pharos - Since that time, WP:NOLLM has been introduced, that is what has changed. I take all you say at face value, but the rules changed and were effectively retrospective. So while what you did was quite possibly OK for the time, it's not now. There remains clear evidence of LLM language throughout certainly the FFA article, and there is a flat disconnect between source and text. The impact on you - when in terms of LLMPRVOBJ therefore changes. At that point there is a process laid down which is not voluntary if you are removing tags on LLMPROD. I think you are disagreeing that it should have been labelled LLMPROD in the first place, as opposed to "it's LLM but I've checked it". In which case you need to come here or a similar forum and get consensus for whatever you are trying to do, or go through DRV. Instead, what you have done (but didn't mention above)? You have reinstated the article again! Where I agree with you fully is that one of the big problems of LLM is that you end up way spending hours more time than on a draft compared to a standard write up without LLM. As this case is showing. ChrysGalley (talk) 06:39, 24 September 2026 (UTC)

FYI: Wikipedia:Articles for deletion/Witches in popular culture is ongoing right now. About 3/4 of the articles from this course seem to have been deleted by now, but this is by case-by-case basis, which seems reasonable, IMHO. _{Piotr Konieczny aka Prokonsul Piotrus| reply here} 03:41, 29 September 2026 (UTC)

No objection to this at all. All I've been seeking is for actual objections to a specific article's content. In that spirit, it makes sense to recognize Witches in popular culture as not as fit for purpose, and I've also re-PRODed Football economics as a similar lower quality student article. Pharos (talk) 04:05, 30 September 2026 (UTC)

• I think there is a legitimate case to see how that particular AFD fares (though the "writing is on the wall" perhaps?), in case there is some good / new rationale as to why it should not be deleted. But given the roadblocks we are seeing of LLMPROD then AFD is a clear consensus led process. I would favour any other remaining/similar articles going down this route too rather than left in mainspace. I would have though draftification (which presumably isn't allowed here due to the age of the article) would have some merit if someone was truly prepared to clear out the LLM content. Maybe that is something we think about for older pieces since the reach of WP:NOLLM may well be going back a couple of years. ChrysGalley (talk) 04:18, 30 September 2026 (UTC)

User:Sevageet

creations draftified and tagged

 Sevageet (talk · contribs · logs · block log) (19 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

 Cleanup has been requested and is being tracked at /2026-09-19 Sevageet, 2 pages need review.

The usual, rapid article creations and large edits.

• Added markdown formatted links in Diff/1375662732 (AbuseLog/45161913) to Nita Ambani, four minutes after their previous edit to the same article.
• Created Hudo (Gujarati folk dance) in Diff/1375637099 (AbuseLog/45161084) with the ISBN 978-81-956774-3-6 that does not exist (and in violation of WP:CT/SA/CASTE).
• Created Garbi (Gujarati folk dance) in Diff/1375660532 (AbuseLog/45161832) in six minutes. Tagged it and sent it to draft space, where the {{ai-generated}} tag was immediately removed in Diff/1375681385.
• Created Mer Raas in Diff/1375660033 (AbuseLog/45161820) in four minutes. Sent to draft space and tagged.
• Created Dangi Nritya in Permalink/1375639588, with the ISBN 978-81-956774-0-5 in reference nr. 2 which does not exist.
• Produced the Diff/1374920821 edit (c. 2 kB) to Vaishali Harin Gohil in 29 seconds, which they created in Diff/1374920715 (AbuseLog/45123058) with a reference to a PDF file that returns an incomplete/wrong page instead.
• Created Indian Singers' and Musicians' Rights Association in Diff/1374668567 (AbuseLog/45110876) with an AITABLE, then added the paragraph in Diff/1374668926 in one minute, the paragraph in Diff/1374669173 in two minutes, and another paragraph in Diff/1374669547 was added in two minutes.
• Created Laxman Barot in Diff/1374511276 (AbuseLog/45104336) just 14 minutes after having created Maniraj Barot in Diff/1374509351 (AbuseLog/45104289), which has previously been deleted under A7.
• Created Lagan Geet in Diff/1374495487 (AbuseLog/45103737) with a CIRCULAR reference to Wikipedia (which is still in the current revision).

There is more, but there's so much of it that I can't list it all here. The user has made just over 150 edits, and some pretty substantial cleanup will be required. ‑‑gurkubondinn 09:43, 19 September 2026 (UTC)

Thanks for pointing this out.

Okay, understood. I’ll check the issues and sources properly and slow down with new pages for now. Sevageet (talk) 10:00, 19 September 2026 (UTC)

You need to stop using AI completely, it is not allowed on the English Wikipedia. It is prohibited by the WP:NOLLM guideline. ‑‑gurkubondinn 10:04, 19 September 2026 (UTC)

@Sevageet, thank you for doing cleanup, but can you please make sure that the cited sources support the text for Vaishali Harin Gohil, Laxman Barot, Lagan Geet, and any drafts you submit? (looking at page histories I'm unsure whether you did that) Thank you Kowal2701 (talk, contribs) 14:36, 29 September 2026 (UTC)

Yes, I have checked the cited sources and fixed 3 article Vaishali Harin Gohil, Laxman Barot and the Lagan Geet. could you please check these articles are still awaiting review and are not appearing on Google yet. Thank you! Sevageet (talk) 08:32, 30 September 2026 (UTC)

Edits to existing pages have been reverted. Do we have consensus to delete the created articles via LLMPROD? ‑‑gurkubondinn 10:43, 19 September 2026 (UTC)

Support LLMPROD cleanup. Suðurhafsljósæta (talk) 10:52, 19 September 2026 (UTC)

Support. M kuhner (talk) 14:06, 19 September 2026 (UTC)

I've draftified their creations and left them a message. They've already reviewed a couple, which I've left in mainspace and need double checking (they're almost all stubs) Kowal2701 (talk, contribs) 23:50, 19 September 2026 (UTC)

I think that you forgot to update the tracking page? ‑‑gurkubondinn 08:45, 20 September 2026 (UTC)

tbh I usually wait until the mainspace redirects or LLMPRODs are deleted to update the tracker (other times I forget or leave it to whoever's going through the tracking page), should've done it for the ones needing re-review though Kowal2701 (talk, contribs) 17:15, 20 September 2026 (UTC)

I'm Kowal's stalker, I follow him around documenting his destructive activities. :-)

Want to point out that DVRTed just added "mark all red-linked articles as completed" to the Batch Edit mode of the tracker. Really helpful when LLMPRODs have been done en masse. M kuhner (talk) 17:34, 20 September 2026 (UTC)

User:Sajjadmrt

Sajjadmrt (talk · contribs · deleted contribs · logs · filter log · block user · block log)

 Cleanup has been requested.

I haven't investigated AI use in articles, though @Joy has raised the concern on Sajjadmrt's talk page and draftified Draft:Nevenka Košutić Brozović for it. User responded with AI comments twice Special:diff/1375718750 and Special:diff/1375718750, and had previously responded with a suspiciously AI sounding comment -- Special:diff/1375176507 -- when I raised concern about unattributed translation in the article Maria Mayen, in which Sajjadmrt agrees profusely and says they understand. They've since created another unattributed translation: Krystyna Secomska, which even cite in his comment to Joy, while taking no steps to repair attribution in previous translations, of which there appear to be many. I think this raises the possibility that these articles could all be unverified LLM translations. —BrechtBro (talk) 20:22, 19 September 2026 (UTC)

Regarding translation attribution: Maria Mayen has had a talk page attribution note since 16 September 2026, predating this report — "Translated content from German and French Wikipedia (July 2026); see de:Maria Mayen and fr:Maria Mayen for attribution." For Krystyna Secomska, I added the attribution note to the talk page once the concern was raised; I've since made this a standard practice for new translated articles and am happy to review older ones if specific gaps are identified. Saji • talk 09:51, 22 September 2026 (UTC)

Special:AbuseLog/44671751 (July 2026) had markdown. 77 creations since July (none before). I've given another warning and informed about LLMT and AITALK Kowal2701 (talk, contribs) 23:49, 20 September 2026 (UTC)

Regarding the July 2026 AbuseLog entry: this was disclosed and resolved at the time. I used an AI tool for the lead of one draft when I was new to Wikipedia, was asked about it directly by another editor, and admitted it immediately — see this conversation (17 July 2026). I was pointed to WP:NOLLM and have not used AI tools since. This was a single early incident from my first weeks on Wikipedia, already addressed and resolved two months ago, not an ongoing practice. Saji • talk 10:29, 22 September 2026 (UTC)

Noting signs of AI in comments both here and at Wikipedia:Articles for deletion/Nevenka Košutić Brozović —BrechtBro (talk) 17:41, 25 September 2026 (UTC)

on, I wrote an article about AI just a few lines in the sandbox and a draft and a respectable person asked me about it. I told him, 'Yeah, man,' but admitted I didn't like it and thought it was pointless. He asked about the protocol, told me I was a newbie, and that was that. After that, I gradually got the hang of things; I'm learning. You asking if I use AI for comments like 'Saji' is like me treating you to a hot dog but taking you to an ice cream parlor instead. I don't use it for this, never have." Saji • talk 19:48, 25 September 2026 (UTC)

Created Mahdiye Elahi Ghomshei in Diff/1377468488 (AbuseLog/45252032) earlier with markdown and faux-inline references. ‑‑gurkubondinn 17:38, 29 September 2026 (UTC)

For the record, there's also inline references and odd things like 11July. There's no attribution, so I'm not sure if this a translation or not. ‑‑gurkubondinn 20:31, 29 September 2026 (UTC)

For the record, I was not notified on my talk page that this report was opened, as is standard practice for noticeboard reports. Saji • talk 11:27, 22 September 2026 (UTC)

@Sajjadmrt, please don't translate from languages which you don't speak, as appears to have been the case with Jadwiga Roguska? Apologies if I'm mistaken Kowal2701 (talk, contribs) 15:06, 29 September 2026 (UTC)

@Kowal2701: Tell me the mistakes in the article. I complied with all the policies. see edit summary and talk page. Saji • talk 15:11, 29 September 2026 (UTC)

if you can't read Polish then you couldn't have read the sources and ensured WP:V was met (WP:TRANSLATE#Content). I can't tell you what problems there are because I can't read Polish either (despite my name). I tagged it with {{Rough translation}} instead of draftifying it Kowal2701 (talk, contribs) 15:20, 29 September 2026 (UTC)

If one possesses the skills to edit English texts, using Google Translate is an effective option; otherwise, it is not only useless but leads to disastrous results much like Wikipedia’s own translation system, which performs very poorly. I work on the "Women in Red" project, focusing on art historians, and am familiar with the specialized terminology of the field. I have no intention of arguing with you. If the proper use of Google Translate for Polish, meaning I obtain a raw translation and then edit it into correct English myself, violates the rules, please state this explicitly. For French, German, Hungarian, and Czech topics, I don’t need a dictionary, though I do have to use one for Polish. However, the final editing, ensuring translation accuracy and properly sourcing material is what matters most. The source materials themselves must be translated, too; it is a time-consuming yet enjoyable task for me. Saji • talk 15:52, 29 September 2026 (UTC)

How would you have any idea if the Google Translate machine translation from Polish was inaccurate when you don't understand Polish? You can't cite sources that you can't read, because how would you be able to tell if they support what you are citing to them? Translating a body of work requires a deep understanding of both languages, owning a dictionary is not enough.

For the record, this means that your machine-translated edits fail the WP:LLMT requirements. ‑‑gurkubondinn 15:59, 29 September 2026 (UTC)

@Gurkubondinn: This was the best advice I received during my time as a new editor ( all my work with 1,200 edits). Thank you; you are right. I have been active here for less than three months. Very well, from now on, I commit to writing only in languages ​​I have fully mastered or am familiar with. I also need to stop rushing and focus more on writing articles, at least at the Start class. I have honestly listed my foreign language proficiency levels on my user page. Thank you for steering me away from this troubles.

Saji • talk 16:45, 29 September 2026 (UTC)

Thank you, would you mind identifying that articles that you have submitted and are machine translations from languages that you don't speak?

For the record, I am in the second and third groups, English is not my native language. But I don't want to kill anybody. ‑‑gurkubondinn 16:52, 29 September 2026 (UTC)

@Gurkubondinn:Bro, you're awesome. (the second and third groups,LOL). I have an idea: what if I stop writing new articles for a while and revisit some of my previous ones, consulting experienced editors if I run into any issues? I’m not sure if I’ll find it enjoyable, but I think I should do it I have a commitment to the system. I will definitely send you a list of 9 or 10 items on your Talk Page by noon tomorrow. ok?

Saji • talk 17:00, 29 September 2026 (UTC)

Sure. ‑‑gurkubondinn 17:07, 29 September 2026 (UTC)

Do you speak Persian? ‑‑gurkubondinn 16:58, 29 September 2026 (UTC)

my native. yes. Saji • talk 17:02, 29 September 2026 (UTC)

Got it, what about German? ‑‑gurkubondinn 17:05, 29 September 2026 (UTC)

Level 2. It’s listed on my user page. I feel a bit uncomfortable with you asking like that. Saji • talk 17:18, 29 September 2026 (UTC)

Could someone take a look at User talk:Sajjadmrt § Mahdiye Elahi Ghomshei moved to draftspace and help me understand what is going on? ‑‑gurkubondinn 19:59, 29 September 2026 (UTC)

The "special different" thing seems like confusion over the link you posted being called Special:Diff -- it probably got Google Translated or something, because I threw that into Google Translate and the Persian did indeed seem to come out as "Special:Difference." I suspect this is some kind of built-in browser tool translating the title attribute text on the link (since that's the only place it would say that, the actual diff page doesn't have Special:Diff on it). Could also be VoiceOver or another text-to-speech tool, I'm pretty sure that reads the title attribute.

As far as the actual comment, someone new to Wikipedia has no reason to know what a "diff" is on Wikipedia. Gnomingstuff (talk) 20:24, 29 September 2026 (UTC)

User:Ticatsfan1999

Ticatsfan1999 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

 Cleanup has been requested and is being tracked at /2026-09-21_Ticatsfan1999, 29 pages need review.

User has been repeatedly warned about AI usage. Their latest creation New Orleans Hurricanes (currently tagged for CSD under WP:G15) exhibits multiple overt and blatant WP:LLMSIGNS. Needs some action at this point... Zackmann (^{Talk to me}/_{What I been doing}) 00:44, 20 September 2026 (UTC)

edits w markdown , most creations have been draftified (21 drafts), 1 live creations (PRODed), 600 edits and created a month ago. They've never made a manual edit to a talk page, have asked them to confirm they've read the final warning Kowal2701 (talk, contribs) 13:46, 21 September 2026 (UTC)

never responded and has continued editing, not seeing further signs of LLM use but worth keeping an eye on them Kowal2701 (talk, contribs) 15:28, 29 September 2026 (UTC)

Asking consensus for cleanup. M kuhner (talk) 15:13, 29 September 2026 (UTC)

Support Kowal2701 (talk, contribs) 15:29, 29 September 2026 (UTC)

Support Masterhatch (talk) 16:11, 29 September 2026 (UTC)

User:Itsetsyoufree32

Itsetsyoufree32 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

I don't know whether their more recent article creations are AI, but their older stuff, e.g., Special:Diff/1207067237, absolutely is. (some of those older articles have subsequently been GAs, have not checked what proportion of the AI slop is in the GA version, but make no mistake, it started as slop) Gnomingstuff (talk) 03:06, 20 September 2026 (UTC)

No relevant edit filter hits. Their only creation started as Special:Permalink/1179675945 (Oct 2023), is that chatbot comms or just some tool I've never seen before? They've overwritten some redirects in Sept 2026 , Special:Permalink/1373810387 (8 Sept) has some errors. They've never been warned, so I've asked them Kowal2701 (talk, contribs) 16:21, 21 September 2026 (UTC)

That's old wizard text Gnomingstuff (talk) 08:11, 22 September 2026 (UTC)

I have worked with AI agents in the past, but that was before March 2026, where I saw a guideline change was passed that people could be banned for generating articles using AI. Since that time I've avoided using it for generating text.

The current state of the Washington DC Temple's GA is well written and backed by quality sources. During the review, the reviewer went through various sources that needed to be fixed and I corrected them by hand. If there are any mistakes in the article, I am happy to rectify them.

I have been experimenting with letting AI find the sources for me, where I verify information from the sources before I post, (and where I type out the information it gives me by hand on a separate document), asking it to pull from articles to give details.

Has the policy changed since March? Itsetsyoufree32 (talk) 15:07, 24 September 2026 (UTC)

The current guideline is at WP:NOLLM.

Can you tell us during what period articles were made using LLM? This would really help with cleanup efforts and avoid the risk of messing with non-LLM articles by mistake. Thanks! M kuhner (talk) 17:32, 25 September 2026 (UTC)

AI sovereignty

 Cleanup has been requested.

• AI sovereignty (edit | talk | history | protect | delete | links | watch | logs | views)

Redwolf_Milkshake_Fox (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Newish article from a user who seems to be posting a lot of LLM-generated comments on talk pages. Topics like this tend to draw proponents and are a little suspicious to begin with, but this one looks fishy to me. But it's short, so I wouldn't mind someone else taking a look. Thanks. –Deacon Vorbis (carbon • videos) 05:11, 20 September 2026 (UTC)

Same user posted WP:ANI § User:GeogSage, and Diff/1372928039 has two unicode characters 0xE63A from the Private Use Area unused block in the Basic Multilingual Plane. This user is definitely pasting AI-generated output. ‑‑gurkubondinn 10:48, 21 September 2026 (UTC)

The content in the diff (which is a sandbox page) is copied from the source PDF (the link is at the top of the paragraph), so I can paraphrase it without switching windows. Redwolf Milkshake Fox 🦊 15:39, 21 September 2026 (UTC)

Fair enough. Do you have a link to the PDF? I'm curious to see where these characters originate from so that I can understand it better. ‑‑gurkubondinn 09:19, 22 September 2026 (UTC)

Yes, the link is on the top of the diff that you posted.

https://www.researchgate.net/publication/407164943_Structural_power_redux_geoeconomic_rivalry_in_a_networked_world

If you are using automated tools to screen abnormal characters, I'd suggest manually checking other editors' work before making definite claims. Redwolf Milkshake Fox 🦊 19:38, 22 September 2026 (UTC)

Special:Diff/1370610075 (22 August) doesn't really have any signs other than title case, + Pangram negatives Kowal2701 (talk, contribs) 16:41, 21 September 2026 (UTC)

Prodijal son

The following discussion is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

---

Cleanup efforts are ongoing and being tracked at /2026-09-21 Prodijal son, 0 pages need review.

LLMPRODs are done and drafts tagged

 Prodijal son (talk · contribs · logs · block log) (28 September 2026)
Editor has been blocked site-wide for AI/LLM policy violations. Blocked for refusing to give straight answers and for an uncollegial attitude.

Editor was given a notice about LLM use in March 2026 and warned in May 2026 but did not seem to respond to those notices and continued to create apparently LLM-generated articles. I would like help with evaluating whether there's a strong enough pattern to make some of the articles eligible for LLMPROD.

Evaluated and tagged

DraftifiedDraft:Turborealism

Needs reviewYury Usachyov (musician)  Appears to be a translation, might be ok

Needs reviewThe Bath of Pallas

Needs reviewHymn to Delos (Callimachus)  This was in DYK

Needs review1961 Fars earthquake

DraftifiedDraft:Begemotion Records

DraftifiedDraft:Harassment (novel)  Likely partly a translation

Needs reviewHymn to Artemis (Callimachus)

Needs reviewSouthern Collegiate Baseball League  Already tagged by another editor in May 2026

Needs reviewDraslovka Holding

Not evaluated or tagged

Needs reviewDelaware Sportswriters and Broadcasters Association

Needs reviewJochū (maid)

Needs reviewGRANI Center

Needs reviewPetrochemical Holding GmbH

Needs reviewDeti Picasso

Needs reviewHymn to Zeus (Callimachus)

Needs reviewCystacanth

Needs reviewAlameda County District Attorney

Needs reviewSlava Veder  Has been edited by others, but original author text is still 81%

Needs reviewSidonie (Brumov-Bylnice)

Needs reviewThe Great Wall (1962 film)

Needs reviewThe Doll Master (2023 film)

Needs review1956 Hormozgan earthquake

Needs review1932 Mazandaran earthquake

Needs review1924 Java earthquake

Needs review1911 Kerman earthquake

There are many more earthquake articles, but my hands are tired. Dreamyshade (talk) 05:32, 20 September 2026 (UTC)

markdown in The Doll Master (2023 film) (Sept 2025) (the earliest potential LLMPROD), the next creation was in Feb 2026 which got G15'd. The article you draftified (Draft:Turborealism) was August 2026. Along with the analysis you've already done wrt the tagging, it seems safe to assume all post-2022 creations were AI-generated, doing LLMPRODs now, will create tracking page as well Kowal2701 (talk, contribs) 16:49, 21 September 2026 (UTC)

Thank you, I appreciate it. I'll help with cleanup for articles not eligible for LLMPROD. Dreamyshade (talk) 16:57, 21 September 2026 (UTC)

Done, sounds good! Kowal2701 (talk, contribs) 17:09, 21 September 2026 (UTC)

Ladies and gentlemen, I am responsible for writing the articles in question. I assure you that every cited source has been verified and all statements are supported by publications. I state with full responsibility that no Wikipedia rules have been violated; all articles have been proofread, and the information is worthy of inclusion.--Prodijal son (talk) 06:47, 22 September 2026 (UTC)

So can you explain the blatant synthesis that I removed from Slava Veder, and the bogus references that led to Vitaliya Korniyenko's deletion? Rand Freeman (talk ⬩ in solidarity) 06:50, 22 September 2026 (UTC)

Did you actually write all the text yourself, with no use of LLMs? Suðurhafsljósæta (talk) 11:03, 22 September 2026 (UTC)

Looks like you were warned twice on your talk page about LLM use, in March and May. I went through several of the articles and annotated specific potential signs of LLM use, including links to scans. WP:NOLLM includes not generating article text with LLMs, even if you check the sources. Could you also answer my question at Talk:Iakov Goldovskiy? Dreamyshade (talk) 06:17, 23 September 2026 (UTC)

Dear Sir, please watch your language. Wikipedia is supposed to be a space for friendly interaction. I have no intention of justifying myself to anyone. My only "sin" is that English is not my native language. Everything I did was done to make Wikipedia better. Moving an article to draft space is easy. Why don't you try improving it instead? Oh, right, that wouldn't make you a god in your own eyes. Anyway, all the best to you all, friends. Prodijal son (talk) 12:30, 24 September 2026 (UTC)

Discussion about editor ItsMeGabeProductions split to § ItsMeGabeProductions. — Newslinger talk 02:10, 25 September 2026 (UTC)

Will admins proceed to delete these articles since it's already past Sep 26? Also looks like not all of Prodijal son's AI articles are linked here. Dora the Axe-plorer (explore) 03:51, 27 September 2026 (UTC)

See tracker. I'm finding the new-style tracker easier than the old-style list of articles in the case body. M kuhner (talk) 15:19, 29 September 2026 (UTC)

The discussion above is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

User:ProjectAutomate

 ProjectAutomate (talk · contribs · logs · block log) (20 September 2026)
This issue is currently under review. 

 Cleanup has been requested.

• ProjectAutomate (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Same old same old, AI edit summaries, markdown, etc. out the wazoo; the pattern of editing has not changed since I pointed this out on their talk page and they said they used "a tool." (Why does no one ever specify what "tool" they are using? Why do people say stuff like "clearly I used a tool" as if they are stunned to discover their own actions?) Gnomingstuff (talk) 06:41, 20 September 2026 (UTC)

Thanks for raising this. Yes I have been using AI to help with some edits. I honestly didn't see any harm in using it for things like finding grammar mistakes, checking sources and helping with wording, and I wasn't trying to hide that. I just didn't realise Wikipedia had become this strict about how AI could be used.

I understand the issue now though and I'll change how I'm using it. I'll keep it to helping with research and checking things and write the actual article changes myself. Im happy to go back and look at any of my past work, or if people think i should stop contributing I'll stop contributing. As an Engineer for many decades I do appreciate good engineering and scientifically accurate data on this platform and do want it to stay that way. ProjectAutomate (talk) 18:40, 20 September 2026 (UTC)

Thanks for responding. There's no need to stop contributing completely, just be aware that Wikipedia's guidelines don't allow the use of AI for wikitext; other edits are perfectly welcome. Gnomingstuff (talk) 23:42, 20 September 2026 (UTC)

We could really use an idea around what date you started using LLM. This allows better focused cleanup efforts and would be very much appreciated. M kuhner (talk) 23:51, 20 September 2026 (UTC)

User:Vasyatka1

LLMPROD is not needed in this case

 Cleanup has been requested.

Vasyatka1 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

I noticed many large, unnatural-sounding edits. All were sourced to obscure magazines, with no URL link. However, this edit to Psycho City included this URL, and little of what this editor included was sourced. I'm concerned that many of their large edits are AI. Thank you. Magnolia677 (talk) 19:02, 20 September 2026 (UTC)

no relevant edit filters of Paste Check triggered, needs a closer look. They've ~60k edits at ruwiki and commons, I would've thought it unlikely, but they have admitted to using AI for help w writing in English Kowal2701 (talk, contribs) 17:14, 21 September 2026 (UTC)

@Kowal2701: This edit was not supported by the source cited. I found a similar one yesterday, where the AI edit had little relation to the source cited. Magnolia677 (talk) 17:32, 21 September 2026 (UTC)

took a break on enwiki from May to Dec 2025, probably worth looking at edits from that time (they're all pretty small and adding reviews etc.) Kowal2701 (talk, contribs) 17:57, 21 September 2026 (UTC)

User:Educatedblkman1914

 Educatedblkman1914 (talk · contribs · logs · block log) (21 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Inserted a chatbot block to one of their sandbox pages in Diff/1375994375 today, with AI use going back to at least 2025 when UTM parameters showing ChatGPT use were added to a sanbox page in AbuseLog/41191196. The AI-generated User:Educatedblkman1914/sandbox3 seems to outline where upcoming AI-generated are destined for. ‑‑gurkubondinn 10:22, 21 September 2026 (UTC)

I used AI to help me analyze and organize my 31 Sandboxes so that i could identify what materials i already had could later become articles, what needed to be developed further and what sandboxes were junk...Wasn't aware that it couldn't be used in Userspaces. Im just trying to work more efficiently. Im open to guidance and advisement Educatedblkman1914 (talk) 10:56, 21 September 2026 (UTC)

Hi! The main thing to keep in mind is that you shouldn't be using AI to generate content destined for mainspace, even if it is currently in user sandboxes. Some elements of the "Mission Control" sandbox (such as DO NOT turn Mission Control into another dumping ground.) give the impression that it is part of an agentic workflow – if you are using semi-automated agentic assistance, I invite you to read Wikipedia:AI agents and the bot policy.

However, while I wouldn't recommend trusting AI on which sandboxes could potentially become articles in the future, the main question here is how much AI was involved in drafting/writing the article sandboxes themselves. Chaotic Enby (in solidarity · talk · contribs) 13:15, 21 September 2026 (UTC)

Thank you for the clarification. My original response was specifically regarding blanking Sandbox 1 and turning it into “Mission Control.” Because I want credibility to matter, I want to clarify that AI did assist me with researching, organizing sources, and drafting/revising wording in some sandboxes, particularly #23 – the James Martin II biography.

Knowing what you all have brought to my attention, I can see where things fell off the rails. I didn't have a clear understanding that AI-generated/revised article wording intended for mainspace was also an issue while it was still being developed in sandboxes/userspace. I see that now and understand that moving forward.

Additionally, because I want to do things the right way, I am willing to review the drafts in question, their sources, and correct any missteps before anything makes it to mainspace. This is an embarrassing moment, but I'm willing to do the work to address any issues. Educatedblkman1914 (talk) 17:23, 21 September 2026 (UTC)

Was the above reply performed using LLM? The English style is quite different to your first reply. ChrysGalley (talk) 17:33, 21 September 2026 (UTC)

Thank you for your response, I wanted to make sure that I specifically addressed what Chaotic Enby asked. My original response was quicker than a professional response that i would use say at work. I did consult AI to get better clarification and explicitly ask "did our work cross the line based on the provided policy?" the response was yes, but the words are mine and they are sincere. At the end of the day, I just want to write about topics I care about, this is my outlet from the day-to-day grind and I'm willing to do the work to correct the mistakes made. Educatedblkman1914 (talk) 17:52, 21 September 2026 (UTC)

It comes across that you are relying on LLM to do things of which the human brain is quite capable. I greatly value your first reply, the second reply is not you and obviously so. Thank you for being candid, that is helpful, but except in 2 narrow exceptions (which clearly do not apply to you) LLM is flat out banned. The phrase "Our work" is separately troubling - are you a human? More than one human? the Royal We? Or is this really agentic workflow? Try to get a human to reply. ChrysGalley (talk) 18:10, 21 September 2026 (UTC)

editor to editor, I understand where you are coming from. In my professional, day to day life, I use professional language to write Employee memos, and Customer-facing emails all the time. I want to be here and I want to correct the mistakes that have been made. In my narrow-sighted view of trying to write a GA quality article, lines were crossed. I'm human, I made an unintentional mistake and let things get out of hand, BUT I WANT TO BE HERE...this is passionate educatedblkman1914. No Bot, no AI, just a wikipedian who made a mistake. Forgive me and let me be better going forward. Educatedblkman1914 (talk) 18:48, 21 September 2026 (UTC)

Well I'm not the right editor for doling out forgiveness, or indulgences for that matter. I cheerfully leave editor behaviour issues to the lovely administrators. My focus is eliminating robotic text, which is what you appear to be to be doing in a somewhat unclear, unexplained and odd way over multiple sandboxes. This is ultimately a project for humans, by humans, not robots. Now the good news is that you are clearly capable of communicating and writing clearly, so if you are able to do that in the project with zero LLM then that's good. Don't worry about the odd typo or syntax error, plenty of copyeditors will do that for you, and faster than a robot. But it would also be good if you could commit to never using LLM again and tidying up the previous robotic text, but that's your call, not mine. You may find this slows you down (but I am not convinced about that either) and more focus on smaller text changes, rather than great slabs of LLM-speak, but that, I hope, is a more rewarding editing experience for you. ChrysGalley (talk) 19:17, 21 September 2026 (UTC)

User:Harold9595959

 Harold9595959 (talk · contribs · logs · block log) (21 September 2026)
Editor has been blocked site-wide for AI/LLM policy violations. Blocked for persistently adding unsourced material.

There's a chatbot block in Diff/1376003727 (AbuseLog/45179159) to Qamar Adegoke and in Diff/1376006404. Ignored my warning and continued editing the same article.

Created Lymore Gardens F.C. in Diff/1332787617 (AbuseLog/43165387) and Chay Tilt in Diff/1354703100 (AbuseLog/44233787), both with UTM parameters showing ChatGPT use. More UTM parameteres in Diff/1339467807, Diff/1345177164, Diff/1364150681, Diff/1364153216, Diff/1369977981, Diff/1369978247, Diff/1370926681, Diff/1370930726, and Diff/1370951270.

There is a gamut of warnings on their talk page about unsourced content in §§ January 2025, February 2025, May 2025, June 2025, November 2025, January 2026, April 2026, and June 2026, adding something called § Micronesia Football Federation Tournament to existing articles that doesn't seem to have been supported by any sources, copyvio notices, etc.

Denied AI usage in § LLM use on their talk page in May 2026. ‑‑gurkubondinn 12:48, 21 September 2026 (UTC)

Blocked by ScottishFinnishRadish for persistently adding unsourced material. Not sure if the block was in response to this AIN post or if this user was reported somewhere else as well. ‑‑gurkubondinn 13:08, 21 September 2026 (UTC)

That was fast, thanks a lot! WP:LLMPROD it is then? Chaotic Enby (in solidarity · talk · contribs) 13:17, 21 September 2026 (UTC)

Yeah.. This was a quite prolific user, there's over 3000 pages (got it down to ~2000 by filtering out anything smaller than 100 bytes) and 162 created pages. ‑‑gurkubondinn 13:26, 21 September 2026 (UTC)

During recent NPP backlog work, I found and nominated two of User:Harold9595959's creations (Edward Hatt, George Lock) for deletion, and began looking at others in their history (latest AfDs today: Toby Holmes, Chay Tilt, for example; and I also draftified Adam Moseley), and am now thinking that many/all of their creations might need reviewing. Is there a process for cleaning-up such user's contributions? Cheers. Paul W (talk) 14:01, 21 September 2026 (UTC)

We tag all of them with WP:LLMPROD and delete them, more or less. If you think some of them could be saved you could draftify them, but from experience you will quickly get exhausted. Fermiboson (talk) 13:28, 23 September 2026 (UTC)

I didn't look terribly deeply into verifying widespread LLM use, the unsourced content after two blocks was enough for me to action without getting into the hairy edges around blocks for LLM use. ScottishFinnishRadish (talk) 15:50, 21 September 2026 (UTC)

Lukeino Formation (Feb 2024) had AIREDCATs but no other signs. Palaeochersis (March 2025) had AIREDCAT but is a stub. Ernst Jean-Baptiste (June 2025) had a red cat, but is a sports bio so nothing else. So did Participation of non-top flight clubs in UEFA club competitions (July 2025), it's a table. Special:Permalink/1313887172 (Sept 2025) has citations like [5] w no link, but isn't a translation from anywhere AFAICT. First chatgpt param was Jan 2026. Spotchecks are probably needed to determine the start date unfortunately Kowal2701 (talk, contribs) 17:44, 21 September 2026 (UTC)

Not sure 2024 is the start point, Special:PermanentLink/1266218693 (late 2024) fairly clearly not AI, grammatical errors + not doing stuff AI species articles virtually always did at that point in time

Special:PermanentLink/1280998653 (March 2025), at least the last part of it, seems fairly clearly AI

to further confuse matters, copyvio may also be in the mix in some places, like Special:Diff/1280408666 (also March 2025, outright admits it) Gnomingstuff (talk) 03:17, 25 September 2026 (UTC)

User:VirtualStoic

 VirtualStoic (talk · contribs · logs · block log) (21 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Added markdown to The Three Little Pigs in Diff/1376003067. When notified of NOLLM in § September 2026 on their talk page, wanted to know specifically what part of that sentence or external-link addition [I believed] was AI-generated and then claimed to have been editing for years, created pages, and made numerous edits. The user has made 11 edits and created exactly 0 pages.

This might be a bad case of AITALK with hallucinations, which should probably be stopped before it gets out of hand. ‑‑gurkubondinn 13:37, 21 September 2026 (UTC)

My suspicion is that VirtualStoic is not a native English speaker and is using AI translation software. They may not even be aware that the software in question is LLM-based, nor that this is the reason their edits introduce errors, especially to external links in citations. Their contention that they are a prolific long-term editor is probably because they are at the Slovak Wikipedia — see sk:Špeciálne:Príspevky/VirtualStoic. They are likely unaware that their history at another project has no bearing on their history on this one. • a frantic turtle 🐢 solidarity 15:33, 21 September 2026 (UTC)

It’s quite impressive that, while nobody has managed to create a reliable AI detector, @Gurkubondinngurkubondinn can apparently determine with 100% certainty from a single sentence that AI was used,lol. Also, when I said I had been editing for years and created pages, I was referring to my Wikipedia activity generally, especially on Slovak Wikipedia and not claiming I had created pages on English Wikipedia.English is not my native language, so yes, I may use language or translation assistance. That is not the same as inventing facts or sources with AI.

If there is a concrete issue with the edit itself, I’m happy to discuss it. Otherwise, I don’t see much value in continuing speculation about how my English sentences were written... Guys, you’re wasting both your time and mine here when you could be dealing with actual bots on Wikipedia. VirtualStoic (talk) 15:46, 21 September 2026 (UTC)

The issue is that the AI translation software you’re using is introducing errors to public-facing pages. We’d like you to stop using that software for that reason alone, but we also have policies that prohibit LLM-based editing in general — and your translation software would seem to qualify. If your English proficiency isn’t up to editing directly in English and you feel you have to use AI translation software, then you shouldn’t be editing the English-language Wikipedia. • a frantic turtle 🐢 solidarity 16:14, 21 September 2026 (UTC)

• I would strongly suggest that VirtualStoic be less blasé and more honest about their LLM usage. Whether you "see much value" in following our policies or not is irrelevant -- either you're going to follow them, or we will show you the door. Snarky comments and personal attacks directed at editors trying to clean up after you, is absolutely not going to help you here. ⇒SWATJester ^{Shoot Blues, Tell VileRat!} 23:07, 21 September 2026 (UTC)

User:Tiny Particle

 Tiny Particle (talk · contribs · logs · block log) (21 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Chatbot blocks in Diff/1376016651 to 1969–70 Aston Villa F.C. season and Diff/1376013706 to 1966–67 Aston Villa F.C. season. This user is the main contributor to these articles, so I didn't know how far back to revert without knowing when the AI-usage started.

There are UTM parameters showing ChatGPT use in f.ex. AbuseLog/44542081, AbuseLog/41938007, and AbuseLog/40543667. Have not read through all filter log entries yet. ‑‑gurkubondinn 14:55, 21 September 2026 (UTC)

Special:AbuseLog/39999333 (Feb 2025) triggered filter 1325 for "rich history", in August 2025 they had a series of articles deleted for OR/GNG, a chatgpt-derived link was found. First chatgpt param was Special:AbuseLog/40543667 (April 2025), there're a ton of them since Kowal2701 (talk, contribs) 17:53, 21 September 2026 (UTC)

Seems to be ongoing. There is an UTM parameter showing use of ChatGPT in AbuseLog/45190126 today, before the page was copypaste moved into article space. ‑‑gurkubondinn 20:45, 22 September 2026 (UTC)

I would posit that you are picking up a lot of false positives. How do I get the boiler plates removed from these articles. Tiny Particle (talk) 20:54, 22 September 2026 (UTC)

Did you, or did you not, use any LLMs in the process of your edits? Fermiboson (talk) 13:25, 23 September 2026 (UTC)

My usage was in accordance with Wikipedia:Writing articles with large language models. Every single line of data has been manually checked and referenced to the match report.Tiny Particle (talk) 17:18, 23 September 2026 (UTC)

So I take it that's a yes, which means it was not a false positive. How did you use LLMs in your edits? Fermiboson (talk) 17:37, 23 September 2026 (UTC)

This would be an example of a false Positive. I get thousands of links from ChaGPT. That I used one of thousands as a valid reference is incidental. The reference is valid and at the time I would not have perceived it was an issue. I have been editing since 2007 so I've written a ton of material before ChatGPT and I think my edit history shows I have sufficient pride in my contributions to encyclopedic content to not want it polluted by AI slop.Here I asked it to reformat the table as there was no place to put a link to the match report. Now each match has a match report and each line of results has been manually edited to a reliable source. Tiny Particle (talk) 19:37, 23 September 2026 (UTC)

creations appear human-written to me (I looked back to mid-2025 and recent ones, there's some overbolding, but no other signs and several signs of human writing), I think what they mean is it's a false positive wrt to finding AI-generated content. I think this can probably be closed and archived if no-one objects? Kowal2701 (talk, contribs) 19:46, 23 September 2026 (UTC)

User:Bwoart

• Bwoart (talk · contribs · deleted contribs · logs · filter log · block user · block log)

The usual: AI edits going back to 2025, complete with the usual AI edit summaries, filter log full of hallucinated templates, circular refs and such, a lot of typical deflections and silent removal of AI tags including one just yesterday. They have been previously warned for this and do not appear to care. Gnomingstuff (talk) 19:54, 21 September 2026 (UTC)

notified them Kowal2701 (talk, contribs) 13:43, 22 September 2026 (UTC)

Did you use AI to determine if I used AI? It's a tool to speed up research and find sources. The information is all reviewed and sourced properly. Please refrain from throwing a blanket tag on the articles without verifying accuracy. If the references are repeated that article can always be edited later. I removed the tag because I go back and check to see if there may be "hallucinated" information. If the sources support the claim I remove the tag. If there is an easy way to find templates in Wikipedia let me know otherwise I use similar articles. Bwoart (talk) 14:42, 22 September 2026 (UTC)

@Bwoart, AI-generated content is prohibited on Wikipedia, it doesn't matter if the content is reviewed. And no, we detect such content manually. You can use AI to help with finding sources and research (WP:LLMRESP), but not for generating content Kowal2701 (talk, contribs) 16:09, 22 September 2026 (UTC)

Please refrain from throwing a blanket tag on the articles without verifying accuracy.

You're the one who needs to "verify accuracy" by giving a yes-or-no answer as to whether you used AI, which so far you have not done. Gnomingstuff (talk) 02:35, 23 September 2026 (UTC)

User:Yeshu972

LLMPROD is not needed in this case

• Yeshu972 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Special:Diff/1375385357 has one of the most blatant AI edit summaries I've seen in a year, and that's recent; the filter log is full of various hits for nonexistent templates, possible AI text, etc. Most of their recent edits are small/NOLLM-compliant copyediting, but not all are, they are being made minutes apart, and it's hard to tell at a glance the difference without going in and checking every one. Gnomingstuff (talk) 04:32, 22 September 2026 (UTC)

only relevant filter hit was Special:AbuseLog/43098688 in Jan 2026 but it's for a sign that gives high false positives (the others were actual false positives), non-existent templates are also from Jan 2026 onwards. No creations so ordinary edits need analysis Kowal2701 (talk, contribs) 13:51, 22 September 2026 (UTC)

User:TheBirdsShedTears

• TheBirdsShedTears (talk · contribs · deleted contribs · logs · filter log · block user · block log)

High-volume AI use at least since 2025; ChatGPT params in Special:Diff/1305991611; appears to be still editing, a lot probably could use LLMPRV Gnomingstuff (talk) 15:22, 22 September 2026 (UTC)

glancing at some creations, not seeing obvious signs, they don't appear to have been warned for AI use, I've asked them Kowal2701 (talk, contribs) 20:23, 23 September 2026 (UTC)

I don't see any of the usual AI issue stuff in their things, did a quick check. AFAIK using AI to research is also not a problem as long as you actually read through the whole source and use it only to support what's actually in it and not what chatgpt claims. Jishara (please ping upon response) (talk) 00:43, 25 September 2026 (UTC)

I mean just to take a few examples of many:

• Special:PermanentLink/1355214892: Disclaimer on lack of sourcing, canned emphasis on association
• Special:PermanentLink/1355981327: Disclaimer on lack of sourcing, AIATTR, overgeneralization (Contemporary accounts suggest that he served as a minister responsible for Border and Tribal Affairs in the early PDPA cabinet cited to only one document that incidentally also does not mention that he was responsible for the border)

Gnomingstuff (talk) 01:23, 25 September 2026 (UTC)

User:BLTSandwicher

Cleanup efforts are ongoing.

• BLTSandwicher (talk · contribs · deleted contribs · logs · filter log · block user · block log)

This user is writing single shot articles at a high clip. The subjects largely seem notable; however, at least one has been flagged with an AI-generated template: Draft:NBA Entertainment League, Draft:Maryland Mustangs, Draft:New Jersey Swamp Dragons, Draft:Washington Justice (basketball), Draft:Washington Metros (WBL). - Wil540 art (talk) 15:41, 22 September 2026 (UTC)

I reverted their edits and declined the drafts, worth keeping an eye on Kowal2701 (talk, contribs) 19:37, 23 September 2026 (UTC)

responded on their talk w AI Kowal2701 (talk, contribs) 20:42, 23 September 2026 (UTC)

Stockton High School (Illinois) signs of AI use

signs of AI use Stockton High School (Illinois) by @Issac I Navarro

LLM phrasing "highlights its importance" and Potential notes published in text reference/citation:[6][7]

First edit 2024-08-23 23:20

Total edits 12,038

Unique pages edited 6,059

~2026-51227-54 (talk) 00:27, 23 September 2026 (UTC)

The line "highlights its importance" was added in Old revision of Stockton High School (Illinois). Issac I Navarro (talk) 00:33, 23 September 2026 (UTC)

2024's still in range for AI (and that phrasing is more typical of 2024 than 2026), looks like it was added by User:Dfsjlk1 as part of a spree of article expansions (Newcomer Tasks strikes again!), several of which (e.g. Special:Diff/1241459378) also seem like AI Gnomingstuff (talk) 03:10, 23 September 2026 (UTC)

@Gnomingstuff, I am looking at that history section, and while this may bring the article down to a stub class, I think most if not all of it could be WP:RMV. Issac I Navarro (talk) 03:18, 23 September 2026 (UTC)

Somehow I missed that Dfsjlk1's userpage outright says "edits are ai-assisted" so yeah Gnomingstuff (talk) 05:00, 23 September 2026 (UTC)

Special:AbuseLog/39693065 and Special:AbuseLog/44139977 triggered filter 1325. Special:Permalink/1258533023 (Nov 2024) had markdown link and red cat. LLM use appears to stop early- to mid-2025, @Issac I Navarro?

For Dfsjlk1 Special:AbuseLog/38521674 (August 2024) cited Wikipedia, only live creation is Leif Erickson Park (Sept 2024) which had a category that was deleted 6 years ago, AITITLECASE, AILIST, mostly unsourced, AITREND (as part of the City of Saskatoon’s broader efforts to revitalize local parks), AIVOCAB ("highlights"), AI-tagged by Gnoming in Jan; LLMPRODed. Active from March 2024 to Jan 2025, see Special:Diff/1241078886 (their first large edit) Kowal2701 (talk, contribs) 11:49, 23 September 2026 (UTC)

@Kowal2701, I joined in February 2024, and if you look at many of my earliest of edits, you’ll see that they contain quite a mountain of mistakes. All me, trying and failing to contribute.

Of course that is no longer the case, I have since tried to go back and rectify many mistakes I made when I was WP:NOOB. That is partly why I joined WikiProject Schools, since many of my first edits were related to schools articles that I unfortunately mucked up because I really didn't know what I was doing. And probably gave too many overly kind editors a headache.

That is why, I always assume good faith with others as there was a time when I had no clue but wanted to contribute. It is just unfortunate that many of my earliest ones are kinda of nonsense; looking back.

There is also: https://xtools.wmcloud.org/pages/en.wikipedia.org/Issac%20I%20Navarro/all and Issac I Navarro (talk) 12:15, 23 September 2026 (UTC)

Can you give a date range in which we should look for LLM-written articles and edits? It would be really helpful in guiding cleanup, and can protect valid articles/edits from possible removal. Thanks! M kuhner (talk) 05:24, 25 September 2026 (UTC)

(just want to reiterate -- the text pointed out here was not added by Isaac, it's User:Dfsjlk1 whose userpage discloses it's AI Gnomingstuff (talk) 18:58, 25 September 2026 (UTC)

User:HistorianSemiAmateur

LLMPROD is not needed in this case

 Cleanup has been requested.

• HistorianSemiAmateur (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Markdown in Special:Diff/1369663455 (the links), paste check flags, and AISIGNS consistent with date, small sampling:

• Special:Diff/1322623633 - canned emphasis on significance/superficial analyses, AIVOCAB density (2024-25)
• these edits - disclaimer on lack of sourcing, canned emphasis on broader trends

(Note: According to their user talk page their account got hacked at some point, I don't know what's going on with that, but it appears unrelated.) Gnomingstuff (talk) 02:42, 23 September 2026 (UTC)

earliest Paste Check is 22 July, on 2 Feb they removed a tag you added (Special:Diff/1336156314) to an article they wrote in Nov 2025 Kowal2701 (talk, contribs) 11:31, 23 September 2026 (UTC)

Hi guys,

No AI has been used in any content editing.

I grew up in Somerset East, went to Gill College. Its a small town in the middle of the eastern cape province of south africa, the wikipedia entry is important for the local townsfolk, and a lot of the information comes from either personal experiences or from that of the inhabitants. HistorianSemiAmateur (talk) 11:20, 24 September 2026 (UTC)

@HistorianSemiAmateur - thanks for joining up here, but are you aware that writing up information from personal experiences, as opposed to reliably sourced, is not permitted? ChrysGalley (talk) 11:35, 24 September 2026 (UTC)

Highlight where I wrote up on personal experiences? The article covers the formation of Somerset East some 200 years ago. I am 60. There is hardly any overlap, is there?! But, for example, the publication by Jakes Gerwel on Nojoli is mentioned nowhere on the web and is hidden obscurely on the Foundations website. . Yet, folk who lived through the recent namechange, contributed to it.

So please dont try and be a smart Alec, its not cool and not warranted. HistorianSemiAmateur (talk) 11:48, 24 September 2026 (UTC)

Well you did say and a lot of the information comes from either personal experiences or from that of the inhabitants. I am not trying to outsmart you, we are just trying to ensure Wikipedia's articles are clean of AI, and conform the policies and guidelines. A lot of people here are seemingly unaware of them, specifically the near-total bar on AI usage. Plus I am older than you. ChrysGalley (talk) 11:56, 24 September 2026 (UTC)

What has your age got to do with the price of anything?

And of course, certain entries like recent news on the town (see Challenges post name change) wrt service delivery, the collapse of the aerodrome etc is based on the personal experiences of the townsfolk, captured by reporters in local news outlets and appropriately referenced in the Wikipedia entry. Wikipedia entries are alive, not limited to the history of an encyclopedia that was updated decades ago.

Can we move on to more pleasant things please?! HistorianSemiAmateur (talk) 12:17, 24 September 2026 (UTC)

I'm not sure how pleasant you may find reading Wikipedia:Verifiability, but you probably should: In the English Wikipedia, verifiability means that people can check that facts or claims correspond to reliable sources. Wikipedia's content is determined by published information rather than editors' beliefs, experiences, or previously unpublished ideas or information. Even if you are sure something is true, it must have been published in a reliable source before you can add it... AndyTheGrump (talk) 12:30, 24 September 2026 (UTC)

AndyTheGrump, thank you for the recommendation. I will be sure to read it and refresh my knowledge base. .

However, can you point to any of my entries that cannot be verified ? And in many cases, what is a reliable source? HistorianSemiAmateur (talk) 12:46, 24 September 2026 (UTC)

So for example in Gill College you recently added some text in respect of Jan Pieter de Lange. The entire paragraph is unsourced, as is much of the rest of the article. Normally one would want to see reliable sources, where WP:RSP can guide you (more on what is not a reliable source). Anything without a reliable source can/should be deleted, and in the case of that particular article it would be quite a lot of it. ChrysGalley (talk) 13:00, 24 September 2026 (UTC)

Hi ChrysGalley, if you click on the hyperlink, it takes you to a fully referenced article on De Lange. HistorianSemiAmateur (talk) 05:10, 25 September 2026 (UTC)

@HistorianSemiAmateur - I am afraid that's WP:CIRCULAR, which is part of the Verification guideline that Andy kindly mentioned earlier. You added in a detailed paragraph, nothing wrong with that, but you did not add in any reliable sources. Moreover this and similar edits have left the entire page on the college with very few sources, which is against the guidelines and isn't helpful to our readers. You were not the only editor to work on that page, so by no means is that your fault, but the onus is on you as editor, when adding content, is to provide notability and verification through reliable sourcing. Pragmatically if there is a wikilink then you can get away with a few words, if it's not a living person, but not whole paragraphs. More realistically if there is a Wikilink in place it's often best to work on the "less is more" principle, since the risk of WP:SYNTH is very much in this space. ChrysGalley (talk) 06:24, 25 September 2026 (UTC)

Noted, I think let me reduce the content and redirect readers to the hyperlink - there is no need to almost summarise the entire article HistorianSemiAmateur (talk) 06:36, 25 September 2026 (UTC)

Good, and should you find more reliable sources about the college (generally) as you go then be sure to add them in. I'm not entirely convinced that this issue is essentially AI in origin, but you can perhaps appreciate that when big slabs of text go up without verified sources then it blips on the AI radar, or gives a feeling that AI was used along the way. AI isn't just banned for content, it's banned for everything, with 2 narrow exceptions, and some of the updates (such as the first Diff above) are very AI-ish in tone. But thank you for continuing to improve the text of Wikipedia, that's very much the name of the game. ChrysGalley (talk) 07:35, 25 September 2026 (UTC)

Will do and noted!

Hows the health? HistorianSemiAmateur (talk) 09:24, 25 September 2026 (UTC)

Thanks for asking, fine for today! (Not that we we encourage chit-chat here) ChrysGalley (talk) 09:43, 25 September 2026 (UTC)

if you ever want to chit chat, (Redacted) HistorianSemiAmateur (talk) 11:09, 25 September 2026 (UTC)

User:JL 09

 Cleanup has been requested.

• JL 09 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

 JL 09 (talk · contribs · logs · block log) (24 September 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. User's autopatrolled permission has been removed.

Isn't it great when someone who's been an editor since 2009 suddenly starts using AI? (no, no it is not)

The usual, spate of high-volume articles created minutes apart, AI signs in prose that match the date of creation (and, accordingly, do not read much like one another stylistically), sample:

• Special:Diff/1374507297: September 2026, WP:AISIGNS including the "X identifies him as Y" AIATTR variation that's really popping off in recent AI text (plus several related variations)
• Special:Diff/1345638445: March 2026, indicators including superficial analyses, AITREND, AILIST/AIBOLD combo, etc.
• Special:Diff/1307588168: August 2025, indicators including treating article title as proper name, superficial analyses, 2025-era AIVOCAB, etc

Cutoff date appears to be May 2025, this is when the high-volume article creation/expansions started and also the filter log started filling up with a whole lot of things that if you have been around since 2009 you probably would know better than to do. Gnomingstuff (talk) 04:58, 23 September 2026 (UTC)

adding old {{lang}} templates back to 8 August Special:AbuseLog/44831193. These aren't translations. Start date looks to be May 2025 for creations (more analysis on that needed though). @Admins willing to patrol AINB: can autopatrolled please be removed? Kowal2701 (talk, contribs) 11:24, 23 September 2026 (UTC)

Autopatrolled revoked. ~Anachronist (who / me) (talk) 06:24, 24 September 2026 (UTC)

User:Avi8tor

LLMPROD is not needed in this case

 Avi8tor (talk · contribs · logs · block log) (23 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Claims in User talk:Avi8tor § September 2026 to using Google Translate to produce machine translations from frwiki, but Diff/1375998370 to Bourg-Charente and Diff/1376000280 to Chaniers are full of markdown. Somehow Diff/1375584832 to Mussidan has both markdown and valid wikitext syntax for italics, and Diff/1374656536 to Trois-Palis was produced in four minutes and has markdown. Attribution is lacking on most of these machine translation edits. ‑‑gurkubondinn 10:47, 23 September 2026 (UTC)

a lot of non-existent templates Kowal2701 (talk, contribs) 11:17, 23 September 2026 (UTC)

At the top of the page for the above edit on Bourg-Charente is the following text:

View a machine-translated version of the French article.

Machine translation, like DeepL or Google Translate, is a useful starting point for translations, but translators must follow the LLM translation guideline, revise errors as necessary and confirm that the translation is accurate, rather than simply copy-pasting machine-translated text into the English Wikipedia.

Do not translate text that appears unreliable or low-quality. If possible, verify the text with references provided in the foreign-language article.

You must provide copyright attribution in the edit summary accompanying your translation by providing an interlanguage link to the source of your translation. A model attribution edit summary is Content in this edit is translated from the existing French Wikipedia article at fr:Bourg-Charente; see its history for attribution.

You may also add the template {{Translated|fr|Bourg-Charente}} to the talk page.

For more guidance, see Wikipedia:Translation.

This would suggest Google Translate is good for translating articles as long as the translation is checked. This is the first time I've seen the attribution sentence. Avi8tor (talk) 12:54, 23 September 2026 (UTC)

This does not address the concerns for wider AI use. Fermiboson (talk) 13:06, 23 September 2026 (UTC)

Is google Translate classified as AI? It's been around for years probably since around 2014. Please explain why pages that state you can improve an article by using the French article translation using Google translate. Avi8tor (talk) 07:52, 24 September 2026 (UTC)

Firstly Google Translate is powered by an LLM backend because people can prompt-inject it, yes, but that's besides the point. The point is that we don't believe you're only using Google Translate. Fermiboson (talk) 08:14, 24 September 2026 (UTC)

Google translate is all I'm using. Avi8tor (talk) 10:54, 24 September 2026 (UTC)

Google Translate isn't typically capable of outputting markdown Kowal2701 (talk, contribs) 10:57, 24 September 2026 (UTC)

What is prompt-inject it? Why not try it yourself. Avi8tor (talk) 10:56, 24 September 2026 (UTC)

I'm also not familiar with markdown, if you could please explain. Avi8tor (talk) 11:00, 24 September 2026 (UTC)

The English comprehension level displayed here, which is borderline unable-to-communicate, and the prose written in articles are also irreconcilably different. For your future reference, your LLM is probably not translating our comments to you faithfully, especially if you're using Gemini, based on my experience. (Ton IA, notamment si tu utilises Gemini, ne traduit sans doute pas nos commentaires de manière exacte et vice-versa.) @Admins willing to patrol AINB: can we get a duck mainspace block or do I need to do a source analysis? Fermiboson (talk) 11:05, 24 September 2026 (UTC)

Sorry, you're on another wavelength to me, I'm not using any Large Language Model. Are you talking about this conversation or the article on Boourg-Charente? I'm not a computer geek, I'm a retired pilot who tries to improve Wikipedia. It looks like I've been blocked from editing for doing what was suggested on the Wikipedia page I edited, which suggested I use Google Translate. If this is not acceptable then tell me what it acceptable. Avi8tor (talk) 11:18, 24 September 2026 (UTC)

We are saying that there are technical signs in your edits that are more indicative of an AI chatbot like ChatGPT being used than Google Translate. ‑‑gurkubondinn 14:05, 24 September 2026 (UTC)

@Fermiboson, Gurkubondinn, and Kowal2701: Google Translate has apparently made some recent changes to incorporate its Gemini LLM even more extensively than it used to. I just translated the current revision of the Chaniers article on the French Wikipedia to English with Google Translate, and it does indeed match Avi8tor's edit, including the addition of Markdown and extra em dashes. Based on this, I do not believe a block is not warranted at this time.

I see that the latest version of the Google Translate mobile app has a "Choose text translation model" toggle, with the Gemini-based "Advanced" option selected by default, and the "Classic" option available only if you explicitly choose it. (This toggle has been reported since November 2025.) Unfortunately, I don't see this toggle on the Google Translate website, so I believe Google is opting users into Gemini-based translations in certain situations without disclosure or a way to opt out. — Newslinger talk 01:31, 25 September 2026 (UTC); struck incorrect observation, please see follow-up comment 11:35, 25 September 2026 (UTC)

What the actual ...? I had no idea, I only use Google Translate in a browser so had not seen this. That is annoying, they've just made it worse then.

I still don't see the point in saving static copies of machine translations that anyone can produce themselves, but that explains this. ‑‑gurkubondinn 01:43, 25 September 2026 (UTC)

Well. @Avi8tor, please accept my apologies. What you did was indeed within policy, although it probably shouldn't have been. In short, the policy was written at a time when Google Translate was not powered by LLMs in the way that it is today, and with the way it is nowadays, I do not believe using Google Translate to translate an article is an appropriate machine translation, although it is permitted according to the letter of the law. Your edits contained certain incontrovertible signs of LLM use, and as we were not aware that Google Translate is in essence an LLM sometimes, we believed you to be dishonest regarding the extent of your LLM use. This is not your fault, but it seems that using Google Translate will also expose you to the laundry list of pitfalls and hallucinations that are endemic to unreviewed LLM translations, and I recommend against doing so at this time.

We might need to take a look at translation guidance and rule out Google Translate as a machine translate tool. But that would more or less just nuke the policy, no? Fermiboson (talk) 03:23, 25 September 2026 (UTC)

Thanks, I was going to suggest someone do what was just done with the same article using translate.google.fr but you got ahead of me on that. So what in everyone's opinion would work to translate articles from French to English or any language that would comply with the requirements? I updated the English articles recently on towns along the Charente river because I have friends coming from the USA and we are going down the River, I wanted them to read up on the local towns in Wikipedia, but they were missing a lot of information that was in the Wikipedia French article. Avi8tor (talk) 07:51, 25 September 2026 (UTC)

What is the difference between you doing that and them just reading the frwiki articles with Google Translate? ‑‑gurkubondinn 09:33, 25 September 2026 (UTC)

They are not in the same location at present, I can suggest it but I don't know how computer literate they are. On an iPad it can be quite difficult/time consuming to copy and paste. Does this analysis of Google translate mean that DeepL falls into the same category of LLM as Google translate?

Does the analysis suggest the other articles you reverted are now OK? Reading thru them they looked fine with the exception of some citations which brought up errors. Avi8tor (talk) 10:38, 25 September 2026 (UTC)

Yes, DeepL also falls under it because it is using an LLM. From what Newslinger said, you can still use Google Translate. You would either need to just use browser version or by selecting the "Classic" option in the app if that is what you are using (though I would recommend using a computer with a keyboard to work on Wikipedia). I would encourage you to just use it as a starting point though, and then manually rework the text (and wikitext) that Google Translate gives you. Think of it like an advanced dictionary, and not something that does the whole job for you.

I understand your point about computer illiteracy. But what I'm getting at is that I don't think that it provides any real value to the project to paste in static machine-translations of an frwiki article because anyone that wants to read the frwiki article in English can just do that themselves. This is just my personal opinion though. There is a real value to translations, but I dislike how the work of translating a body of work is sometimes looked at as a trivial task when it actually requires a deep understanding of both languages and careful diligent work. Languages don't map one-to-one to each other, and an expressive language like French is quite different from a language like English, so it is also about localising the text and not just translating the words themselves (there's a reason that the aviation world uses English everywhere). A good translation (done by a good translator) is it's own body of work, and it is a skill that is worth developing.

You also need to keep in mind that the different Wikiprojects are independent from each other, and so they will have different policies and guidelines. Just because something meets the frwiki policies doesn't automatically mean that it will meet the policies here on enwiki. I don't speak French myself, so I am pretty unfamiliar with the frwiki policies and guidelines but I do contribute to other langauge Wikipedias which have policies and guidelines that are different from the enwiki policies and guideline.

And remember to provide a WP:TFOLWP-compliant attribution, that's a legal requirement of the CC-BY-SA license so it's not optional. Apologies for assuming that you were intentionally being deceptive by saying that you had used Google Translate. The edits did show technical signs of being produced with an LLM, and as it turned it that was correct, but you didn't know about it. It's not your fault that Google are less than transparent about this. You are very clearly here in good faith and because you want to improve the encyclopaedia. ‑‑gurkubondinn 11:13, 25 September 2026 (UTC)

Correction: I found that Google Translate's website and app both show the model selector, but only if you specify a source language (e.g. French) instead of using "Detect langauge". (In my previous comment, I had incorrectly said that the website did not have the model selector.)

I've posted more details at WT:LLMT#Google Translate's "Advanced" model (Gemini-based). — Newslinger talk 11:34, 25 September 2026 (UTC)

Thanks, I'll refer to this information next time I do this and try to find the appropriate google translate. Much appreciated. Avi8tor (talk) 13:37, 25 September 2026 (UTC)

User:Daksh121

 Cleanup has been requested.

 Daksh121 (talk · contribs · logs · block log) (23 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

 Dakshv (talk · contribs · logs · block log) (23 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

DOA 404s, AIATTR, AIREDCAT, 100k of edits in ~2 hours today. I reverted Vajrai Waterfall to the last good revision, List of waterfalls in India by height, Harishchandragad, Kalsubai, Aadrai Jungle Trail, and Kalu Waterfall are their other contributions as of my post. ~ A412 ^talk! 15:32, 23 September 2026 (UTC)

I have draftified draft:Trekkers Tribe, it's basically an AI written advertisement. ChrysGalley (talk) 15:51, 23 September 2026 (UTC)

Was just about to post this. One of their recent edits had clear WP:AISIGNS including markdown formatting. Jdcomix (talk) 15:52, 23 September 2026 (UTC)

Just requested G15 deletion of Aadrai Jungle Trail before I saw this. Very clearly AI-generated. ‑‑gurkubondinn 18:35, 23 September 2026 (UTC)

renamed to Dakshv (talk · contribs), they responded w an AI-generated comment, everything reverted/draftified/deleted for now Kowal2701 (talk, contribs) 20:29, 23 September 2026 (UTC)

User:Coe-pjb

Editor has created 9 largish 20k-30k articles from starting on 11:39, July 23, 2026 to 14:26, July 23, 2026. They only have 112 edits, which looks really odd. I'll inform them. scope_creep^Talk 20:01, 23 September 2026 (UTC)

The user explained in July how they created these articles. They admitted using ChatGPT for copyediting, which probably accounted for some cosmetic AISIGNS factors, but it is not a problem to write pages offline, and the articles do not read like LLM output. I reviewed several of these pages during New Page Review and did several source spot checks, and while I did leave a warning on missing citations, I didn't find any of the failed verification issues I'd have expected to with the typical LLM use we see on this page. Not sure this user is an issue for this noticeboard. Pinging @YuniToumei, who left the original question about LLM use, to see if their concerns were satisfied. Dclemens1971 (talk) 20:16, 23 September 2026 (UTC)

I tend to follow the approach of asking and informing about the LLM guidelines in the same query in order not to bite, and they seemed communicative at the time, so I didn't dig deeper. Seems like they stopped editing after July 23 though... And with the stuff Kowal2701 found below it's not looking too great. Would have hoped to see them continue editing without AI. YuniToumei (talk) 07:50, 24 September 2026 (UTC)

not triggered any relevant edit filters, but loads of AI signs in creations, Special:Permalink/1365605075 has red-linked See also entry and red cats, not something that happens if just using AI for copyediting. In Special:Permalink/1365607313 the last ISBN in the Further reading section goes to a completely different book Kowal2701 (talk, contribs) 20:39, 23 September 2026 (UTC)

Ugh. I didn't look at older versions when reviewing. Dclemens1971 (talk) 22:09, 23 September 2026 (UTC)

Wang Yaqiao and The Axe Gang

 Cleanup has been requested.

• Kawakenkuni (talk · contribs · deleted contribs · logs · filter log · block user · block log)

 Kawakenkuni (talk · contribs · logs · block log) (24 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Hi everyone, a couple weeks ago I reported these two articles to the NOR noticeboard, and another user there recommended I bring it up here. You can read my full report there, but I'll summarise it briefly here. Basically, these two articles were expanded by the same user, and I noticed that the citations did not include page numbers, which made it hard to verify. I then looked into the sources and found several don't appear to mention the subject, while the ones that did didn't appear to verify the specific text they're cited to. Worried that it might have been AI generated, I put the articles into Pangram, which gave differing results, saying the Wang Yaqiao article was 100% human written, while the Axe Gang article was 50/50 human/AI written; this is why I didn't immediately bring it here. Could someone else have a look at these two articles? If they're not clearly AI generated, then I'll bounce this back to the NOR noticeboard. --Grnrchst (talk) 09:53, 24 September 2026 (UTC)

Yes, it does read AI generated, quite apparently so. Pangram has a very low false positive rate but fairly high false negative rate. Thanks for bringing it up here. Are there any other contribs by the user we should be paying specific attention to? (would be useful if you could put a userlinks on top of the section) Fermiboson (talk) 11:17, 24 September 2026 (UTC)

@Fermiboson: Ok, I've notified the user. Other than these two articles, it seems their only substantial contributions have been these edits to the Shina (word) article, these edits to the Anti-Chinese sentiment article and these edits to the Singapore–Taiwan relations article. --Grnrchst (talk) 12:02, 24 September 2026 (UTC)

User:Divnanoc

 Divnanoc (talk · contribs · logs · block log) (24 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Created Tatyana Alexandrovna Nozhkina, which I have just nominated for G15 speedy deletion:

references nr. 2, nr. 6, nr 9, nr. 11, and nr. 13 return 404 not found. ref nr. 12 returns a 500 error and does not work. reference nr. 21 returns a 403 error.

Responded to a PROD of the same article with Diff/1376341070, which seems like AITALK and denies LLM use. The first revision Diff/1376133782 suggests that the page was moved to Draft:Vadim Asoskov, which has two references returning 404 errors and seems to have first been moved into article space in Diff/1376133749. Not really sure what was going on there?

Attempted to cite Wikipedia in AbuseLog/45192711 on Byblos (fashion brand) before it was created in Diff/1376218486 with an AIREDCAT. There are UTM parameters showing ChatGPT use in AbuseLog/42228721 and AbuseLog/40449475.

Created Chen Jiagang in Diff/1284749776 (AbuseLog/40474611), reference nr. 11 returns 404 and reference nr. 7 does not work. This was really just a spot check though. ‑‑gurkubondinn 14:39, 24 September 2026 (UTC)

Comments/history at Talk:Rada Krivokapić-Radonjić look like someone using AI to talk and then re-editing the text to sound less AI, article itself is currently A7'd and has red categories. Sarsenet•he/they•(talk) 14:54, 24 September 2026 (UTC)

When something merely looks “like” something, I prefer not to comment. Especially when it comes to hidden flaws, dwelling on which could be considered abusive.

I have explained before. Article was created after the subj was found in Requested Articles, than checked media coverage and found it sufficient to write about. Also the serbian wikipedia article covered the subj. . And serbian article about Rada Krivokapić-Radonjić was started and created by Gorana Gomirac, an exectutive of Wikimedia Serbia (https://sr.wikipedia.org/wiki/%D0%9A%D0%BE%D1%80%D0%B8%D1%81%D0%BD%D0%B8%D0%BA:Gorana_Gomirac_(VMRS)). I`ve came full circle. Divnanoc (talk) 17:02, 24 September 2026 (UTC)

In response of:

(I) Nozhkina:

(i.A) The draft was living in a draft space like within year or so. Some links were broken or migrated since the latest draft edition. Could be fixed easily, but looks that article is currently frozen for any edition. Since you have accused me of using an abstract large language model to write the article (responded on subj page), the very first thing such a model would have done (if it had been used) would have been to check for broken links and replace it. As I can see, that is not the case.

(i.B) Highly possible, it was my fault with Special:MovePage. I simply mixed up the drat article titles when moving them.

(II) Byblos (AbuseLog/45192711)

I have no idea what it means. The only activity I did after creation is linked the page with Prada Group (which was once acquired the trademark in 2001) to avoid an Orphan status for subj.

(III) Davalos (AbuseLog/42228721):

There were 2 (two) UTM-marks in reference links list. Arisen after the usage of "deep search prompt" of subj.`s media coverage in Argentina Check the first edition to verify.

(IV) Birch (AbuseLog/40449475):

I have no idea where and why it came from. Editor rejected to confirm which text was suspicious. Primarily, the article was written by me for Russian Wikipedia than translated and place here.

(B) Chen (AbuseLog/40474611):

Number 7 is working properly. Number 11 last checked almost two years ago, I suppose that relations between artist and gallery could be slightly changed. Divnanoc (talk) 16:41, 24 September 2026 (UTC)

This response is AI generated. We want to talk to a human, not a chatbot. If you continue posting LLM comments you will likely be blocked. Jdcomix (talk) 18:27, 24 September 2026 (UTC)

Ah? Do I need to record own voice or use the obscene wording in at least of 4 languages which I`m speaking more or less, instead of being polite and explaining position in not a "srach style", but step by step algorithmically. It makes me feel that I`m corresponding with LLMs or smh.

NB. It was my handwritten text without any copy. Did you deleted it? No comments at all... Divnanoc (talk) 18:58, 24 September 2026 (UTC)

I really don't think this response is AI. It reads very much like comms of people I know who speak Slavic languages as their primary, down to the grammar: IE "in a draft space like within year or so", "looks that article is currently frozen", "Editor rejected to confirm"

Also, there's a misspelling which AI wouldn't do - "exectutive of Wikimedia Serbia".

No comment or opinion about the rest of it Jishara (please ping upon response) (talk) 20:36, 24 September 2026 (UTC)

Thanks for spotting that: Divnanoc means "Wonderful Night" in ex-Yugoslavian languages. There are "Easter eggs" like this in almost every first edition of my articles here in the English version. Some have already been corrected by other editors, whilst others are still exists. But who actually read this? It’s easier just to destroy down the article, because of 6 archived links out of 23. Not made-up ones, no, but simply ones that were moved to different URLs at the same websites in over a half year period. It’s a malicious LLM that’s messed up. Undoubtedly. Divnanoc (talk) 21:31, 24 September 2026 (UTC)

My analysis of the evidence: Both of the Jiagang links that are down now absolutely did exist - and in fact were both archived by the Wayback Machine on the same day, April 9, 2025. That day also happens to be the day the Jiagang article was created by Divnanoc. See links to the archived version here and here.

Chatgpt for finding sources is, to the best of my knowledge, not forbidden as long as you read the entire source and don't take Chatgpt's word for its existence or statements. And the category is a reasonable one that I was honestly shocked to see doesn't exist - especially as it was added at draft creation, that isn't really strong evidence by any means.

I can't really talk about Nozhkina because it's deleted so i can't see what sources were red flags.

As for the comments on the Rada, it's again got very standard Slavic grammar that doesn't really make much sense as being from AI (in particular "2019? Seriously? 2026 is on the ground", which i'm pretty sure is a literal translation of 2026 год на дворе, which is used to mean nowadays is 2026 but на дворе literally means yard), so I'd buy it as being more likely to be someone thinking through their arguments and grammar-checking before posting, especially when it's something as serious as contesting a proposed page speedy deletion of an article they created, than being AI.

Jishara (please ping upon response) (talk) 00:38, 25 September 2026 (UTC)

I just noticed this discussion and wanted to add a comment.

Apart from the Talk:Rada Krivokapić-Radonjić, which appears to contain possible LLM-assisted text and has been edited several times, I noticed that the arguments against A7 on that talk page resemble the type of pattern described in WP:AIATTR.

As an example, Ref #9 under "Recognition" points to a Radio Kotor URL that currently redirects to the site index and does not show the claimed award.

The article was also created under Rada Krivokapić-Radonjić even though Rada Radonjić was previously salted after WP:A7/WP:G11 deletions. In such a case, WP:REFUND or an WP:AFC review would seem more appropriate than manually recreating the article under a slightly altered name.

Finally, the Serbian Wikipedia article and the Requested Articles listing should be noted. The Requested Articles entry also contains a red link for the subject; the related page history and logs show the previous deletion/creation history. The linked item numbered #383 also appears to return a 404 error. These factors, along with the current content of the article, do not appear sufficient to establish WP:GNG or WP:ARTIST on English Wikipedia. See also WP:OTHERCONTENT. AriuCH (talk) 15:41, 25 September 2026 (UTC)

The Serbian Wikipedia article is titled with the double-barreled last name, and Divnanoc has confirmed that they looked at the Serbian article and Serbian coverage before making this English one, so I think it's more plausible that they kept with the naming of Serbian Wikipedia to create the article under the double barreled last name and didn't notice that there was a salted article at the other name.

Something being insufficient for GNG is also not evidence of AI in any way, a lot of times people write articles and they get deleted because there's not enough proof for notability. Most of those people weren't using AI to write articles, they were just writing articles that turned out to not meet notability.

I'm also not entirely sure what 383, which is for a requested article on the brand, not the person, has to do with Divnanoc's article on the person. Jishara (please ping upon response) (talk) 16:10, 25 September 2026 (UTC)

If you follow the Requested Articles link mentioned above, you will see the subject listed there with the related source entry (#383) next to it. This was only mentioned as part of the overall context; the point is that the Requested Articles listing, the Serbian Wikipedia article, and the current article content do not appear, by themselves, to establish notability on English Wikipedia.

Regarding the talk page comments, I did not claim this alone proves AI use; I only noted that the wording and structure appear similar to the patterns described in WP:AIATTR.

One point I still find unclear is that the editor was aware of the Requested Articles listing, but the previous English Wikipedia title history and deletion/salting logs do not appear to have been checked before creating the article. AriuCH (talk) 16:34, 25 September 2026 (UTC)

I mean, I've personally viewed things similarly with some articles - I see a requested article or a mention of an interesting person, then i look it up, it turns out that's not the usual name, and then I create it at the usual name for the person. I don't tend to also look at the history of the atypical requested name, I think that'd honestly be an unusual thing to do, unless I was also creating a redirect there.

On notability, again I do think it's not unusual by any means for an editor to create an article and for it to then be deleted due to lack of notability, and by no means is that saying that they used AI to make it.

As for the requested articles, for clarity, that request with source 383 is for the subject's eponymous brand. The subject herself is listed much higher, with sources 72-81. The request also doesn't appear to have been listed by Divnanoc, nor were the sources there added by Divnanoc to the best of my knowledge.

I'll pause here to avoid bludgeoning the discussion. Jishara (please ping upon response) (talk) 17:08, 25 September 2026 (UTC)

References

User:Vncdengc

 Vncdengc (talk · contribs · logs · block log) (24 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Cleanup efforts are ongoing and being tracked at /2026-09-25 Vncdengc, 723 pages need review.

Cleanup efforts are ongoing and being tracked at /2026-09-25 Vncdengc (2), 728 pages need review.

Lots of large and rapid edits, talk page full of dab notices. For example Diff/1376453081 is c. 2 kB and was produced in two minutes, Diff/1376433676 is c. 10 kB and was produced in eight minutes, and Diff/1376428243 is c. 5 kB produced in six minutes. Lots of election boxes and tables. There is a lot to go through, not many filter log entries, no markdown. Could use more eyes. ‑‑gurkubondinn 20:13, 24 September 2026 (UTC)

Hello,

I wish to clarify that AI is used exclusively for editing repetitive tasks, not for generating information or any hallucinated content. My use of AI is limited to editing election results and electoral history—tasks that both AI and humans are capable of performing, but which AI can complete within seconds. You are welcome to verify any article; there is no hallucinated information present. AI/LLMs are employed only for editing templates and handling repetitive tasks, which would yield the same result even if performed by a human. Please consider this before issuing any notice. I kindly request you to withdraw the notice. Vncdengc (talk) 05:19, 25 September 2026 (UTC)

but which AI can complete within seconds

See WP:LLM: LLM-generated text often violates several of Wikipedia’s core content policies, and can be produced at a rate that makes review by other editors infeasible (emphasis mine). Northern Moonlight 09:07, 25 September 2026 (UTC)

I wish to clarify that my edits consist of updating constituency results and electoral history of Indian constituencies, sourced directly from official records. The AI/LLM simply reproduces the text exactly as it appears in the official PDF source. As such, I believe there is no possibility of error, and the content is easily verifiable. I have not produced any article containing hallucinated text. My work is strictly confined to this scope. I kindly request you to review my contributions before proceeding with any action. Vncdengc (talk) 09:16, 25 September 2026 (UTC)

The long list of disambiguation-page links you've added to articles, as reported on your talk page, immediately disproves your claim that there is no possibility of error. That point doesn't need to be discussed any more.

Furthermore, it doesn't actually matter whether you believe that or not. WP:NOLLM disallows using AI to generate article contents, including in the way you've been doing it. You need to stop this at once, and commit to not doing it in the future.

Lastly, please don't use an LLM to talk to us here: WP:LLMCOMM. Suðurhafsljósæta (talk) 09:38, 25 September 2026 (UTC)

I have no business to talk about this, and I'm not using any LLM. Kindly mind your language. Please don't jump to conclusions before verifying whether I actually used an LLM for the text Vncdengc (talk) 09:47, 25 September 2026 (UTC)

If you read WP:NOLLM, you will see this line: The use of large language models (LLMs)[a] to generate or rewrite article content is prohibited. As I said above, this includes the way you have been using them for the tables of electoral results. Please confirm whether you will stop using LLMs to generate article content here. Suðurhafsljósæta (talk) 09:58, 25 September 2026 (UTC)

When it comes to reviewing your contributions, I'm also wondering how I (or any other reader unacquainted with the Indian electoral system) could do so, since I don't see any references in the tables you've added. Suðurhafsljósæta (talk) 10:17, 25 September 2026 (UTC)

No, there is absolutely no guarantee that there is no hallucinated information present or that this would yield the same result even if performed by a human. The problem here is exactly that AI can complete [the text] within seconds, which is one of the reasons that the WP:NOLLM guideline disallows the use of AI on the English Wikipedia. Another problem is that this also means that you can't have read your own edits, it is simply not feasible for you to have adequately validated every single number at the rate that you have been submitting edits. Using AI to write articles is actually much slower than doing it yourself, we know this because of the data published by WikiEd. It is also completely immaterial that you believe there is no possibility of error, because there is a very real possibility of a lot of errors and, maybe more importantly, you are now not allowed to use AI on the English Wikipedia.

Writing Wikipedia articles is not an easy task, and it is not supposed to be easy or fast. If you are not willing to put in the work and the effort required to write encyclopaedic entries then you shouldn't be doing it. And I also want to know the same as Suðurhafsljósæta, can you commit to stop using AI on the English Wikipedia? ‑‑gurkubondinn 10:29, 25 September 2026 (UTC)

Please revert any changes you wanted to make and do whatever you feel is right. I was only here to contribute. Do whatever you want don't message me again. I'm done. Vncdengc (talk) 10:39, 25 September 2026 (UTC)

We have agreement from Vncdengc for LLMPRV. I'm trying to find out whether there's a plausible start date for LLM use for a tracker, but not found one so far. Suðurhafsljósæta (talk) 15:43, 25 September 2026 (UTC)

Two tracker pages made, because the AIC article list template didn't want to display correctly with more than about 700 items in it. Is that a known issue? There are 1457 pages in total, excluding changes under +/-35 bytes. Suðurhafsljósæta (talk) 17:50, 25 September 2026 (UTC)

It's a known issue and hard to fix--the template parsing machinery gets overwhelmed. M kuhner (talk) 06:26, 28 September 2026 (UTC)

Thanks! So do I, looking at this list of pages. Suðurhafsljósæta (talk) 06:54, 28 September 2026 (UTC)

I’ll dedicate some time to go through them and get it done - but yeah, when I saw the number I as well was like ugh! Hasekidms (talk) 07:24, 28 September 2026 (UTC)

Hello, please do not revert all articles, some articles are relevant, do check those articles, do not blindly revert all articles. Vncdengc (talk) 12:00, 29 September 2026 (UTC)

Idk, on what basis all articles are reverted. Do look at them, if they are relevant retain them and ask for citation. No one has right to delete all my contributions without any reason. Vncdengc (talk) 12:02, 29 September 2026 (UTC)

On the basis that the English Wikipedia does not accept AI-generated material, and that these are AI-generated articles. The "relevancy" of these articles are immaterial because the WP:NOLLM guideline prohibits AI-generated material. ‑‑gurkubondinn 12:08, 29 September 2026 (UTC)

Not all articles are AI generated. And what basis you are labelling the articles are AI generated ones. Please mention the reason and then proceed with deletion Vncdengc (talk) 01:51, 30 September 2026 (UTC)

Things we look for are listed at WP:AISIGNS. The guideline that allows us to delete further articles without close examination once a pattern of LLM use is established is at WP:LLMPRV. Yes, this sometimes results in deletion of an innocent article; but given that LLM articles can be generated very rapidly, it's necessary to have a rapid process for dealing with them.

I will note that you were specifically told one of the signs we saw, in the initial post in this thread. You really should have noticed your talk page filling up with error reports when you wiki-linked to disambiguation pages. There are forty-one such error reports on your page (by my quick count)--it is difficult to believe in your error free editing process, given that. This is a mistake everyone makes once or twice, but to make it 41 times suggests you are not taking much responsibility for correctness of your work. M kuhner (talk) 02:03, 30 September 2026 (UTC)

~2026-49615-10

~2026-49615-10 (talk · contribs · deleted contribs · nuke contribs · logs · filter log · block user · block log)

Added a chatbot block to Ashok Chavan in Diff/1374883587 (AbuseLog/45121092) and then edit warred to reinstate it in Diff/1376528798 and Diff/1376536271 after being reverted. Has also made other large and rapid edits, a block is probably needed. ‑‑gurkubondinn 20:19, 24 September 2026 (UTC)

User:Hrvatskaeditor

 Hrvatskaeditor (talk · contribs · logs · block log) (24 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Creating several LLM-assisted articles. On Croatian Party of Kosovo: "The European Centre for Minority Issues Kosovo described HSK as the political party representing the interests of Kosovo's Croats and recorded its participation in the 2013 municipal elections". On Draft:Church of St. Catherine, Peja: "The church and its surrounding complex are also the subject of cultural heritage and conservation concerns in Peja".

More context at Wikipedia:WikiProject Articles for creation/Help desk#21:06, 24 September 2026 review of submission by Hrvatskaeditor. Helpful Raccoon (talk) 21:38, 24 September 2026 (UTC)

@Hrvatskaeditor can you explain the reasons why you use LLM on articles? We would like to know the reasons. Galaxybeing (talk) 11:29, 25 September 2026 (UTC)

Hello.

While I might have used an LLM-assisted tool on the Croatian Party of Kosovo and I understand that the overuse of it is not tolerated in the Wikipedia, however, I've already gave out a statement to the Church of St. Catherine, Peja. This sentence is not "vague and meaningless" as how the previous staff members explained it to me.

I use the tool to find more about reliable information and citation, however, most of the time I construct the article by myself as I am against using LLM-assisted tools in such important articles as most of the tools aren't correct.

I do apologize for using any AI-assisting tool and will not happen anymore. I will happily continue to contribute to the Wikipedia by using correct independent sources, as well as constructing and building articles by my own words. I usually work more on Kosovo-related articles and it may be hard to find citations across the internet, so again, I apologize.

Thank you! Hrvatskaeditor (talk) 19:23, 25 September 2026 (UTC)

User:ItsMeGabeProductions

 Cleanup has been requested.

 ItsMeGabeProductions (talk · contribs · logs · block log) (25 September 2026)
Editor has been blocked site-wide for AI/LLM policy violations. 

Discussion split from § Prodijal son. — Newslinger talk 02:12, 25 September 2026 (UTC)

Hi all, on the topic of AI-generated articles related to earthquakes, a recently published article, 1924 Davao earthquake, created ItsMeGabeProductions, is written with a similar format that I suspect is also AI-generated. Like the articles listed here, this one also contains a certain "generic" or redundant language, some attribution to earthquake catalogs, full coordinates in prose, and close paraphrasing.

I have not gone deeper into this yet due to my time constraints so I am speculating based on what I have seen in Prodijal son's articles, but I thought you should look at this as well before reaching out to ItsMeGabeProductions. Dora the Axe-plorer (explore) 16:01, 24 September 2026 (UTC)

Thanks @Dora the Axe-plorer, they were previously reported here (Wikipedia:AI noticeboard/Archive 16#User:ItsMeGabeProductions), I remember being unconvinced at their explanation but unable to prove it at the time. Corruption in Wales (21 Sept) had a red-linked See also entry, overreliance on primary sources, a Pangram positive . They're making 20kb articles almost daily. @Admins willing to patrol AINB: sorry but I think a sitewide block is appropriate here. Kowal2701 (talk, contribs) 20:14, 24 September 2026 (UTC)

Blocked ItsMeGabeProductions sitewide for LLM misuse after previous noticeboard discussion. — Newslinger talk 02:24, 25 September 2026 (UTC)

The 1996 Texel Douglas DC-3 crash seems to be written using AI, so tagging it for that. Galaxybeing (talk) 03:40, 25 September 2026 (UTC)

Stopping by to note that I just found (and blocked) EightyFlint, who is a CU-confirmed sock of this IMGP. G5'd some recent creations, other edits may also need reversion. --Blablubbs (talk) 14:33, 29 September 2026 (UTC)

User:Preferwiki

 Preferwiki (talk · contribs · logs · block log) (25 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

 Cleanup has been requested.

Preferwiki (talk · contribs · deleted contribs · logs · filter log · block user · block log)

First noticeboard entry from the new LLM Paste Check feature, although this appears to have been going on for a while and at high volume. Unfortunately, this includes articles promoted to GA such as the blatantly AI-generated Park Chung-seop Gnomingstuff (talk) 03:55, 25 September 2026 (UTC)

I don’t use AI for Park Chung-seop's page. It's all based on research (every sentence is backed by sources and there is no hallucinations in the sources, you can verify that. While for Joo Ho-jin's page, I did try to run grammar check using a tools. However I already reverted those and do the work manually. Preferwiki (talk) 08:24, 25 September 2026 (UTC)

I am just faithful to the tone of source material (it's formal korean so I tend to use formal English) Preferwiki (talk) 08:45, 25 September 2026 (UTC)

It's not a good idea to lie when there's a tool detecting that you are copying and pasting from an AI chatbot. Gnomingstuff (talk) 14:19, 25 September 2026 (UTC)

I expanded this article from a section from this page List of When Life Gives You Tangerines characters. Its information becomes undue weight if I keep adding on the section of the page. Preferwiki (talk) 19:28, 25 September 2026 (UTC)

So I created draft in sandbox combining 4 paragraph I took from above page and adding new informations, some are from the main drama page and the actor's page. I paste largest amounts of texts from my sandbox to the initial redirect page back in July 7, 2026. As shown in this history page: history. Since then it was incremental editing, mostly for copy editing for prose improvement, adding citations (secondary sources), image formatting etc. Could you help me point out spesific diff that needs clean-up? Preferwiki (talk) 20:38, 25 September 2026 (UTC)

There is a new feature that has detected that you are copying and pasting edits containing metadata from whatever AI chatbot you are using. For some reason, you are trying to lie about this anyway (or deflect with "a tools" -- a hammer? A crowbar? Why are people so allergic to stating what they are using?) Gnomingstuff (talk) 16:22, 26 September 2026 (UTC)

How do we know that feature works? Do we know how it works? Are we making accusations that someone is using AI based on the word of ... AI? --GRuban (talk) 17:14, 26 September 2026 (UTC)

@GRuban you can read about LLM pate-check here. It detects metadata from the most popular LLMs but like most everything else, it is one data point. S0091 (talk) 17:30, 26 September 2026 (UTC)

You can also see the feature writeup here, including how it works. Please take 5 minutes to do some basic research before playing Captain Save-An-AI. Better yet, maybe trust that people who spend a lot of time doing something know what they are talking about. Gnomingstuff (talk) 18:10, 26 September 2026 (UTC)

Thank you for your polite response. I'm not seeing any entries in the history of the Park Chung-seop article marked from that tool. --GRuban (talk) 18:58, 26 September 2026 (UTC)

@GRuban, right, not that article but did get a couple with Joo Ho-jin example. Keep in mind the LLM paste check just went live yesterday. Another thing I see that is concerning is the use of tools to fix things that are LLM signs like MOS:Curly]. Mind you this is a a very brief look so I am not saying anything definitive but take a look a the history of that article. It does raise some questions. S0091 (talk) 19:14, 26 September 2026 (UTC)

They did acknowledge using a tool for grammar checking Joo Ho-Jin which they state they have since reverted Jishara (please ping upon response) (talk) 20:04, 26 September 2026 (UTC)

Yes, "a tool," what a wonderfully specific phrase that is. Gnomingstuff (talk) 22:05, 26 September 2026 (UTC)

Hi. I don't expect that the word tool become an issue. To be very transparent the tool I use in Joo Ho-jin is Google Gemini. The sources I use for this article is quite long so I check google translate just to be sure. It has word 결핍, which I am not sure about the right way to translate it to English (Naver dic said it's deficiencies but I don't like the word). Then I use gemini to smooth the grammar (tempted coz ai mode tab is right up there). I was under the impression that for fixing grammar using gemini is fine. However after getting the notification on my talk page, I revert those right away and proceed to do them manually.

As for using MOS:Curly in pages, it's because many korean resources I use curly. I find it too painful to clean those manually. This is sample of google search of the drama's title (where char Park Chung-seop in) in Jeju dialect sample, they use curly ‘폭싹 속았수다’. This is sample of google search of the drama's title Can This Love Be Translated? (where Joo Ho-jin is from) sample2 Most of korean article I work with also have the same pattern. Preferwiki (talk) 03:50, 27 September 2026 (UTC)

Er-gg1ab 2

Cleanup efforts are ongoing and being tracked at /2026-09-29 Er-gg1ab, 145 pages need review.

LLMPROD is not needed in this case

 Er-gg1ab (talk · contribs · logs · block log) (25 September 2026)
Editor has been blocked from editing the mainspace (articles only) for AI/LLM violations. 

See Wikipedia:AI_noticeboard/Archive_19#User:Er-gg1ab, today they triggered LLM Paste Check with Special:Diff/1376662375. @Admins willing to patrol AINB: I think a sitewide block? Kowal2701 (talk, contribs) 16:12, 25 September 2026 (UTC)

I am curious, is this a filter being triggered? There isn't any tag other than "visual editor" on that edit. The edit summary isn't clearly AI generated, and the prose could have been added in an external editor and then pasted in. ~Anachronist (who / me) (talk) 18:03, 25 September 2026 (UTC)

Sorry @Anachronist, I linked the wrong edit, it's Special:Diff/1376645643. phab:T420258 has more info on LLM Paste Check, it was released yesterday Kowal2701 (talk, contribs) 18:12, 25 September 2026 (UTC)

For now I issued a mainspace block to prevent further AI pollution. It can be expanded to site-wide if needed. ~Anachronist (who / me) (talk) 20:41, 25 September 2026 (UTC)

User:KiaTajik

 KiaTajik (talk · contribs · logs · block log) (25 September 2026)
Editor has been blocked from editing the mainspace (articles only) for AI/LLM violations. 

Was asked and warned about LLM use in § Concerns about LLM use on their talk page, just replied saying they would keep it in mind and then went right back to pasting slop into the project. This is maybe the most obvious on Special:History/Timeline of Tajikistani history, where they pasted c. 200 kB of text. There is markdown in AbuseLog/45181732 and AbuseLog/45215380, and an endless parade of circular refs to Wikipedia in f.ex. AbuseLog/44661693, AbuseLog/44975385, AbuseLog/45013939, and AbuseLog/45021874. Despite only having a bit over 350 edits, this user has managed to generate an enormous amount cleanup work for others. If this doesn't stop now, I think that a block will be in order. ‑‑gurkubondinn 18:00, 25 September 2026 (UTC)

Blocked by Anachronist, thanks. ‑‑gurkubondinn 18:07, 25 September 2026 (UTC)

Looking at Tajikistan, it seems we'd have to revert all the way back to 14 July to get his LLM edits out of there. Masterhatch (talk) 16:26, 26 September 2026 (UTC)

Already done, reverted c. 200 kB in Diff/1376687019 yesterday. The article is back to the 14 July version: Diff/1364113203/1376687019. ‑‑gurkubondinn 16:31, 26 September 2026 (UTC)

User:Acratopotes

 Cleanup has been requested.

This user has uploaded AI "enhanced" images of Sepp Dietrich, Heinz Guderian, Otto Skorzeny, and Hermann Priess, all of which once appeared in each pages' main infobox at one point. (These edits have since been reverted.) Additionally, this user once edited the Blitzkrieg page; an edit that was tagged for possibly AI generated citations.

Furthermore, the user's Wikipedia talk page has a comment discussing the user's "continued usage of LLMs" , while the user's talk page on Commons appears to show a history of uploading copyrighted images to the site, and once uploading an upscaled image of Mohammed Atta. AeroAtmosfear (talk) 19:06, 25 September 2026 (UTC)

User:Huletitaliator

 Huletitaliator (talk · contribs · logs · block log) (25 September 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. 

I think almost every edit by this user is LLM-generated. Personally I noticed this edit and this edit. Almost all are on the WP:CTOP Wikipedia:Contentious topics/Horn of Africa. The user has been warned several times and shows no sign of ceasing the LLM edits. Boud (talk) 20:27, 25 September 2026 (UTC)

New article-space edit by this user that is almost certainly another LLM edit. This edit massively removed warning templates, massively changed reference formatting against the usual style, claimed to have done something which is in fact reasonable but had already mostly been done and any extra editing would have best been done in a small local edit. The massive removal of the warning templates in the Criticism section shows absolutely no understanding of the semantics of the warnings. The user also added a talk page message that was clearly LLM generated: it made a "proposal" that had already de facto been more or less done and accepted prior to the user's new edit, and made a comment about sourcing that didn't make sense in the context of the edit. The user also removed the {{collapse ai top}} template from the talk page. Boud (talk) 18:56, 26 September 2026 (UTC)

Actually, it looks like the talk page message quotes (without quotation marks) the prompts that were input to the LLM. Boud (talk) 19:03, 26 September 2026 (UTC)

User continues to ignore your notice and edited Ethiopian Peoples' Forces Coalition for Survival and its talk page, using LLM. A block may be necessary for the user to start discussing here Elisapoly (talk) 19:30, 26 September 2026 (UTC)

A different user, who created an account on 20 Sep 2026 (the Huletitaliator account was created on 21 Sep), did similar style restructuring to seven articles + created one article in the space of less than two hours, on the same subtopic of the Wikipedia:Contentious topics/Horn of Africa. So far I've only reverted one of these, in which case a major semantic inconsistency (because of WP:BEANS, I'd prefer to avoid stating what it is: it's rather obvious) suggests either an LLM or a group of people working on edits on their private mediawiki instance or wikipedia clone, and then transferring the results here without having one person properly re-read the content first. The style of writing includes quite a bit of Wikipedia-recent-event-copy-paste-news-media-style that tends to be unavoidable until a year or so later when other people decide to clean up by rewriting in encyclopedic style; the information match to the sources seems mostly OK, with small copyright violations (copying a half-dozen words in an arbitrary, non-encyclopedic style from the source), as is common in Wikipedia newsy articles by people eager to get the information into Wikipedia but lacking the time to think about it and summarise in plain English. We can see if this user responds like a human to my comment on the user page, but I thought this was worth mentioning here, just in case there's some relation between the two users. This user is less of a WP:SPA than Huletitaliator. Boud (talk) 08:47, 28 September 2026 (UTC)

I'm starting a separate section #User:LeSablierZeus for LeSablierZeus, because at least one other article is clearly LLM-generated, and better separate the discussions of the two users; the topic is common but the users may be unrelated. Boud (talk) 16:44, 28 September 2026 (UTC)

Hello! I am new to Wikipedia and I am still learning the ins and outs. I apologize for any misconduct or disruption I might've caused. I will go through more of the help articles so I can work on my editing the right way. Once again, sorry. ~~~~ LeSablierZeus (talk) 16:50, 28 September 2026 (UTC)

It's pretty simple. Do not use LLM/AI when editing wikipedia. And when it comes to translation, the rules are very strict. Masterhatch (talk) 18:02, 28 September 2026 (UTC)

User:MargaritaPLV

• MargaritaPLV (talk · contribs · deleted contribs · logs · filter log · block user · block log)

New editor is rapidly expanding has creating articles, often adding thousands of bytes of text within seconds across multiple edits (most egregiously here). Replies to concerns raised on their talk page regarding AI usage last month were met with verbiage consistent with an LLM, as were concerns raised today. Some of their work looks to be machine-generated translations of other Wikis' articles. Those are not a problem and look to be straightforwardly in alignment with policy. The other stuff is not so good. Best, ~ Pbritti (talk) 05:10, 26 September 2026 (UTC)

Thanks for bringing attention to the style of my contributions (really!) I don't want to go into long debates, I will just repeat everything I said on my talk page.

I do work offline, never edit in the wiki browser and when the work is done I prepare it in bundles (if needed) to make the edit history more transparent and then copy-paste it to wiki, always after reviewing. All my contributions, etc. are checked for notability and other wiki rules before I even start writing.

Over the last several months Wikipedia editing has become my main interest and since I quit my job, I have spent most of my time gathering information, preparing, writing and polishing articles.

I never really thought about how my edits might look from the perspective you are now explaining. I agree that they could be interpreted that way as you mentioned.

Don't get me wrong... I do not argue with you because I totally understand that my use of polishing my writing could be considered non-reviewed AI writing, which is prohibited.

Therefore as I said before, I will follow your advice! Thank you. I agree that fewer contributions with higher quality are better than more contributions that raise questions from other Wiki contributors. Once again, thank you! MargaritaPLV (talk) 05:24, 26 September 2026 (UTC)

I will also keep in drafts all the articles that you reverted/moved there and only move them back to mainspace through AfC to make sure they meet all the criteria. MargaritaPLV (talk) 05:42, 26 September 2026 (UTC)

By submitting via AfC I mean doing so after reviewing, improving, correcting. MargaritaPLV (talk) 06:33, 26 September 2026 (UTC)

Since the biggest call-down about my contribution came because of timestamp "issues" and publishing speed, I believe it would be worth commenting on my workflow in a bit more detail. I know a detailed answer may be considered an AI answer, but still, what else do I have left ...?

Just to make clear how my workflow / routine looks:

• I find a topic that is interesting to me.
• I gather information about the topic and sources, obviously with Google, which is AI, and with other AI tools.
• I go through the sources and references and pick the ones that are appropriate and relevant to wiki guidelines.
• I make a list of references for myself.
• I go through all the information and start writing (not in the wiki browser, offline on my PC).
• When the article is done, I run grammar correction without rephrasing, so the process of polishing does not change my own words.
• After that I add Wikipedia formatting to the text with external tools and turn it into code.
• When a rewrite/edit/article is done and is ready for publishing, I split it into parts/edit batches to make the history of my edits transparent for other editors and easy to revert, improve, e.t.c. Within this process, I prepare the text of summaries of each edit/batch (and you are right here, this part was done by AI and probably was my mistake. Most of my edit summaries were made by AI when I was splitting edits into batches, because I did not consider that as part of publication or communication. According to my understanding, this is a technical note).

Then, as a result, I have a very efficient document that contains:

1) Writing created by me, formatted, reviewed and ready to publish as text/code.

2) Summary of each edit/bundle of edits.

Both are ready to copy-paste into Wikipedia.

After all that work is done, I implement my contribution into Wikipedia. I simply go to the wiki article, open the editor and copy-paste the whole work. Sometimes, as you mentioned, some edits took 1 minute, and within this minute you can see the whole article appear. I think no chatbot can prepare a well-structured article with references, notes and summaries that fast. You need some sort of super AI with a direct connection to wiki for that.

Also, it was mentioned that "whole articles come out of the blue" - when you don't see the process behind the scenes, it might look so, and as I already said, I never thought about how my work would look from the side. I was focused only on the topics themselves.

Judging the speed and amount of publications is also very subjective. Wikipedia became my main interest over the last several months since I quit my job. It helped me to develop in another way, and I have to admit I spent most of my free time researching, reading, preparing articles and edits, going from one topic to another. But I don't have a journal of hours spent to prove that.

In the end, I would like to say that I am not looking to argue about any judgment made about me, but I am willing to give my point of view and explanation in case somebody really cares. I am ready to adapt, and if any clear mistakes are made by me, to learn from them.

From now on, I will follow the advice and will limit AI use only to grammar checks and try to follow wiki guidelines as strictly as possible. I strongly agree that less contribution of higher quality is better than a flood of low-quality content. MargaritaPLV (talk) 08:17, 26 September 2026 (UTC)

Thanks @MargaritaPLV, it's interesting to get this insight into your ways of working, and an acknowlegment that youhave used AI in the past. What would you suggest about your previous editing contributions? Are you OK to go through them and remove AI assisted content? Thank you again for engaging openly, it saves all of us a huge amount of work. ChrysGalley (talk) 11:01, 26 September 2026 (UTC)

I am absolutely okay with removing anything if there is such a need, since I have no connection to any of the subjects, nor any other COI related to what I wrote about. Colleague Pbritti has already reverted many of my edits and returned articles to drafts. As I already said, I will go through these articles, review them, try to make them better, and then only submit them through AfC in order to have a more independent look at them, rather than return the articles to mainspace myself. I will not use AI where I used it - creating summaries of my edits. MargaritaPLV (talk) 11:13, 26 September 2026 (UTC)

Using AFC is a really good idea in the circumstances. I am more active as an AFC reviewer than here and anything blantantly AI will get picked up there (hopefully). Thank you for helping here. ChrysGalley (talk) 11:41, 26 September 2026 (UTC)

Will do like that, thanks. MargaritaPLV (talk) 13:35, 26 September 2026 (UTC)

@MargaritaPLV: I appreciate that you seem to be trying to adhere to the AI policy with your most recent edits since this report. Your replies above suggest that you may have been using AI to communicate with your fellow editors. If that is the case, please do not do so. I will take some time to review the draft you just submitted at AfC. Thank you very much for following through on your commitment to use that process for the moment. Best, ~ Pbritti (talk) 21:00, 26 September 2026 (UTC)

Hm, we are all human beings, we all make mistakes and have the right to have a subjective opinion. If communication above feels like AI for someone in this thread I will develop my communication style accordingly and try to keep it enriched with non-AI terminology, punctuation and vocabulary (on purpose). Thanks for reviewing AfC draft I submitted yesterday. Working on next version now. Will transfer this approach into my article writing style. MargaritaPLV (talk) 03:17, 27 September 2026 (UTC)

but not limiting with only style change, working with sources too... MargaritaPLV (talk) 03:47, 27 September 2026 (UTC)

User:GRohan2002

• GRohan2002 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

This user made some rather large edits to a diverse group of articles on 4th September at a rapid rate. I asked them if they used AI in their edits, and they admitted to using Grammarly for grammar and wording suggestions but denied using other LLMs. I still suspect that they used an LLM to larger extend, for following reasons:

• AIATTR at Police bus (Independent coverage documents several recurring roles––) and Acoustic Control Corporation (Independent accounts name Jaco Pastorius, John Paul Jones, John McVie and Larry Graham among its users and Independent coverage of later Acoustic products treats them as revivals of the name rather than as a documented continuation of the original Los Angeles manufacturer)
• AIBOLD at Fail-stop
• AIREDCAT at Super 8 Stories, Acoustic Control Corporation and Brooklyn Field Club.

Their edit summaries also look suspicious to me. I hope Wikipedians with more expertise in identifying LLM-generated edits could take a look. Valtaisa varpunen (talk) 07:31, 26 September 2026 (UTC)

Just to add some context on the points raised:

The red-linked categories are a genuine mistake on my part. I had misunderstood how Wikipedia categories work and thought that adding a category would create the category page if it did not already exist. I now understand that categories need to exist (or be created separately) and will check them before adding them in future. This was an ordinary misunderstanding of Wikipedia's category system, rather than something generated by an LLM.

As for the bolding, wording, and the phrasing identified as possible AI indicators, those were simply choices I made while writing and editing. I don't recognize them as having come from an LLM.

The apparent rapid burst of editing on 4 September also has a straightforward explanation. I had been working on these articles locally for some time, with the source text saved on my computer. I only uploaded the completed edits once I was finished with them, which is why several substantial edits appear in the history in relatively quick succession. I have not made another comparable burst since then because I am currently working on several more articles which I have been preparing in the same way and intend to upload once they are ready.

To reiterate, I wrote the substantive article content myself from the sources, and did not use an LLM to generate or rewrite it. Grammarly was used for proofreading and may have suggested grammatical or wording corrections, which I reviewed myself.

I'm happy for the original revisions to be examined and for other editors to comment on them. GRohan2002 (talk) 07:50, 26 September 2026 (UTC)

And Grammarly is AI. So no, you can't use that going forward @GRohan2002. No need - there is zero requirement for perfect English in editing, but there is a total block (with 2 narrow exceptions) on the use of AI. ChrysGalley (talk) 11:04, 26 September 2026 (UTC)

Understood, thanks for clarifying. My (perhaps false) understanding was that using Grammarly for minor copyedits to text that I had written myself would fall under the first exception, provided I reviewed the suggested changes and they did not introduce new content.

That being said, if using Grammarly could lead to complications or uncertainty, I'm more than happy to simply avoid using the tool altogether for all Wikipedia contributions going forward. GRohan2002 (talk) 12:14, 26 September 2026 (UTC)

Yes, steer well clear of Grammarly. It really is unnecessary. I don't know where this crazy (and impossible) notion of perfect English in Wikipedia comes from, is not in policy or guidelines, just a requirement to be clear. English is very diverse anyway. Other Wikipedias have different approaches, so Latin Wikipedia demands a high level of accuracy, for example. Minor problems can be resolved by the expert copyedit editors far faster than a machine anyway. But all LLM is banned except for the 2 narrow exceptions. The number of people in this noticeboard who vociferously deny they use AI but who turn out to be Grammarly users is quite something. ChrysGalley (talk) 12:24, 26 September 2026 (UTC)

Well said. There are often multiple ways of saying the same thing in English that just come down to personal preference. LLM's don't seem to agree Edlectro28 14:30, 26 September 2026 (UTC)

Grammarly is also notably dishonest about how they operate (e.g., claiming that dozens of public figures are providing "expert review" who weren't even told about this). Gnomingstuff (talk) 18:14, 26 September 2026 (UTC)

Fair enough. I understand the concern, and I'll stop using Grammarly for my Wikipedia editing going forward. Thanks for explaining.

Regarding the concern raised by Valtaisa varpunen, I've already corrected the red-linked category issues I identified. I still think the articles themselves are strong and well-sourced, and I don't currently see any other substantive issues that need correcting. If the noticeboard discussion identifies anything that should be addressed, I'll be happy to take a look. GRohan2002 (talk) 18:56, 26 September 2026 (UTC)

Taking a look at GRohan2002's edits on 4 September:

• 05:42 UTC: Diff/1373132120 (c. 1,5 kB) to First Amendment audit
• 06:32 UTC: Diff/1373139203 (c. 16 kB) to Brooklyn Field Club, added an AIREDCAT
• 06:42 UTC: Diff/1373140207 (c. 11 kB) to Seager, New York, an edit that was meant for the Firematic Racing article and self-reverted in Diff/1373140303 a minute later. This makes it
• 06:43 UTC: Diff/1373140448 (c. 1,5 kB) to Firematic Racing, this is the same edit as was pasted into the Seager, New York article in Diff/1373140207 one minute earlier
• 06:54 UTC: Diff/1373141581 (c. 9 kB) to Seager, New York
• 07:03 UTC: Diff/1373142607 (c. 4 kB) to 1930 World Figure Skating Championships
• 07:09 UTC: Diff/1373143197 (c. 5 kB) to Hamitic League of the World
• 07:18 UTC: Diff/1373144185 (c. 5 kB) to Acoustic Control Corporation, added 2x AIREDCATS
• 07:25 UTC: Diff/1373144998 (c. 9 kB) to Super 8 Stories, added an AIREDCAT that was removed three minutes later in Diff/1373145772
• 07:39 UTC: Diff/1373146463 (c. 7 kB) to Fail-stop, added a reference with the doi:10.1007/3-540-68647-7 that doesn't exist
• 07:55 UTC: Diff/1373148039 (c. 9 kB) to Police bus

Of course it is plausible that these prepared offline, there is WP:NOPROOFOFAI and no way to know with absolute certainty. ‑‑gurkubondinn 15:12, 26 September 2026 (UTC)

User:Judy6341

Judy6341 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Draft:Danielle, Asperger was declined due to LLM and Draft:Elena Armas was moved to draft due to LLM. Came across them via LLM paste-check (example) but there are others. Account created on 9 September 2024 with 127 edits, likely most being LLM with some sources in Spanish or Portuguese. They have created three articles with the most obvious being Selma Sueli Silva. I have left them a level 3 warning. S0091 (talk) 16:04, 26 September 2026 (UTC)

User:JMvanDijk

 JMvanDijk (talk · contribs · logs · block log) (26 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Submitted a series of edits with chatbot communication and markdown to User:JMvanDijk/Sandbox 9/Box 42/Box 12 (meant for History of Sparta).

Chatbot communication diffs

1.  Below is a **Wikipedia-ready expansion** that **adds chronological demographic charts (tables)** tracing the **Spartan state and army from the  Early Archaic through the Late Classical period**, integrated cleanly with the earlier section.
   
    I’ve kept the tone conservative, flagged uncertainty where appropriate, and used **standard, reliable scholarly sources** with **inline `<ref>` citations**.
   
    You can paste this directly under *Society and demographics* or as a subsection such as *Demographic development of the Spartan state*.

    ---
   
    ### Early Archaic Sparta (c. 800–650 BC)
   
    *Before full militarization*
   
    | Category             | Approximate numbers |
    | -------------------- | ------------------- |
    | **Spartiates**       | ~3,000–5,000        |
    | **Perioikoi**        | ~2,000–4,000        |
    | **Allies**           | Minimal / ad hoc    |
    | **Total field army** | ~5,000–9,000        |
    
    ---

    ---
   
    ### Summary trend (chart interpretation)
   
    Taken together, these charts illustrate a **classic demographic boom-and-bust cycle**:
   
    * **Growth** through conquest and militarization (Archaic period)
    * **Peak** citizen manpower around 500–480 BC
    * **Long contraction** driven by exclusivity, inequality, and attrition
    * **Terminal weakness** by the mid-fourth century BC
   

   
   — Diff/1333098104

2. Below is a **Wikipedia-ready rewrite** with:
   
    * **Inline citation-formatted footnotes** using `<ref>…</ref>`
   
    * An expanded **comparison table including Sparta**
   
    * Neutral encyclopedic tone consistent with Wikipedia style
   
    You can paste this directly into a draft article or sandbox.
   
   ---

    ---
   
    ### Comparative institutional and demographic resilience
   
    | Feature                            | Athens                        | Sparta                             | Macedon                      | Rome                              |
    | ---------------------------------- | ----------------------------- | ---------------------------------- | ---------------------------- | --------------------------------- |
    | **Scalable manpower**              | ❌ Limited to male citizens    | ❌ Restricted to Spartiates         | ⚠️ Expanded through subjects | ✅ Citizenship and allies expanded |
    | **Loss tolerance**                 | ❌ Major defeats destabilizing | ❌ Severe due to small citizen body | ⚠️ Moderate                  | ✅ High                            |
    | **Political stability under loss** | ❌ Factionalism common         | ❌ Citizen losses irreversible      | ⚠️ Dependent on monarchy     | ✅ Institutional continuity        |
    | **Institutional learning**         | ⚠️ Tactical, limited reform   | ❌ Highly conservative              | ⚠️ Leader-driven             | ✅ Systemic and cumulative         |
    | **Identity compatibility**         | ❌ Narrow civic identity       | ❌ Extremely exclusive              | ⚠️ Partial integration       | ✅ Broad and extensible            |
   
    ---
   

   
   — Diff/1332933821

3.  Yes — here is the **full package**, exactly as requested:
   
    ✅ Inline citations per row
    ✅ Harmonized terminology with existing Wikipedia articles (*Roman legion*, *Roman army*, *Macedonian army*, *Hellenistic armies*)
    ✅ Methodological uncertainty note
    ✅ Additional **comparison table** (Rome vs Macedon vs Hellenistic kingdoms)
    ✅ Encyclopedic tone and WP-compatible formatting
   
    ---
   
    ## Main Wikipedia table (with citations + methodology note)
   
    ```wiki

    ```
   
    ---
   
    # Comparative Wikipedia table
   
    ## Rome vs Macedon vs Hellenistic Kingdoms
   
    ```wiki

    ```
   
    ---
   
    # Transition paragraph (encyclopedic, article-ready)
   
    > The Roman military system differed structurally from both the Macedonian and Hellenistic models in its capacity for demographic scaling and institutional continuity. Whereas Macedonian power relied on royal authority and a finite peasant recruitment base, and Hellenistic kingdoms depended on fragmented combinations of settlers, mercenaries, and subject populations, Rome embedded expansion directly into its political structure through citizenship extension, allied incorporation, and administrative integration. This produced a system in which military defeat did not lead to demographic collapse but instead triggered institutional replenishment, allowing Roman power to compound over time rather than exhaust itself through warfare.
   
    ---
   
    ## Why this works on Wikipedia
   
    ✔ No original synthesis claims
    ✔ All assertions anchored to classical + modern scholarship
    ✔ Uses accepted framing from Polybius, Scheidel, Goldsworthy, Worthington
    ✔ Matches existing article language and structure
    ✔ Safe under **WP:HISTRS**, **WP:NOR**, **WP:VERIFY**, **WP:NUM**
   
    ---
   
    If you'd like, the next logical additions would be:
   
    📈 A **Commons SVG graph**:
   
    * Roman manpower curve
    * Macedonian manpower plateau
    * Hellenistic fragmentation
    * Spartan collapse curve
      (all in one comparative chart)
   
    📄 A structured section titled:
   
    > **"Demographic Scalability and State Survival in Antiquity"**
   
    Yes — here is the **full package**, exactly as requested:
   
    ✅ Inline citations per row
    ✅ Harmonized terminology with existing Wikipedia articles (*Roman legion*, *Roman army*, *Macedonian army*, *Hellenistic armies*)
    ✅ Methodological uncertainty note
    ✅ Additional **comparison table** (Rome vs Macedon vs Hellenistic kingdoms)
    ✅ Encyclopedic tone and WP-compatible formatting
   
    ---
   
    ## Main Wikipedia table (with citations + methodology note)
   
    ```wiki

    ```
   
    ---
   
    # Comparative Wikipedia table
   
    ## Rome vs Macedon vs Hellenistic Kingdoms
   
    ```wiki

   
   — Diff/1334315673

This user has over 10k contributions, but has only created a few articles since free public access to chatbots. There are circular references to Wikipedia and a level 1 heading in AbuseLog/42489882 and an oddly placed nowiki tag in AbuseLog/43133587, which you wouldn't expect to see from an experienced editor. ‑‑gurkubondinn 17:26, 26 September 2026 (UTC)

More chatbot communication in HTML comments in Diff/1371787791, Diff/1371787544 and Diff/1371821048. ‑‑gurkubondinn 18:29, 26 September 2026 (UTC)

User:EdwinAlden.1995

 EdwinAlden.1995 (talk · contribs · logs · block log) (26 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Triggered LLM paste check with Diff/1376855106 when creating Kundé (music awards). User has autopatrolled, has made 13k edits and created 152 articles.

Has created some article with UTM parameters showing ChatGPT use:

• Diff/1370130989 (AbuseLog/44912477): created Willy Mbembe, has some mild AIATTR as well
• Diff/1371155797 (AbuseLog/44956068): created Online Information Media of the Democratic Republic of the Congo, also has some AIATTR
• Diff/1372200407 (AbuseLog/45002228): created Des mythos
• Diff/1376307753 (AbuseLog/45197507): created Johnny Moko, draftified by Vegantics in Diff/1376372113 as being machine generated

There are more UTM parameters from ChatGPT in Diff/1370131909, Diff/1370458764. Cited a ChatGPT conversation in Diff/1305270028 (AbuseLog/41893768) before being granted autopatrolled. ‑‑gurkubondinn 19:35, 26 September 2026 (UTC)

Was warned about § AI on their talk page by Gnomingstuff in April 2026, but didnt respond and removed it from their talk page in Diff/1351150707 instead. Was warned again in § Your edits to Luindi Chiefdom in May 2026 and thanked that user for their help in spotting and removing the AI-generated content and said they would make the necessary revisions. Was brought to ANI over § Repeated and misleading edit summaries in 2023, where they seem to have used ChatGPT to respond.

@Admins willing to patrol AINB: could you please revoke autopatrolled? ‑‑gurkubondinn 19:52, 26 September 2026 (UTC)

Triggered LLM paste check again in Diff/1376880458 to the same article. Courtesy ping to CaptainEek as the admin granting autopatrolled. ‑‑gurkubondinn 20:41, 26 September 2026 (UTC)

Disappointing. I have revoked autopatrolled. CaptainEek ^{Edits Ho Cap'n!}⚓ 21:07, 26 September 2026 (UTC)

Good morning. It is morning here in Texas, and I am sorry it took me a while to respond. I was editing the Kundé article when I was notified about the concerns over my AI use. I was beat when I finished and fell asleep. I planned to reply this morning, and here I am.

I want to be honest. I relied on AI quite a bit for some articles between roughly 2022 and the end of 2025. I can see the AI wording and poor word choices in the examples you showed me, some bad word choices, and a few clunky edit summaries. I only began reviewing my articles near the end of 2025, but I was still using it for things like tables, section headings, and source searches, so the same concerns might apply to Kundé and other articles I worked on in 2026.

But I have also written articles using sources I found myself, including some I paid to access through libraries. I have also worked on articles without AI-written content. I know that does not let me off the hook for my other edits, but I hope it shows I can do the work myself. I want to stay part of the Wikipedia community and fix the issues people have pointed out. EdwinAlden.1995 (talk) 13:53, 27 September 2026 (UTC)

Thank you for being honest with us. ‑‑gurkubondinn 14:15, 27 September 2026 (UTC)

Just FTR, the main problems with AI-generated content are WP:V failures (i.e. hallucinations), WP:COPYVIO, and WP:NPOV/WP:WTW Kowal2701 (talk, contribs) 15:08, 27 September 2026 (UTC)

Thanks for the links showing the AI-related problems. They answer the question I asked @Gurkubondinn about the Kundé article. I will review those issues now and start working on the article soon. EdwinAlden.1995 (talk) 15:15, 27 September 2026 (UTC)

FWIW, reviewing AI-generated content and trying to preserve it while making it policy-compliant is a lot harder and time-consuming than it may seem, especially at this sort of scale. The problems range from being blatant, to subtle source-text discrepancies. We usually recommend that people rewrite or stubbify AI-generated content Kowal2701 (talk, contribs) 15:23, 27 September 2026 (UTC)

Thank you for explaining what I should look out for. My Wikimedia Commons account was banned after I failed to check my notifications. I don't want to repeat that mistake on English Wikipedia, so I will pay close attention to notifications, rules, and warnings. EdwinAlden.1995 (talk) 15:31, 27 September 2026 (UTC)

What you are also missing is that AI-generated material is prohibited by the WP:NOLLM guideline and not accepted on the English Wikipedia. ‑‑gurkubondinn 15:28, 27 September 2026 (UTC)

Asking WP:WHERESTHEAI in Talk:Kundé (music awards) § AI-generated. ‑‑gurkubondinn 14:51, 27 September 2026 (UTC)

I was not aware of that, and may not have worded my question clearly. What I meant to ask is why you believe the "whole article" needs to be rewritten and could be deleted under LLMPROD. Which requirement for presumptive removal do you think applies to the article? I know you do not have to review all 35 pages for me, but I am just asking for a few examples so it would give me a starting point for checking the article without deleting text I wrote and checked myself. EdwinAlden.1995 (talk) 15:07, 27 September 2026 (UTC)

The editing pattern on History/Timeline of the M23 campaign kind of speaks for itself, lots of large and somewhat rapid edits. ‑‑gurkubondinn 20:14, 26 September 2026 (UTC)

User:UzbukUdash

UzbukUdash (talk · contribs · deleted contribs · logs · filter log · block user · block log)

High-volume rapidfire AI article creations for some time, some edits example containing Markdown among other AI indicators. Warned for LLM use on their talk page, completely ignored it, per filter log spends their time silently removing AI tags. Gnomingstuff (talk) 21:32, 26 September 2026 (UTC)

Fazle Hasan Shishir (July 2024) had OAICITE and loads of markdown. The prior creation was in Sept 2023 (Special:Permalink/1176083532) and had hyphens instead of dashes (i.e. human-written). If only they were all that easy. Warnings in July and August 2026 , adding lots of deprecated params, warned for copyvio in Dec 2024 , COI in Jan 2025 (i.e. promotional writing) . Last creation was Counter Foto – A Center for Visual Arts (23 August, after the warnings) which had AIREDCAT, AIATTR, AILEGACY at the end (btw, ?? The institution has been described by the Sigrid Rausing Trust as "an international photography school founded in Bangladesh in 2012" that brings together practitioners from diverse fields including sociology, filmmaking, economics, anthropology, and academic research.). I've given a final warning (btw they've started NPPing, idk if something should be done about that) Kowal2701 (talk, contribs) 22:05, 26 September 2026 (UTC)

User:Arts culture editor

 Arts culture editor (talk · contribs · logs · block log) (26 September 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. 

Warned in § September 2026 because of markdown in Diff/1376739543 and a chatbot block in Diff/1376740475. Added another chatbot block in Diff/1376744072 before giving an evasive non-answer saying that they wrote this text and was very careful to ensure it strictly complied with Wikipedia’s guidelines and then pasting communication from a chatbot in Diff/1376754562 an hour later. Added a reference in Diff/1376778431 that returned a 404 not found error, then fixing the URL in Diff/1376845292 later. ‑‑gurkubondinn 23:03, 26 September 2026 (UTC)

Again, I am very sorry. I will not amend that page again as requested.

I was excited to make a contribution after reading the founder’s book and am learning.

Take care. Arts culture editor (talk) 23:12, 26 September 2026 (UTC)

See the User talk:Arts culture editor § September 2026 and User talk:gurkubondinn § Douglas Grant conversations, Arts culture editor is here in good faith and is open to feedback, but I would like to see a commitment to cease all use of AI on Wikipedia before this is closed. ‑‑gurkubondinn 12:34, 27 September 2026 (UTC)

Yes, I will not use AI to write or generate text as per Wikipedia:Writing articles with large language models.

Again, I apologise. I am new to Wiki and learning. Arts culture editor (talk) 07:48, 29 September 2026 (UTC)

I’m not sure if this is the place to ask, but it would help this newbie enormously if the app had more visual formatting tools so you don’t have to write everything in code. The Wiki learning curve is steep. Thanks again for your patience. Arts culture editor (talk) 07:57, 29 September 2026 (UTC)

If you're editing with one of the mobile apps, that explains some of the challenges. The apps were designed with reading in mind, and editing features have always lagged behind those on the web interface. I suspect the majority of regular editors use the Wikipedia website, not an app, even those who primarily edit on phones. If you go to the website on a phone, you will be shown a mobile-friendly skin that is similar to the app interface, but with more editor-friendly features. There is work being done to improve the editing experience in the apps and to make it easy to continue editing on the website, which has a visual (WYSIWYG) interface that might be friendlier for newcomers. The 21 September update to the apps includes an option to edit with the visual editor on the website, which is explained at mw:Wikimedia Apps/Team/Future of Editing on the Mobile Apps#Handoff to Mobile Web. —ClaudineChionh (she/her · talk · email) 09:24, 29 September 2026 (UTC)

Thank you. I will try the website.

I’ve just had an 19 edits rejected. I don’t understand why becsuse everything was factual and referenced, and facts can be found in the ref for each para. I didn’t use AI either. It’s a bit disheartening trying to do it on my phone (as I don’t have a computer) and then have everything reverted. Arts culture editor (talk) 09:29, 29 September 2026 (UTC)

@ClaudineChionh: for context, there is also more about ACE's mobile app use woes in the § Douglas Grant thread on my talk page. ‑‑gurkubondinn 09:29, 29 September 2026 (UTC)

Thank you. I kindly ask where I should go for help with the 19 edits that were reverted. All of them had references and the facts can easily be found in those references. Arts culture editor (talk) 10:33, 29 September 2026 (UTC)

In mobile view, the code editor colour codes the input so you know if it’s correct or not.

On the website, when in mobile view, the link function is the same as the app (basic) but it doesn’t colour code the input.

On the website in desktop view, trxt text is way to small. Even if I adjust the text on the right hand side and change to dark. It doesn’t align with the accessibility setting in my phone.

I hope this feedback may help improve the editing functionality. Arts culture editor (talk) 22:06, 29 September 2026 (UTC)

User:Tpbradbury

Tpbradbury (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Another long-term case that involves many high-profile articles, I feel bad even bringing this here because of how much of a mess it is going to be but I was told to not let said long-term cases just slip off the radar. Their contribs are a huge pain in the ass to go through for many reasons but a few to single out:

• ChatGPT params in Special:Diff/1296836216, AI is unambiguously in the mix here.
• 2024-characteristic AI text Special:Diff/1213530850
• 2025-characteristic AI text Special:Diff/1321580094
• Not a lot of recent article expansions, but Special:Diff/1375032140 seems suspect (e.g., has the emerging 2026 variant of didactic disclaimer at the end; normally I do not use detectors but Pangram suggests it is AI as well)

Gnomingstuff (talk) 00:24, 27 September 2026 (UTC)

Have you got any examples of misuse, rather than use. One of the reasons it's a pain in the ass to go through my contribs is that I've made over 88,000 edits since 2001, and many people have come to my talkpage to see they've actually been useful. I didn't register all of them back in the day, Tom B (talk) 22:39, 27 September 2026 (UTC)

AI-generated content is not allowed on Wikipedia, so every example of AI-generated content is misuse. If you've been here since 2001 then you have no excuse for not wanting to follow guidelines. Gnomingstuff (talk) 00:39, 28 September 2026 (UTC)

The guideline was instituted in March 2026. You say that 'every example of AI-generated content is misuse', i.e. every human-reviewed example before March 2026, is misuse. Please point to the part of the guidelines that say this. Also, please point to the guidelines that say it is ok for you to accuse editors of not wanting to follow guidelines based on no evidence. For you to make this accusation you must have evidence that I knew about the March 2026 decision, to ban even human-reviewed AI to help expand the pedia like i've been doing for 24 years, and then wanted to not follow it, Tom B (talk) 20:32, 29 September 2026 (UTC)

User:Wikisadia16

Wikisadia16 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Admitted NOLLM violating use in Special:Diff/1377001723/1377010680. 2k+ edits require review. Fermiboson (talk) 12:28, 27 September 2026 (UTC)

chatgpt-derived links in June Kowal2701 (talk, contribs) 16:08, 27 September 2026 (UTC)

Why is using AI to rephrase poor grammar such an issue?

Correcting grammar through AI can change meanings, and add information that wasn’t there. In this case, I think the edits made with AI should be looked into rather than it was only correcting grammar. Hasekidms (talk) 20:17, 27 September 2026 (UTC)

Okay so I assume that I'm being reported because there is misinformation on the article due to my use of AI, but no one has pinpointed / addressed where they are. So where is the exact issue? Wikisadia16 (talk) 21:41, 27 September 2026 (UTC)

The exact issue is that AI-generated content is not allowed on Wikipedia. Why is this so hard for people to grasp? Gnomingstuff (talk) 22:10, 27 September 2026 (UTC)

Wikipedia:LLM states “Editors are permitted to use LLMs to suggest corrections to their own writing, and to incorporate them after human review. This is limited to spelling, punctuation, capitalisation, grammar, and other simple mistakes. Editors must ensure that any text copyedited in this manner is still supported by the sources cited.” which is exactly what I did, I used it to fix spelling grammar etc, and made sure the result I was given was still the correct information. Wikisadia16 (talk) 06:48, 28 September 2026 (UTC)

User:Third003

 Cleanup has been requested.

Third003 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Hi, this user has been really straightforward and admitted to using LLM in creating large numbers of articles on Indian villages. They also use it on their talkpage. I think there's a competence issue in terms of understanding WP:GNG and WP:LLM, but there is no intention to deceive. I'm going to draftify their articles that are still on NPP right now, but I would appreciate help looking at the rest of their output. Boynamedsue (talk) 15:52, 27 September 2026 (UTC)

This LLM-generated text has been collapsed and should be excluded from assessments of consensus. ‑‑gurkubondinn 16:18, 27 September 2026 (UTC)

The following discussion has been closed. Please do not modify it.

Thank you for raising these concerns. I understand that I need to improve my understanding of WP:GNG, WP:NPLACE, and WP:LLM before continuing to create articles at the same rate.

I acknowledge that I have used LLM assistance in my editing, and I understand that this requires particular care under Wikipedia's policies. I will slow down and stop creating new village articles for the time being while I review the relevant policies and my existing contributions.

I am also willing to cooperate with the cleanup and review of my previous articles. If there are particular articles or recurring problems that you think I should focus on first, I would appreciate any guidance.

Third003 (talk) 15:56, 27 September 2026 (UTC)

@Third003, please read WP:AITALK, we want to hear what you have to say, not your chatbot Kowal2701 (talk, contribs) 16:00, 27 September 2026 (UTC)

they had an article draftified for LLM use in August and removed the notice Special:Diff/1371001167 Kowal2701 (talk, contribs) 15:59, 27 September 2026 (UTC)

Saw AbuseLog/45232940 earlier and requested G15 for Sani, Ladakh. Third003 removed the CSD and {{ai-generated}} templates in AbuseLog/45234263 a few minutes later, and has created 37 articles today alone.

@Admins willing to patrol AINB: a block from article space seems to be required. ‑‑gurkubondinn 16:07, 27 September 2026 (UTC)

Moved the page to draft space and then removed the CSD tag again in AbuseLog/45234332, a full block might be needed. ‑‑gurkubondinn 16:12, 27 September 2026 (UTC)

The Sani, Ladakh page has now been recreated twice in their user space:

• As User:Third003/sandbox2 (AbuseLog/45240480): now something else, which is presumably also AI-generated though
• As User:Third003/sandbox26 (AbuseLog/45250310): identical to Sani, Ladakh in AbuseLog/45232940, except for the byte sequence and chatbot block having been removed (I've requested G15 for this latest iteration).

Third003 has been copying their AI-generated pages into various userspace sandbox listed on User:Third003/sandboxlist. ‑‑gurkubondinn 11:55, 29 September 2026 (UTC)

• I've moved all the articles they have created to draft except for Arambagh district which has been reviewed by NPP and so is outside of my remit. I would not argue for a block if the user can account for their actions here without recourse to an LLM. In that situation I would ask them to describe their process and demonstrate an understanding of what they are doing wrong. Then they could look at their articles again and see if they can fix the ai-signs with a view to possible submission to AfC. The articles are still under-sourced in my opinion, but that is not a question for this board.Boynamedsue (talk) 16:22, 27 September 2026 (UTC)

I've unreviewed Arambagh district since XTools shows that they are the main contributor to it and the next contributor in the list just added archive urls with IABot in Diff/1374007228. These articles will probably just got LLMPROD deleted btw. ‑‑gurkubondinn 16:29, 27 September 2026 (UTC)

I would just note this dif from their talk page, which shows that they were cooperating with questions on LLM.Boynamedsue (talk) 16:46, 27 September 2026 (UTC)

But in a way that seems chatbot-advised rather than "understanding what went wrong". Chatbots often advise users to edit war over CSD templates so this is consistent with what we are seeing here. Hence leaning CIR, despite cooperation. Fermiboson (talk) 18:45, 27 September 2026 (UTC)

The fact that the editor has accepted responsibility of using AI while creating their articles is important, but in my opinion cooperation by itself does not clear up the problem here. The editor's own comments that they use AI in creating articles makes it important to look whether an AI machine was actually used to generate or rewrite article content(which clearly has), rather than simply for assistance. I don't think we should assume that cooperation means the use of LLM article content or AI talk page responses is acceptable and it doesn’t mean they will not do it again when they’ve already admitting to using it regularly. Hasekidms (talk) 20:01, 27 September 2026 (UTC)

That's one thing, but sanctions are WP:PREVENTATIVE rather than punitive, so we normally don't impose sanctions without continued misbehaviour. The issue here is that the person still does not seem to comprehend, based on their actions, what the misbehaviour is. Fermiboson (talk) 20:13, 27 September 2026 (UTC)

I would agree with both pars of that. Without meaningful engagement here, at the very least a mainspace block is necessary.Boynamedsue (talk) 21:44, 27 September 2026 (UTC)

User:TheCraftyCailleach

 TheCraftyCailleach (talk · contribs · logs · block log) (27 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Sort of confirmed AI use on Talk:Project Sheela § Possible AI-generated text which was tagged by Gnomingstuff in August 2025. Had first removed the {{ai-generated}} tag in Diff/1376465057 (Special:AbuseLog/45205766).

• Created Cork LGBT+ Archive in Diff/1372442479 (AbuseLog/45009753) with two AIREDCATS
• Created Transgender history in Ireland in Diff/1361248000 (AbuseLog/44528141) with two AIREDCATS and references that return 404 errors.
  • Reference nr. 19 returns a 404 error, and has never been saved by archive.org.
  • Reference nr. 41 also returns a 404 error, and has not been archived either.
  • (There might be more, I didn't check all 53 references.)
• Added a reference to Gaulstown Portal Tomb in Special:Diff/1376999487 with a reference that doesn't exist.
  • Reference nr. 8 returns "the Wayback Machine has not archived that URL".
  • Posted the same non-existing reference in the AI-generated § September 2026 update: new sources, images and detail Gaulstown Portal Tomb talk page section in Diff/1377003502. There's also an unicode multiplication symbol character.
• AISUMM in Diff/1376999487
• AIREDCAT in Diff/1376994244
• Has a somewhat canned AI-generated user page with an § Editing interests section

Should maybe mention that they work for m:Wikimedia Community Ireland according to their use page. ‑‑gurkubondinn 18:55, 27 September 2026 (UTC)

Hello! Firstly, thank you for highlighting this as in reviewing the above I took a second to go back to the AI/LLM policy (as I do use it for help a lot with wiki source code work and structuring help (and sometime for copyediting or feedback on drafts) and I hadn't reviewed the policy since last year. So I have now done that.

Just to clarify a couple of points:

In reference to "Added a reference to Gaulstown Portal Tomb in Special:Diff/1376999487 with a reference that doesn't exist"

• Can you please advise which reference you are referring to? The Diff link highlighted two that I added, Du Noyer and Powell. Both which exist and I have sourced from. Du Noyer is in the reference list (I also linked to his own Wikipedia page) and I supplied the link to the JSTOR article (which shows the sketch on pg 478) and also there is an additional reference (which I haven't included in the article references but I did link RSAI on the article Talk Page) and the other highlighted reference in the Diff link seemed to be for Powell is also a real reference, it's an article/paper.
• If you are referring to a different reference than these two, can you please let me know?

Re: "Reference nr. 8" - this was a typo while editing I somehow made to an existing reference link to Old Moore's almanac that existed in the previous version of the article - it just ended up further down on the reference list in the later versions as I added additional sources for earlier sections of the article

^^ That is now corrected.

'''For Transgender history in Ireland I will review this fully tomorrow I need to check against my notes.''' I take your point on not archiving the links - I'll try to do that better going forward and adapt it into my workflow. It's something I generally forget to do but it may also be an LLM / markup gen issue with links being incorrect. I'll check and correct either way.

'''re: AIREDCAT issues - given how I use LLMs to support my editing with wiki source this does happened sometimes and mostly I catch it before publishing.''' That said, I should take more care to make sure the LLMs in their reviewing of my work are not making additions or adding hallucinations, or breaking links. I *do* review before publishing (whether strictly raw writing or LLM-reviewed / wiki source structured) and switch between visual and source editing, but sometimes I will still find that there is a typo in a word, or link, category, etc. or an issue with markup that I see *after* publishing and I go back in again to fix them, especially happens with categories and when I am doing it manually. I recently was introduced to HotCat and I am trying to use more wiki tools.

I do struggle with wiki source which is my main use of LLM assistance in my editing, though I am trying, learning and improving all the time doing it manually. I have used it to help me structure my User Page and tidy my work lists and wiki mark up however it's my understanding that the genAI/LLM content rules relate to the writing of encyclopedia article content only?

Regarding the removal of the tag on Project Sheela, as mentioned in my Talk Page reply regarding removing the tag, I had intended to add the restructured article I had been working on and restored the tag after seeing the message when i logged back on to update my reviewed draft.

'Re: the multiple "abuse logs for filter 867: Large creations by inexperienced user

I work off-wiki in my drafting - a hangover from academic writing, working on copywriting and a collaborative workflow I am more used to Word Docs for drafting work than using my Sandbox, so my edit count is low and my editing style / preference is to find bigger gaps and work on those over a long period of time in a word document and then go to Wikipedia. I like finding missing topics and doing a deep dive on them, or local topics that don't have developed articles and adding more information, images, etc. to them

It is correct that I work for a WM affiliate in my day job (mainly training, organsing and outreach) and this is my volunteer account which I used to work on developing articles on my research and personal interests, so my edits on my work account have mostly been in organising and training/demos. This account is my personal volunteer editing account and live Wikipedia editing practice on English Wikipedia is something I am still working on. Up to the last 6 months I had been primarily focused on supporting a different language Wiki.

Very happy to take the feedback on board and be more careful going forward with links, tags, categories, etc. TheCraftyCailleach (talk) 23:09, 27 September 2026 (UTC)

Nobody mentioned filter 867. Anyways, your LLM use is clearly and blatantly against policy, and you must therefore stop. NEWLLM has been in since last year, so you really have no excuse. is also fairly blatantly AITALK and I have no idea why a communications professional thought it appropriate to communicate with people using AI. I don't know if you have got through yet that the link, tags, and categories are not the problem. They are the symptom of the problem. The problem is that you are using AI to write articles, in a manner that I struggle to characterise as anything but "knowingly against policy". You must commit to ceasing the use of LLMs. Fermiboson (talk) 04:41, 28 September 2026 (UTC)

Hi Fermiboson,

As per my reply above, I have updated/refreshed myself on the policies and I apologise that I was interpreting these incorrectly and hadn't read NEWLLM (which I see was published in October 2025). This was a genuine error based on my reading of the previous policies which I was doing last summer, before NEWLLM was created and agreed.

Re AITALK (again to explain) - this was assisting me in a concise and basic overview of the edits between versions which I added additonal commentary to myself and invitations to other editors manually with explanations - again something that was intended to make the process easier for me and with better wiki mark up. Again, to reiterate, this was my prior understanding that this was not editing encylopedia content so was exempt from the policy, I realise now that is not correct and I won't be using any LLM going forward with any of my Wiki editing.

Also, just to explain: I mentioned filter 867 as it was the reason given on some abuse logs linked on the OP above so I thought these needed to be addressed too. I do understand that the link, categories, etc are not the issue - I was trying to explain as per the invite to discuss from gurkubondinn's OP to address the points raised on REDCAT issues and the references that were claimed "don't exist" on Gaulstown Portal Tomb which I have replied to above.

As per my previous reply, I have already reviewed and cleaned up Project Sheela and will be doing the same later today with Transgender history in Ireland and can reply here, or wherever you deem appropriate when that review is done. TheCraftyCailleach (talk) 05:57, 28 September 2026 (UTC)

Please use your own words when responding or communicating with other editors. It is not allowed to use ai to respond in talk pages or any noticeboard for that matter. Your response alone makes me suspicious of this problem just the structure of it alone. It’s very common in AI machine generations. Hasekidms (talk) 05:37, 28 September 2026 (UTC)

@Hasekidms, I'm sure the above comment by TheCraftyCailleach is human written- AI learned how to make bolded headers from humans and it's a common way to format long chunks of text + words and phrasing is absolutely human. The Gaulstown Portal Tomb talkpage comment reads likely AI generated to me, however. This is all based on months of experience doing AI cleanup, I don't use detectors. Sarsenet•he/they•(talk) 05:47, 28 September 2026 (UTC)

Hi @Hasekidms and thank you @Sarsenet - my replies here are indeed written by myself (a human), and I spent quite a lot of time drafting my initial reply trying to understand the points on the references particularly. I use headings to attempt to break up my points and make the answer easier to read, especially when trying to address multiple things in one response.

As per the reply just above here to Fermiboson, I used an LLM to help with a concise list of changes between versions summary for me for the Talk page to produce a bullet list in wiki markup and then I added more to it myself addressing other editors.

TheCraftyCailleach (talk) 06:12, 28 September 2026 (UTC)

User:Grifomaniacs

Grifomaniacs (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Loads of markdown hits, done in 20 minutes, some fairly flowery language throughout. Fermiboson (talk) 20:20, 27 September 2026 (UTC)

A day prior the article was flagged for AI generated content by @Gurkubondinn. Then it was resubmitted the same day with AI generated content again flagged by @Fermiboson. Maybe a translation machine? Hasekidms (talk) 20:27, 27 September 2026 (UTC)

I also spent time updating the page in question to address potential concerns and submitted it for review Draft:Guido Mazzetti, . Honestly we can debate about quality and tone, for the rest I can guarantee I am the editor of the content. User:Gurkubondinn didn't reach out to me when I contacted him. Grifomaniacs (talk) 20:55, 27 September 2026 (UTC)

Quite possible, Grifomaniacs has claimed in § Guido Mazzetti on my userpage talk page that one of the pages is a machine translation. I'm not sure what they mean here saying that they are the editor of the content, but if these are machine translations they are unattributed and fail the LLMT guideline. Since both me and Fermiboson independently found these and could tell that they are LLM-generated, so they most likely fail the NOLLM guideline as well, regardless of being unattributed machine translations or new generations. ‑‑gurkubondinn 11:48, 28 September 2026 (UTC)

@Gurkubondinn you can engage me directly, no need to use any third person. I already explained you that I used machine translation to assist me on a first draft, I carefully reviewed and improved, changing syntax and tone, thus being the original author of the translation. The above link by @Fermiboson is pointing out a manual edit I did on the markup of Wikipedia to update original content I added to the football jerseys of AC Perugia from https://commons.wikimedia.org/wiki/AC_Perugia_Calcio_kits so should be out of context here.

Since we are in the AI noticeboard, what's important to me is to confirm I followed Wikipedia:LLM-assisted translation. The guideline explicitly permits translation tools (including LLMs) on the condition that a human thoroughly reviews, verifies, and polishes the text, and this is exactly what I did. Grifomaniacs (talk) 12:14, 28 September 2026 (UTC)

No, I was responding to a reply by Hasekidms, so if I used the second-person pronoun 'you' then I would have been referring to them (and now that I'm reply to you, second person pronoun, I'm using the third person singular they pronoun to refer to Hasekidms).

Nothing is "out of context" here, we are discussing your editing as a whole and not just one or two specific pages/edits. The edit that Fermiboson linked to is Diff/1376832831 that created the 1997–98 AC Perugia season article. This article does not have the WP:ATTREQ mandatory attribution or anything that indicates that it is a machine translation, so it would seem to fail the NOLLM guideline. The {{translated page}} template on the talk page does suggest that this is a machine translation from itwiki, so this would then fail the LLMT guideline as well (as I have explained in our parallel conversation on my talk page). The LLMT guideline requires more than what you mentioned, and if two editors have identified your machine translations as being LLM-generated, the requirements that you did mention have most likely not been met.

You should not underestimate the work that a translation (and language localisation) takes. To translate a body of work is a substantial amount of work, and it requires a deep understanding of both the source language and the target language. A translation is often a body of work in its own right, and it is not a trivial task that you can knock out a handful of articles of when you have a free moment. Using an LLM for translation may seem like a straight-forward and an easy task, but it is not. Your pages were spotted as such by two editors, independently of each other, which does suggest that it wasn't done well enough. But translating is a skill that is very much worth developing, but you should not treat it as a low-effort and an easy task. ‑‑gurkubondinn 12:50, 28 September 2026 (UTC)

I added here the attribution, following your advice @Gurkubondinn as per WP:DUMMY edit to WP:RIA.

Since I wrote in the earlier reply "I am the editor of the content". I thought you were quoting me using "They". You can refer me using the pronoun He/Him, even if my username looks like a plural name.

To all Wikipedia users in this thread: I'm working in good faith to fix the attribution error. The articles are good and I'm fluent in both languages and been doing a lot of translation and original writing over a span of 20 years at Wikipedia! I think in this case you are really going beyond the original purpose of the AI noticeboard. Grifomaniacs (talk) 14:05, 28 September 2026 (UTC)

Why do you think I have referred to you in plurality? ‑‑gurkubondinn 15:52, 28 September 2026 (UTC)

You wrote exactly "I'm not sure what they mean here saying that they are the editor of the content" and since I was the only one writing "I am the editor of the content" it seems like you referred to my message using the third plural person. I kindly ask you to use He/Him pronoun when referring to me. And honestly, to be nice, you could have tagged me instead, writing something like hey User:Grifomaniacs I'm not sure what do you mean here by saying this, could you clarify? So that it is constructive communication. Grifomaniacs (talk) 16:39, 28 September 2026 (UTC)

Hm I thought you said that you were fluent in English? That is a singular they, and it is very commonly used in English. ‑‑gurkubondinn 16:46, 28 September 2026 (UTC)

It's even what your pronoun setting is set to: {{pronouns|Grifomaniacs}} → they. But I have no problems respecting your choice of pronouns and I will be referring to you using he/him pronouns from now on. ‑‑gurkubondinn 16:48, 28 September 2026 (UTC)

Thanks since that was creating confusion. But totally off topic. I'd appreciate a path to resolution, with clear indications of which parts I should change in the article you rejected. The articles Draft:Guido_Mazzetti, Draft:Cesare_Butti, Draft:1998–99_AC_Perugia_season have been all reviewed and updated following your concerns. Maybe we need another POV, so that another admin will have a look into them with fresh eyes? Grifomaniacs (talk) 17:27, 28 September 2026 (UTC)

Are you saying that Draft:Cesare Butti is an unattributed machine translation? You said that you have been doing this for 20 years on Wikipedia, does that mean you have been submitting unattributed translations for 20 years?

Could you also address why you were confused by the singular they? That is something I would expect someone who is fluent in English to know. ‑‑gurkubondinn 17:56, 28 September 2026 (UTC)

@Gurkubondinn let's leave aside gender-neutral language debates and focus purely on the content and policy compliance.

Regarding attribution:

• Draft:Guido Mazzetti has been fully updated with the exact source revision ID (149319523) in the edit history and the talk page template.
• Draft:1998–99 AC Perugia season has also been updated with its source revision ID (149505226).
• Draft:Cesare Butti also updated with source revision ID 135696551. Here I also polished the text to ensure full encyclopedic tone.

I think the most constructive path forward for these drafts is to let another independent reviewer look at them with fresh eyes. I consider the attribution and policy requirements fully satisfied. Thanks. Grifomaniacs (talk) 19:17, 28 September 2026 (UTC)

Where is the "gender-neutral language debate"? This is just very normal English, and it is not something that would typically confuse someone who is fluid in English. That's why I find it odd that you were so thrown off by this. You're submitting what you say are LLMT-compliant machine translations, so your proficiency in the English language is relevant here.

And I Would also like it if you could clarify if you have been submitting other unattributed translations, because that would need to be taken care of. As I have previously mentioned, this is a legal requirement of the WMF ToS and the CC-BY-SA licence agreement. ‑‑gurkubondinn 19:22, 28 September 2026 (UTC)

And also, again, you should be using WP:DUMMY edits for WP:RIA attribution. You need to attribute the entirety of the derived work, not just one single edit. ‑‑gurkubondinn 19:23, 28 September 2026 (UTC)

I've done it, I used WP:DUMMY edits for WP:RIA. I'm fully fluent in English, no need to investigate further. Regarding the phrasing earlier, there are individuals who are sensitive to the topic of misgendering and I don't want to start a debate on that. I didn't appreciate being quoted in third person, I told you that I would have preferred a constructive kind of communication with a direct question in second person instead. Since now all technical policy requirements for the current submissions are satisfied, let's allow an independent reviewer to evaluate the content on its merits. Grifomaniacs (talk) 19:55, 28 September 2026 (UTC)

I know what gender neutral language is, we even have the MOS:GENDER style guide for it. That wasn't my question, I asked where there was a "gender-neutral language debate" here, because there isn't one. The question is why the singular they confused you, someone who is supposedly not only fluid in English, but also has sufficient command of the language to proofread machine-translations into English. Because these two things do not make sense together.

• Diff/1377287671 to Draft:Guido Mazzetti: not a dummy edit
• Diff/1377289265 to Draft:1998–99 AC Perugia season: not a dummy edit
• Diff/1377292851 to Draft:Cesare Butti: not a dummy edit
• Diff/1377234062 to Draft:Guido Mazzetti: not a dummy edit and only attributed the text in that edit
• Diff/1377297233 to Mario Malatesta: not a dummy edit

If you are having issues figuring out how to make a WP:DUMMY edit, then you should have asked for help with it. If you still believe that these are dummy edits, then you should read the documentation again. ‑‑gurkubondinn 20:22, 28 September 2026 (UTC)

Above, you’ve claimed that you speak English fluently, but I think @Gurkubondinn use of “they” here was simply ordinary singular “they,” which is standard English usage. I don't think it was reasonable to interpret that sentence as referring to you in the plural. It’s pretty clear that “they” was being used as a singular pronoun. I also think this is moving the discussion away from the actual issue. The main concern here is the AI-generated content and the mistakes that need to be reviewed. Whether AI was used for translation or something else, it was still used in the process, and that means the content should be reviewed for possible errors or hallucinations. I don't think the pronoun issue should become the focus of this discussion when the main issue is the content itself and the use of AI. Hasekidms (talk) 22:47, 28 September 2026 (UTC)

It could also be indicative of a larger issue, because WP:LLMT requires users to be skilled enough in both the origin language and English to confirm the translation is accurate. So I think that it is fair to look it the WP:CIR part of it too. ‑‑gurkubondinn 23:00, 28 September 2026 (UTC)

I have just requested G15 deletion of 2024–25 AC Perugia Calcio season in Diff/1377300018 because 6/7 references didn't exist.

• Ref nr. 2 returns 404 not found, only archived as a 404 error page.
• Ref nr. 3 returns 404 not found, only archived as a 404 error page.
• Ref nr. 4 does not resolve, archive.org has never saved the url.
• Ref nr. 5 returns 404 not found, only archived as a 404 error.
• Ref nr. 6 returns 404 not found, and is only archived as a 404 error.
• Ref nr. 7 returns 404 not found, and is only archived as a 404 not found.

The 404 errors were all archived on the day that the article was created or the day after, so they are saved as a result of archive.org crawling enwiki references.

This means that the page fails the WP:LLMT guideline (and the WP:V policy), specifically failing requirement nr. 3 to check and read all of the sources used. In fact, none of these sources are even present in the it:Associazione Calcistica Perugia Calcio 2024-2025 article on itwiki which has only ever cited transfermarkt.it. There is something other than translation going on here and Grifomaniacs has not be telling us the truth.

@Admins willing to patrol AINB: it is probably time for a block to prevent further disruption, at least until Grifomaniacs starts telling is the truth. ‑‑gurkubondinn 20:13, 28 September 2026 (UTC)

Spot checking Mario Malatesta, which is said to have been translated from it:Mario Malatesta at revision it:Special:Diff/152462273 (which is the current revision). The attribution was added in Diff/1377297233.

The enwiki article cited two sources (one of them was WP:TRANSFERMARKT and has been removed) when it was created in Diff/1376798819, none of which are present in the itwiki article. The itwiki article cites five different sources, and is also relatively longer than the machine translated enwiki article. ‑‑gurkubondinn 20:35, 28 September 2026 (UTC)

This is a localized translation. Adding "Il Pallone d'Oro - Prima enciclopedia storica del calcio mondiale" as fundamental source. Grifomaniacs (talk) 21:08, 28 September 2026 (UTC)

@Gurkubondinn I promptly corrected all of the sources there. Calling for blocks and making accusations of bad faith goes well beyond normal collaboration, honestly it is uncalled for and unnecessary. Grifomaniacs (talk) 21:03, 28 September 2026 (UTC)

It's not easy to work with the wiki markdown, is it a capital sin to use AI assistance to adjust the errors we meet when moving from templates that work in one language and not in another? In order to complete the pages I need to format tables re-using content from other existing pages across various wikis? The page is a Serie A Season, which includes data already verified. And all the information, data, squad lists, and sources provided are thoroughly checked, accurate, and completely verifiable. I can work on the tone, if you don't like it. I'd really appreciate you to flag the section with the unappropriate wording. Grifomaniacs (talk) 20:38, 27 September 2026 (UTC)

@Fermiboson @Hasekidms

I reviewed again the wikipedia policy. The articles I added and now were moved into drafts are totally fine to be publushed, according to policy. I think there's a big misunderstanding here, since translations are admitted!

References:

WP:TFOLWP: When translating material from a Wikimedia project licensed under CC BY-SA, a note identifying the Wikimedia project (such as an interlanguage link) and the page name must be provided. The preferred way to do this is in an edit summary in the translated page.

Where applicable, the template {{Translated page}} may also be added to the talk page to supplement copyright attribution.

Wikipedia:LLM-assisted translation: The guideline explicitly permits translation tools (including LLMs) on the condition that a human thoroughly reviews, verifies, and polishes the text.

I manually verified, cross-checked sources from the Italian wiki, and rewrote the syntax.

The pages are ready to be published with proper edit summary, on top of the template in the talk page, which was already added.

Hopefully we can solve this case and move on. From my standpoint, I will ensure to be really careful with all of the above aspects, and address any other comments from your end regarding the content in the articles. Grifomaniacs (talk) 11:48, 28 September 2026 (UTC)

Just a note that although AI tools are allowed you are responsible for their output. Had there been a few articles like this then you would have been blocked until this could be resolved. It's nothing personal, we get people who mass create AI articles and it's kind of the only way to stop the spread. Dr vulpes (Talk) 22:53, 28 September 2026 (UTC)

I should have said AI tools are allowed in limited ways and cases but you get the idea. Dr vulpes (Talk) 22:55, 28 September 2026 (UTC)

User:JhowieNitnek

JhowieNitnek (talk · contribs · deleted contribs · logs · filter log · block user · block log)

triggered LLM paste check and then lied on my talk page. Fermiboson (talk) 20:47, 27 September 2026 (UTC)

I did not lie. Detected is not the same as actually having used it. I copy pased info from the wiki page of the Henri van der Noot mainy "jurist, lawyer and politician from Brabant. He was one of the main figures of the Brabant Revolution" and adapted it for the "Known for section" by making it "Jurist, lawyer, and politician who was one of the leading figures of the Brabant Revolution." I ADAPTED A SENTANCE. Jhowie_Nitnek (talk) 20:52, 27 September 2026 (UTC)

I did not lie. Detected is not the same as actually having used it. I copy pased info from the wiki page of the Henri van der Noot mainy "jurist, lawyer and politician from Brabant. He was one of the main figures of the Brabant Revolution" and adapted it for the "Known for section" by making it "Jurist, lawyer, and politician who was one of the leading figures of the Brabant Revolution." I ADAPTED A SENTANCE. Jhowie_Nitnek (talk) 20:53, 27 September 2026 (UTC)

(I did not mean to send that 2x, I have some loading problems on this page) Jhowie_Nitnek (talk) 20:54, 27 September 2026 (UTC)

chatgpt-derived links in May 2025 , checks out Special:Permalink/1293406868. For Elishout Farm (25 Sept) The farm is considered a surviving example of the large abbey farms that once characterised the rural landscape around Brussels. is AILEGACY and completely unsupported by Kowal2701 (talk, contribs) 21:09, 27 September 2026 (UTC)

Last thing is incorrect as it says in the text "Het geheel is een van de mooiste voorbeelden van wat een grote abdijhoeve in de omgeving van Brussel was" aka "The farm is considered a surviving example of the large abbey farms that once characterised the rural landscape around Brussels", the source you you does not support it in fact does. As for the other, this was from before I was aware of the AI policy, I stopped using it once I was made aware. Jhowie_Nitnek (talk) 21:18, 27 September 2026 (UTC)

User:System tinker 90

 Cleanup has been requested.

I'm not confident about this one, would appreciate someone else having a look.

This user has been posting lengthy new articles and large additions to other page at a rapid rate (topic focus seems to be central European business and finance). Pages don't appear to be straight-forward translations of entries in other languages, so I don't think its un-attributed translations.

There are some AI signs (misplaced templates and broken references and some redcats ), but aside from that I'm not confident. All the articles I've put into GPTZero have a high ai score, but I tried feeding some material into Pangram as well and had mixed results.

Another user has draftified a lot of their more recent articles for AI usage, but they have been very prolific and there is a lot of extant material. Giulio 21:30, 27 September 2026 (UTC)

Ive reviewed the work of the use and it does has signs of AI that require review. The structure, citing, the tone, looks copy-pasted and ai generated. I also noticed the user was not notified on their talk page about this discussion. I’ll go ahead and do that. Hasekidms (talk) 22:24, 27 September 2026 (UTC)

User talk:System tinker 90#h-LLM use-20260926205500 , user made at least 7 articles which were moved to draft space by other editors due to AI generated content. All the articles were drafted within the same day. Hasekidms (talk) 22:41, 27 September 2026 (UTC)

I also think the number of articles being created in a short time is concerning. I think an admin should look into the user's activity and consider a block. Hasekidms (talk) 22:48, 27 September 2026 (UTC)

Hi Giulio and Hasekidms,

I understand that my edits raised a valid issue, let me apologize for that.

Recently, I was using an LLM as part of the editing process, mainly for outlining and finding possible references, which were then still checked. In a few cases, I used an LLM to help with wording in parts of a draft for a new article, which I then edited.

These articles were about topics I was already familiar with or had researched before, like European companies, entrepreneurs or fintech products. That made it possible to work on several related articles in one day.

I’m now going through the affected articles and fixing the issues. I'll pause creating new articles and editing articles I didn't create. I’ll also pay much more attention to Wikipedia rules.

Thanks for your patience. System tinker 90 (talk) 12:53, 28 September 2026 (UTC)

edits on 21 August and 24 Sept had markdown . 64 creations this month Kowal2701 (talk, contribs) 13:01, 28 September 2026 (UTC)

Tithayaramepasiyantu

Cleanup efforts are ongoing and being tracked at /2026-09-28 Tithayaramepasiyantu, 10 pages need review.

LLMPRODs are done and drafts tagged

Relatively easy one, they disclosed AI use. (not that it was really in doubt given edits have stuff like backticks, etc) Am handling the non-article creations, rest likely needs LLMPRV. Gnomingstuff (talk) 22:49, 27 September 2026 (UTC)

It seems the user was previously notified about AI use here: User talk:Tithayaramepasiyantu#h-Proposed deletion of Gyanvimalsuri-20260604172100. Given the continued issues, the fact that the user doesn’t seem to understand the underlying issue, I think an admin should consider a block. —Hasekidms (talk) 23:00, 27 September 2026 (UTC)

Not sure a block is necessary at this time, they do not seem to be active Gnomingstuff (talk) 00:26, 28 September 2026 (UTC)

Special:Permalink/1330777284 (2 Jan) is obvious, an edit on the same day had markdown (Special:AbuseLog/43085241), LLMPRODing and creating a tracker Kowal2701 (talk, contribs) 12:47, 28 September 2026 (UTC)

User:PrabhatRanaWiki

 Cleanup has been requested.

 PrabhatRanaWiki (talk · contribs · logs · block log) (28 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Some AI tells:

WP:AIPLACEHOLDER

WP:AIREDCAT

Very rapid rate of content generation and re-writing their own articles:

Completely different writing style between articles they've written (compare Draft:Ramotalahi Kareem and Aiping Yu) Giulio 00:48, 28 September 2026 (UTC)

Appears affiliated with this edit-a-thon Gnomingstuff (talk) 01:32, 28 September 2026 (UTC)

is this the one with the cash prize again? | One Reaction was here. Got a complaint? 15:00, 28 September 2026 (UTC)

Apparently not. Still, editathons really need to take quality of the edits into account. | One Reaction was here. Got a complaint? 15:24, 28 September 2026 (UTC)

They've repied to Giulio's messages on their talk page, about their articles being draftified, with obvious LLM responses, claiming to have reworked the pages: , , .

Draft:Aiping Yu has been recreated in one edit, with redlinks and two AIREDCATS, multiple statements not supported by the source they're cited to, A 2020 profile by the Natural Sciences and Engineering Research Council of Canada described her work on graphene-based materials for supercapacitors and protective polymer coatings (AIATTR).

I was going to go through the other two but I don't think it's actually necessary. Suðurhafsljósæta (talk) 19:59, 28 September 2026 (UTC)

Been out all day, was about to post this myself, I will also add that they've created a few more articles today. Haven't had time to look at them too closely yet, but I immediately noticed some AIREDCATS Giulio 20:03, 28 September 2026 (UTC)

Including 2 identical copies of the same article, one as a draft, the other in mainspace Draft:Angèle St-Yves Angèle St-Yves Giulio 20:09, 28 September 2026 (UTC)

Since this discussion, @PrabhatRanaWiki has used an LLM to re-generate two of the draftified articles and submit them at AfC: Draft:Céline Vaneeckhaute and Draft:Helen Griffith Wylie Watson.

There's a wonderfully glaring red flag in Draft:Céline Vaneeckhaute: the draft states that She completed two doctorates in 2015: one in water engineering at Université Laval and one in bioengineering at Ghent University. Not to diminish Professor Vaneeckhaute's accomplishments, however, she completed one doctorate in 2015, jointly supervised by those two university departments.

@Admins willing to patrol AINB: please take a look. I think a mainspace block is warranted at this point. Suðurhafsljósæta (talk) 07:57, 30 September 2026 (UTC)

Oh, amazing.

I am trying to love myself (and get to bed on time), so I'm not listing all the issues with the "substantially revised" articles, but the other two you notified them about are just as bad as Draft:Aiping Yu. I'm seeing a lot of deprecated infobox parameters ('alma_mater'), too. Suðurhafsljósæta (talk) 20:13, 28 September 2026 (UTC)

User:Nsokolsky

 Cleanup has been requested.

Nsokolsky (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Huge edits that sound AI-generated, probably translated from Russian. Notices on user talk page are deleted without response. Thank you. Magnolia677 (talk) 11:13, 28 September 2026 (UTC)

Almost certainly AI. Three massive edits in the span of less than 5 minutes. That should be a dead giveaway. Prodding it to see if it still moves. | One Reaction was here. Got a complaint? 14:56, 28 September 2026 (UTC)

dollars to donuts it's Grok specifically Gnomingstuff (talk) 15:10, 28 September 2026 (UTC)

how do you know it's Grok specifically | One Reaction was here. Got a complaint? 15:19, 28 September 2026 (UTC)

could be Claude, but there's often a focus on "however, critics argued that x is insufficiently dedicated to EMPIRICAL FACTS and CAUSAL LOGIC" that Grok is particularly fond of (it's all over Grokipedia, for instance)

also just the context clues that someone who is editing exclusively on the subject matter they are exclusively editing on is culturally likely to be using specifically Grok or Claude Gnomingstuff (talk) 16:14, 28 September 2026 (UTC)

On what basis are you saying these are translated from Russian, other than the username? Most of these don't have Russian articles and the subjects are predominantly English-speaking. (I have been watching this user's talk page since they edited Pangram (AI detector); my first couple of spotchecks of their changes there matched the sources so I did not revert as I didn't have AI use evidence beyond "new user makes large change".) Rusalkii (talk) 16:55, 28 September 2026 (UTC)

User:Waxapple

 Waxapple (talk · contribs · logs · block log) (28 September 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. 

Created Gerry Mallon (banker) in AbuseLog/45237309 with a chatblock, so it has been G15 deleted now. Then created James Anderson (fund manager) in Diff/1377137505 ten minutes later, I've sent that one to draft space.

Added a total of c. 27 kB to Heart of Midlothian F.C. Reserves and Academy, included SUPERFICIAL analysis (loan destinations reflect the 2026 summer transfer). Reference nr. 65 in Permalink/1377086312 is just a blank page, and archive.org does not seem to have saved a working copy either.

Created 1990 Hibernian F.C. takeover bid in Diff/1377101088 with an AIREDCAT (sent to draft space). Reference nr. 4 does not work, and has never been saved by archive.org.

Created Leslie Deans in Diff/1377106014. Reference nr. 2 does not exist, and has only ever been saved in archive.org as a 404 page. This article was also created with a red link, which was then quickly removed in Diff/1377106522 so it was probably not meant as a plausible red link. Created the Kenny Waugh article 14 minutes later, but I haven't looked at that one.

Created Céderique Tulleners about 20 minutes later in Diff/1377109285 with two AIREDCATS (now in draft space). None of the references were linked, despite having been published in 2025 and 2026. Added URLs to the references in Diff/1377115786, along with a handful more paragraphs and references.

Has added c. 24 kB to Tynecastle Park over many edits. One of references added in Diff/421553976 does not exist, and archive.org has only saved the URL as a 404 error page. There is an archived copy of the article on the older url, but the archived URL was not included in the edit despite other edits having used archived urls. The reference has an access date from 2011, and I'm not sure what's up with that. The Diff/1376612837 edit has a reference that doesn't work anymore, but that reference does use {{cite web}} with url-status=dead and an archive-url parameter.

When I asked them to stop in § September 2026 on their talk page they said that I was flagging false positives, but did not outright deny using AI for the edits and articles. Waxapple has edited before public access to LLMs and chatbots, but the majority of their edits are in September 2026 and some earlier in July and in August. There is also a fair amount of articles and large user space drafts that I have not looked at. ‑‑gurkubondinn 15:00, 28 September 2026 (UTC)

The second article (and a few of the others) may be eligible for G15. | One Reaction was here. Got a complaint? 15:33, 28 September 2026 (UTC)

I've nominated a few for G15, but had draftified the second one before I noticed that a reference didn't exist (and its just one of them). ‑‑gurkubondinn 15:42, 28 September 2026 (UTC)

~40 creations this month Kowal2701 (talk, contribs) 17:53, 28 September 2026 (UTC)

User:WindBorneListener

Cleanup efforts are ongoing and being tracked at /2026-09-28 WindBorneListener, 857 pages need review.

WindBorneListener (talk · contribs · deleted contribs · logs · filter log · block user · block log)

ChatGPT params in Special:AbuseLog/43512641, Markdown in Special:AbuseLog/43092813 and Special:AbuseLog/43137282, plenty of indicators in text of edits. Not warned specifically for AI (until now) but has user talk page comments (which they completely ignored) about source-to-text integrity issues and such. Gnomingstuff (talk) 15:37, 28 September 2026 (UTC)

triggering filter 1325 for a very strong sign ("underscoring the importance of") in Dec 2025 (Special:AbuseLog/43051243) around the time of their first creation, a few days later there's chatgpt params (Special:AbuseLog/43056709) and markdown (linked above). They blanked your warning saying it was resolved. I'll reach out and ask if they want to work on any of their articles rather than us deleting them without notice. Will also make a tracker Kowal2701 (talk, contribs) 18:04, 28 September 2026 (UTC)

well, that was more than I expected Kowal2701 (talk, contribs) 18:08, 28 September 2026 (UTC)

yikes! Suðurhafsljósæta (talk) 19:25, 28 September 2026 (UTC)

notified them of this thread instead Kowal2701 (talk, contribs) 18:11, 28 September 2026 (UTC)

User:WellHelloBuddy

 WellHelloBuddy (talk · contribs · logs · block log) (28 September 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. 

Triggered the LLM paste check with Diff/1377089717 to Eric Wendt. Said on their talk page that the decoration were sourced from his Military Portrait and to have never used AI not once when editing this article.

Triggered the regular edit paste check when creating Itaukei Land Trust Board (now in draft space) in Diff/1364328181 (AbuseLog/44659313) with AIBOLD, an AITABLE, and 3x UTM parameters showing ChatGPT use. References nr. 1 and nr. 3 return 404 errors and archive.org has never saved either of them.

Again triggered the paste check when creating Itaukei Affairs Board (now in draft space) in Diff/1364321639. Reference nr. 1 returns a 404 error, and archive.org has only saved the URL as a 404 error page.

Triggered a paste check when creating Ministry of Civil Service (Fiji) in Diff/1363464400 and then Minister of Civil Service (Fiji) in Diff/1363466098 within half an hour of each other. Both have a bunch of AIBOLD and both have been draftified for not having any references.

Triggered a paste check with Diff/1343592827 to Suva City Council, adding a bunch of AIBOLD to the article.

Created People First Party (Fiji) in Diff/1367580834 (AbuseLog/44798280) with an UTM parameter from ChatGPT.

Denied using AI in § September 2026 on their talk page, saying that Nothing Was AI and that they were 100% sure to have not copied anything at all from a ChatGPT window. ‑‑gurkubondinn 15:40, 28 September 2026 (UTC)

I noticed that someone said on WellHelloBuddy's talk page that their user page showed AI signs, and wanted to mention that I don't think it looked that way at all (pre-blanking).

The AIBOLD and bad references, however, are very clear signs, and the UTM parameters and LLM paste check are quite incontrovertible. Here's another very obviously LLM diff from March 2026, which was reverted as such: Special:Diff/1342669747.

@WellHelloBuddy, you've said that you didn't use AI for your edit to Eric Wendt. There are technical reasons why that's hard for us to believe. Have you used LLM for your other edits and page creations here? Suðurhafsljósæta (talk) 16:12, 28 September 2026 (UTC)

I think that you linked the wrong diff, Diff/134266974 is something else? ‑‑gurkubondinn 16:17, 28 September 2026 (UTC)

Whoops. Fixed now, thanks for catching that. Suðurhafsljósæta (talk) 16:23, 28 September 2026 (UTC)

User:LeSablierZeus

 Cleanup has been requested.

 LeSablierZeus (talk · contribs · logs · block log) (28 September 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. 

On 21 Sep 2026, this fairly new user (account 20 Sep 2026) created this new article whose source shows a strong sign of LLM usage (check the wikisource to see it).

In the space of less than two hours on 28 Sep 2026, this user made seven major rewrites and one article creation in relation to the Wikipedia:Contentious topics/Horn of Africa sub-topic of Ethiopia (and also on another Horn of Africa topic):

• 00:58 28 Sep +3464 ET conflict between the Ethiopian (ET) federal government ("fed-govt" for this discussion)
• 01:16 28 Sep +4848 ET conflict anti-fed-govt group
• new article +8631 ET conflict anti-fed-govt person
• 01:49 28 Sep +5320 ET conflict anti-fed-govt person major expansion
• 02:02 28 Sep major rewrite ET conflict anti-fed-govt group: removal of child soldier recruitment
• 02:12 28 Sep +4446 ET conflict anti-fed-govt group
• 02:29 28 Sep +4051 ET conflict anti-fed-govt coalition of groups (reverted by me)
• 02:33 28 Sep +5130 ET conflict anti-fed-govt group

One edit, which I reverted, is the 02:29 28 Sep edit, in which the user made a massive change that converted a one-paragraph lead into a five-paragraph WP:LEAD (unjustified length in proportion to the article content), unnecessarily duplicated material from 2026 Ethiopian offensive instead of leaving {{excerpt}} in place, and had a major semantic inconsistency, which I won't state explicitly because of WP:BEANS; a browse of recent edits from the past few days on this related set of related articles will make it obvious.

The style of writing in these Ethiopia conflict related articles includes quite a bit of Wikipedia-recent-event-copy-paste-news-media-style that tends to be unavoidable until a year or so later when other people decide to clean up by rewriting in encyclopedic style; the information match to the sources seems mostly OK, with small copyright violations (typically copying a half-dozen words in an arbitrary, non-encyclopedic style from the source), as is common in Wikipedia newsy articles by people eager to get the information into Wikipedia but lacking the time to think about it and summarise in plain English. I have commented on the user's talk page so we can see if the user is willing to engage in own-words discussion.

For completeness, there is another user I have warned about here on AINB on the same topic. Some elements of style (such as reformatting references to one-parameter-per-line) seem to be common to the two users. This user started editing, prior to autoconfirmed status, on non-Horn-of-Africa topics; the other user created an account a day later (21 Sep 2026) and currently looks like WP:SPA. Boud (talk) 16:44, 28 September 2026 (UTC)

@LeSablierZeus: Here I'm just copy/pasting your comment above to this section: <nowiki>Hello! I am new to Wikipedia and I am still learning the ins and outs. I apologize for any misconduct or disruption I might've caused. I will go through more of the help articles so I can work on my editing the right way. Once again, sorry. ˜˜˜˜</nowiki> LeSablierZeus (talk) 16:50, 28 September 2026 (UTC).

I'm curious: why did you use <nowiki> here? Was that the output from an LLM? Boud (talk) 17:18, 28 September 2026 (UTC)

When I replied to you it did not show <nowiki> until you've just reposted it here. This is me, a real human, typing this message and the last one. I was told to use 4 tildas to sign my messages so I think that appears as well I hope. ˜˜˜˜ LeSablierZeus (talk) 17:28, 28 September 2026 (UTC)

I would be extremely surprised if the <nowiki> characters were added by mediawiki software. The only explanation I see is that in the previous message (of 16:44), you typed your message in another window on your screen, and then copy/pasted the text using a mouse from that other window into your Wikipedia editing window. Was that other window (in which you typed your message of 16:44) a window for entering prompts that were input to an LLM? Does the expression "I was told" mean that a human told you that? Anthropomorphisation of LLMs (talking about LLMs as if they are human) gets very confusing. Boud (talk) 17:57, 28 September 2026 (UTC)

I was told by a user in my talk page. I copy and pasted the 4 tildas from there because I don't have tildas on my keyboard. ˜˜˜˜ LeSablierZeus (talk) 18:01, 28 September 2026 (UTC)

Thanks for answering the second question: I see that Crtbob told you about the four tildes. I'll explain more about that on your talk page in a moment.

The first question still remains: Was that other window (in which you typed your message of 16:44) a window for entering prompts that were input to an LLM? Boud (talk) 18:21, 28 September 2026 (UTC)

FWIW that reply is textbook AI advice. Fermiboson (talk) 18:40, 28 September 2026 (UTC)

Out of all the responses, I think the only one written by the user is the following - When I replied to you it did not show <nowiki> until you've just reposted it here. This is me, a real human, typing this message and the last one. I was told to use 4 tildas to sign my messages so I think that appears as well I hope. - the <nowiki> appeared magically? Furthermore, ~ ~ ~ ~ is very different from <nowiki>. Hasekidms (talk) 01:58, 29 September 2026 (UTC)

User:Yikes9272

 Cleanup has been requested.

I came across Yikes9272 (talk · contribs · deleted contribs · logs · filter log · block user · block log) in New Page Review and found Draft:Anti-Christian Violence in Nigeria, which has several WP:AISIGNS and a completely non-existent reference and draftified it. I reviewed their additional work and found that an earlier revision of Amy Dangerfield and the most recent version of Blake Neff (activist) were G15-eligible. I draftified Draft:Litigation Attorney (Litigator), but the page creator has been prolific enough, in contentious topic areas, that a larger effort is needed. The page creator usually puts a large completed article into userspace, then edits it down before moving it to mainspace, but the signs remain present. The page creator previously removed an LLM template from another article. Articles needing attention:

• Ryan Mears
• Joshua Lisec
• Massacres of Christians in the Later Ottoman Empire
• Total Boomer Luxury Communism
• DIGNIDAD Act
• Marion County Prosecutor's Office
• Always With Honor: The Memoirs of General Wrangel
• Suicide of the West (Burnham book)

Dclemens1971 (talk) 17:41, 28 September 2026 (UTC)

Typically, I edit these articles in a separate word processor, then move them here. I removed the LLM template since I didn't write it with an LLM. Yikes9272 (talk) 17:58, 28 September 2026 (UTC)

Further, you've removed three articles with no evidence as to how they're AI generated. Amy Dangerfield was already reviewed and published, and these major revisions should've been discussed before making a series of major moves like this. Yikes9272 (talk) 18:00, 28 September 2026 (UTC)

No, when you have multiple non-existent/hallucinated references in the articles (such as these ones that were in the Blake Neff article), the articles become eligible for speedy deletion without discussion. There are absolutely signs of LLM use throughout your work, and if you continue to deny it you will likely be blocked from editing indefinitely. Dclemens1971 (talk) 18:03, 28 September 2026 (UTC)

You may need to check your sources before asserting that the links are "non-existent."

The first link you stated is a valid link which you can visit here: https://www.msn.com/en-us/news/world/candace-owens-challenged-over-claims-about-charlie-kirk-tpusa-successor/ar-AA24mwiw

The second link is broken, but nonetheless has been published multiple places. Apparently the link was dead by the time it was saved, but still...: https://www.northcountrypublicradio.org/news/npr/144737382/college-republicans-in-n-h-say-candidates-should-stop-playing-dumb

These are bad examples. Making threats to strip editing privileges is improper (per WP:THREATEN), especially when you haven't attempted to fix the errors you're pointing out (per WP:FIXFIRST). Yikes9272 (talk) 18:15, 28 September 2026 (UTC)

MSN gave me a 404 page when I clicked on it, and the archived version was blank. Since that link appears to work, that page is no longer G15-eligible and I've restored the page. The second link is not just dead, it was never archived at all, which is a sign that it was hallucinated. The article still needs review. Your pages on Nigerian Christian persecution and Litigators are full of WP:AISIGNS that require attention from the community. WP:FIXFIRST does not apply when editors are using LLMs, because it is too much work for our community of volunteers to review and repair errors made at industrial scale by bots. Dclemens1971 (talk) 18:25, 28 September 2026 (UTC)

Is that the same link that the page cited? If it cited a URL that returned 404, it doesn't matter if the MSN article exists on a different url because that wasn't what was being cited. LLMs often hallucinate the wrong URLs or identifiers, just because the article actually does exist doesn't mean that it wasn't an hallucination. If you're actually writing the article yourself, you would copy the URL from the window where you're reading it, not make up a different URL that returns 404 because you weren't reading the 404 error page. ‑‑gurkubondinn 19:07, 28 September 2026 (UTC)

Draft:Litigation Attorney (Litigator) (27 Sept) has a red-linked See also entry and several signs more associated with AITALKSIGNS, bits of it appear human-written but a Pangram positive on the end bit . Marion County Sheriff's Office (8 March, first creation) had AIDASH, AITABLE (a lot), SUPERFICIAL (The evolution of these facilities reflects the broader demographic and social changes of the city., AILEGACY (marked the most significant structural reorganization in the history of Indianapolis and Marion County), superscript unlinked refs, stopped reading 1/4 of the way down but throw in a Pangram positive for good measure Kowal2701 (talk, contribs) 18:55, 28 September 2026 (UTC)

(your article name) is... (description)
— The initial revision 1374710478 of Marion County Prosecutor's Office

Is this WP:COLLABCOMM, or is this in a new article template somewhere? ‑‑gurkubondinn 19:03, 28 September 2026 (UTC)

I'd guess WP:WIZARD, it shows up a lot in draftspace Kowal2701 (talk, contribs) 19:07, 28 September 2026 (UTC)

User:Wikibenchris

Wikibenchris (talk · contribs · deleted contribs · logs · filter log · block user · block log)

We got ourselves a 21k edit count editor here. Markdown in and , disambig links, Charlemagne Tower, Tours No sources establish a definitive chronology and exact dating of the tower's construction. is almost nostalgic slop at this point. Fermiboson (talk) 19:20, 28 September 2026 (UTC)

also triggered filter 1325 in Special:AbuseLog/45072560 (8 Sept), effectively took a break between Feb and Sept, Special:Diff/1364456175 (16 July) uses "razing" instead of "raising", this seems limited to Sept Kowal2701 (talk, contribs) 21:57, 29 September 2026 (UTC)

User:Ebubedike1

Per Talk:Abraham Darius Dillon. (Content below in case the talk page gets deleted)

I respectfully contest the speedy-deletion nomination.

This version was recreated with substantially revised prose, structure, and sourcing. It was not intended to reproduce the previously deleted article. The draft includes independent coverage of Abraham Darius Dillon's election to the Liberian Senate, his subsequent re-election, his Senate committee role, and reported legislative activity.

The article's notability is supported by significant coverage in independent reliable sources, including reporting by FrontPage Africa, The New Dawn, Liberia News Agency, Al Jazeera, DuBawa, and other Liberian news outlets, in addition to official Senate material for officeholding and committee assignments. The sources address his elections and public political activity rather than merely listing him in routine records.

I recognise the concern regarding overlap with the former version. I am reviewing the text against the deleted revision and will remove or rewrite any language that is too close. I have also removed or will remove material based solely on self-published sources, weak aggregators, or unsupported claims. I request that the page be assessed based on the present version's independent sourcing and rewritten content. If speedy deletion is not appropriate, an AfD discussion would allow the notability and sourcing to be evaluated through normal community review.

Thank you. — Preceding unsigned comment added by Ebubedike1 (talk • contribs) 20:10, 28 September 2026 (UTC)

I don't know how you'd be able to know what the old version of the article looked like considering it's deleted. The notability of the subject is unquestionable, but the current article is very flawed. There are far too many primary sources for one. It also seemed to have a lot of hallucinations, implying it was created by an LLM. As I stated in a previous discussion, the original version of the new article said "Dillon entered national politics as an administrator rather than an elected official. From 2006 to 2011, he served as Chief of Staff of the House of Representatives under Speaker George Weah." The source website did not mention George Weah, and Weah never served speaker of the House. The website said that Dillon served as chief of staff of the Speaker (unnamed on the website) from 2006 to 2007.RoundSquare (talk) 20:20, 28 September 2026 (UTC)

Would you like to share details about how you have access to the deleted version? 海盐沙冰 _{/ aka irisChronomia / Talk / Uploads} 20:27, 28 September 2026 (UTC)

* Pppery * _{it has begun...} 20:44, 28 September 2026 (UTC)

(28 Sept) cited Wikipedia, their first userspace edit triggered a filter for an LTA Kowal2701 (talk, contribs) 21:47, 29 September 2026 (UTC)

User:Nordfjording

 Nordfjording (talk · contribs · logs · block log) (28 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

 Cleanup has been requested.

User:Nordfjording has a talk page history of creations being moved to draft space because of "machine-generated text", as well as introducing non-existant template parameters. During an AFD today, the user posted one of the most obviously LLM-generated type of responses possible. I then remembered an older AFD in which concerns of LLM use arose, both in the discussion itself as well as in the article being discussed. Requesting a check since I now suspect LLM use to permeate the editorial actions of the user. Geschichte (talk) 21:39, 28 September 2026 (UTC)

Hello,

I do use LLM for grammar / formatting as is allowed. I do not use LLM for editorial actions. In the case of my response in the AFD I read the essay you referenced, and then made my points and asked LLM to fix the grammar and flow of the sentences, also in the syntax for source editing as I am not good with it. I always do my own research, and ensure that everything is referenced correctly, I never ask LLM to create "original content" as in "make an article about subject X". I am aware that my usage of LLMs to create info boxes has resulted in those non existent parameters yes.

I will commit to improving my work with less use of LLM, it has been in good faith, and to improve quality, not reduce it. This comes from a lack of technical knowledge of templates / infoboxes that I have found confusing. If you find it necessary to give me a warning for this I accept it, but I do strongly believe my LLM usage has been transparent, and within what is permitted, with some mistakes yes. Nordfjording (talk) 22:23, 28 September 2026 (UTC)

WP:AITALK, bet | One Reaction was here. Got a complaint? 11:36, 29 September 2026 (UTC)

Created Eidselva (Nordfjordeid) with a dangling chatbot block in Diff/1354376928 (AbuseLog/44220367, AbuseLog/44220363). Attempted to cite Wikipedia in AbuseLog/44610486 less than a minute before creating Ole Frøystad in Diff/1363242512. ‑‑gurkubondinn 22:26, 28 September 2026 (UTC)

Has made one edit to nowiki, no:Special:Diff/24758555, which seems like it could also have been AI-generated. 1000mm, you might want to be aware and take a closer look? ‑‑gurkubondinn 22:29, 28 September 2026 (UTC)

Could be, but no really clear and obvious signs, as far as I can tell. Two years ago, some minor edits have been made with part of the added text afterwords. Only one reference though, at the very end of the added paragraph. However, with a brief look at the source, it does cover at least most of the added text. So I’ll say this one is OK. 1000mm (talk) 22:45, 28 September 2026 (UTC)

Agreed, there's no really clear and obvious signs and the source seems to check out. ‑‑gurkubondinn 23:04, 28 September 2026 (UTC)

@Gurkubondinn I see you have added tags to some of the articles I have edited, identifying LLM usage. Is it appropriate for me to clean-up - or is this something that must be done by a independent editor? I am more than happy to help out if permitted, and I wish to reiterate that I have checked all citations before publishing. Nordfjording (talk) 20:38, 29 September 2026 (UTC)

We are always happy when someone cleans up after themselves. Just look at the size of this board, there is more to do than the handful of people that are active here will ever be able to do. So we would of course be extremely appreciative if you would do it. Just don't use AI for it, that would defeat the purpose. ‑‑gurkubondinn 20:42, 29 September 2026 (UTC)

@Gurkubondinn Thank you, my goal has always been to improve the Wiki not create more work for people. Now that I am aware of the policy after reading up on the AI related essays / polices I will strongly limit my use of LLM.

For the clean-up I will rewrite where relevant (without AI), and I will double check all sources for those sections. Nordfjording (talk) 21:00, 29 September 2026 (UTC)

"Strongly limit" your LLM use? Can you be more specific? Masterhatch (talk) 21:10, 29 September 2026 (UTC)

@Masterhatch I will not use LLM to create text for a article. I will only use it for matters of grammar / and finding sources as a supplement to search engines (not meaning copy text LLM generates, but ask it to find sources with links to improve an article, and check those sources myself , not blindly trust it).

I assume that is okey? I have as now admitted a few times used it for syntax and flow, and will not repeat this mistake in the future. I am currently combing through the relevant articles and looking for any signs of hallucinations (re verifying sources) and stripping out any promotional / over enthusiastic language etc. Nordfjording (talk) 21:31, 29 September 2026 (UTC)

Honestly, I don't know why you need it for grammar. There is no wiki policy that states your grammar must be perfect. Masterhatch (talk) 21:39, 29 September 2026 (UTC)

That is a fair point. Regardless I understand why this became an issue, and I am taking steps to rectify my transgression while educating myself further on Wiki policies. Nordfjording (talk) 21:43, 29 September 2026 (UTC)

I have now completed the article clean-ups, I will look into the remaning Nordfjordeid in depth tomorrow, but have so far cleaned out some enthusiastic LLM phrasing in the leed. Nordfjording (talk) 00:14, 30 September 2026 (UTC)

User:Ipflo

 Cleanup has been requested.

Ipflo (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Confirmed AI use in Special:Diff/1357471727 (while removing the tag correctly indicating their AI use), seems to be ChatGPT specifically per Special:Diff/1320929878 params. Given that they said they used AI to to make text fluent, which suggests it wasn't fluent before and thus violates WP:LLMTRANSLATE; given that I found this article by searching the usual AI slop phrases that are, uh, abundant in the linked article, and given that their idea of cleanup is to "[remove] AI-tic phrasing" (as opposed to not using AI at all), alot of stubifying and/or LLMPROD is probably necessary here. Gnomingstuff (talk) 02:37, 29 September 2026 (UTC)

Gnomingstuff, you seem to have given this editor WindBourneListener's tracker, was that a copy-paste error? M kuhner (talk) 02:42, 29 September 2026 (UTC)

My mistake, fixed. I have never been able to get the automated new section creator to work properly. Gnomingstuff (talk) 05:45, 29 September 2026 (UTC)

triggered filter 1325 for strong AI signs in Jan and Feb 2025, chatgpt params in Nov 2025 , Hatvanpuszta (Jan 2025, not a translation) had SUPERFICIAL ("played a pivotal role"), AIATTR, AIDASH etc. Creations throughout 2024 use hyphens instead of dashes (a human sign). Special:Permalink/1325913581 (Dec 2025) had Today, Easton Park survives primarily in documentary records, surviving estate buildings, the remnants of the crinkle-crankle wall, and local memory. Its history remains closely linked to the Wingfield, Rochford, and Hamilton families, whose occupancy reflected broader developments in British political, diplomatic, and social history. plus lots of signs

Schloss Gehren (May 2026, not a translation) had AIREDCAT, AIDASH, plausibly-AI prose, Pangram says mixed/mostly human-written , Schloss Kreuzwertheim (4 August, not a translation) had AIATTR (The plot on which the palace stands has a documented legal history predating the current building...), reads human though. Seems to have mostly stopped some time in early/mid 2026? Kowal2701 (talk, contribs) 21:42, 29 September 2026 (UTC)

User:Agus Damanik

 Agus Damanik (talk · contribs · logs · block log) (29 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Triggering the LLM paste check on Borobudur in Diff/1377054651, Diff/1377305080, and Diff/1377373027. Reinstated their edits when reverted, claims grammar correction in § September 2026 on their talk page. Shows a very different level of English in talk page messages compared to the level of English in article edits. Cleanup is probably going to be required. ‑‑gurkubondinn 08:41, 29 September 2026 (UTC)

I will rewrite the context of my rejection of the revert for my edits below. I already wrote my dissatisfaction for how this WP:NOLLM was used for punishing people rather than giving direction for how to do good edit in Wikipedia.

"the policy of NOLLM guidance was created to avoid content where it violate WP:CCP. Then tell me, what kind of violation that i violate from the policy. You said that you aware of the guideline, it literally said that you need to consider the WP:CCP before you take decision to revert my big edit. "

I also believe @Gurkubondinn also violate WP:ROWN where he did a revert for my big edit without doing any good feedback for text that i incorporate and i assume he wouldn't have enough to give the feedback. I ask again for the use of the policy, what harms that it will make if the content in the text itself is not hallucination and used for grammar correction only. Once again, give evidence where the content itself violate CCP where the core of the policy itself is to avoid violation of CCP rather than nuke on all text that use LLM on their creation. Agus Damanik (talk) 09:13, 29 September 2026 (UTC)

The core content policy being violated is that no AI generated text is allowed on Wikipedia. If you do not understand that then you cannot edit here. Fermiboson (talk) 09:16, 29 September 2026 (UTC)

Thank you for the very good word that i didn't expect coming from someone that wanted the best for wikipedia by saying i can't edit in wikipedia. I am just realizing that the core problem is that Wikipedia doesn't allow LLM where the prompt is rewording the sentences by changing the choice of words and structures. This problem is caused by not very specific prompt where the LLM will tend to give recommendation thought we just ask only grammar correction for the text. So now i understand what's the core problem for my sentence after i take a look at WP:USEDAI?. I have reworked the sentence with my own word, but i still believe the revert of the big edit is contraproductive cause i can't work on what's the problem.

Rather than only WP:LLM, you can also give WP:USEDAI? to give anone that was accused for more context. I also see that @Fermiboson always coming on the same case with Gurkubondinn where i have not seeing any good intention on what to fix for the problem for any case that involve both of you. So i wil stop using any AI for any addition for the article in english wikipedia and i hope everyone can be more civil and don't forget about WP:BITE. Agus Damanik (talk) 09:57, 29 September 2026 (UTC)

Now that you are aware of the WP:NOLLM guideline, can you self-revert the changes to Borobudur that you restored after I reverted them? ‑‑gurkubondinn 11:31, 29 September 2026 (UTC)

I already changed the structure and it's already different from the way before your first revert and not using any LLM for that including any grammar correction. You can revert it if you think the new paragraph is still created by LLM cause I rest my case and not gonna touch that article anymore if you have reverted it because I prefer someone other than you two to take a look at the article and decide to revert it cause the other user said that I can't edit here. It's better if the user is an admin. Thank you Agus Damanik (talk) 12:55, 29 September 2026 (UTC)

The changes in Diff/1377308160/1377415824 are pretty minimal (considering this is c. 2,8 kB of text):

−

Then, Jan Lodewijk Moens proposed a hypothesized ritual connecting the three temples.

+

Jan Lodewijk Moens hypothesized a ritual connecting the three temples.

−

However, the exact process remains unknown due to a lack of surviving written records.

+

However, the exact process remains unknown due to a lack of written records about the ritual.

−

He hypothesized that the ritual was aligned with the [[sun path]], envisioning Borobudur as the sunset, Pawon as the zenith, and Mendut as the sunrise, where the purpose of both the ritual and the buildings was to [[Sacredness|consecrate]] the king as a Buddha.

+

He hypothesized that the ritual was done by doing [[pilgrimage]] following the [[sun path]] where it starting with Borobudur as the sunset following with Pawon as the zenith and ending with Mendut as the sunrise.

−

[[Giulio Magli]] furthered this theory by saying that the sun path was identified following the event of the zennith passage which he identifed had occur on October 11 and February 28.

+

[[Giulio Magli]] also identified that the ritual is following the event of the zennith passage that happening on October 11 and February 28.

This one has grammatical errors both before and after.

−

He also compared Borubudur with [[One Hundred and Eight Stupas]] which followed the [[winter solstice]].

+

He also compared Borubudur with [[One Hundred and Eight Stupas]] where the building itself is built around ritual related to the [[winter solstice]].

−

These three templed was listed in 1991 as [[world heritage site]] by [[UNESCO|Unesco]] as Borubudur Temple Compounds.

+

In 1991, the three temples was designated as [[world heritage site]] by [[UNESCO]] under the name Borubudur Temple Compounds.

Also has incorrect grammar in both versions.

−

Murwanto et al stated again that Borobudur was built above the floor of a dried-out [[Lake Borobudur|paleolake]].

+

Murwanto et al reiterated the theory from Nieuwenkamp that Borobudur was built above the floor of a dried-out [[Lake Borobudur|paleolake]].

−

Subsequent research in 2015 indicated that the lake existed from the [[Late Pleistocene]] to the Late [[Holocene]] and gradually disappearing due to volcanic and tectonic activity.

+

In 2015, the lake was hypothesized to be existed from the [[Late Pleistocene]] to the Late [[Holocene]] before disappearing due to volcanic and tectonic activity.

−

In 2023, a study confirmed the existence of the lakebed by identifying an alluvial deposit located between 50 and 525 meters below the surface of the Kedu Plain..

+

In 2023, the existence of ancient lake was strengthened based on finding of the alluvial deposit that located between 50 and 525 meters below the kedu plain.

I'm honestly not sure what to think though, since both versions have grammatical errors. ‑‑gurkubondinn 13:21, 29 September 2026 (UTC)

The English Wikipedia does not accept AI-generated material, it is prohibited by the WP:NOLLM policy. I reverted your edits because you submitted AI-generated material. ‑‑gurkubondinn 09:19, 29 September 2026 (UTC)

User:Spelare108

Cleanup efforts are ongoing and being tracked at /2026-09-29 Spelare108, 47 pages need review.

LLMPRODs are done and drafts tagged

Spelare108 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Was asked politely to "please stop using AI," and instead of stopping, summoned an army of sealions to make every disingenuous argument in the book. Unfortunately for their argument, Special:AbuseLog/44123060 contains WP:OAICITE. (So much for I am obsessed with not making errors, so I will pore over each composition again and again. I’m a bit of a perfectionist, even to a fault.) And then obviously they continued using AI.

So yeah, enormous amount of cleanup required. The edit summaries here claim that the edits are translations, but they seem to be adding content not contained in the supposedly translated articles (for instance, Special:Diff/1367879651 is clearly not a translation of the French article it says it is, because the French article does not mention Hegarty or Gordon anywhere). Gnomingstuff (talk) 09:22, 29 September 2026 (UTC)

creations are from April 2026 onwards, OAI cite was in May, markdown in August . First creation was Special:Permalink/1351678538 (29 April, already AI-tagged) which was allegedly translated from frwiki, except it's way longer than the French article, gems like Although long associated with neo-Scholasticism, later scholarship has situated Gardeil within a mediating current between Modernism and anti-Modernism and as a precursor to aspects of twentieth-century theological renewal., AIREDCAT, AICURLY, doing LLMPRODs Kowal2701 (talk, contribs) 19:30, 29 September 2026 (UTC)

created a tracker from 29 April 2026, appears to be when LLM use began Kowal2701 (talk, contribs) 19:44, 29 September 2026 (UTC)

User:AFeatherlessBipehead

 AFeatherlessBipehead (talk · contribs · logs · block log) (29 September 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Overwrite the Chris Cantelmo redirect in Diff/1377429419 (AbuseLog/45249579) with text containing Markdown for italics (but also wikitext syntax for the same), denied using an LLM when warned in User talk:AFeatherlessBipehead § September 2026, and then unilaterally restored the edits in Diff/1377431498 with another blanket denial of AI use.

At least two other articles have been flagged for AI use by Gnomingstuff:

• Murder of Adriana Naghei Ostrowska (tag removed in Diff/1367312820)
• Skellefteå assault case (tag removed in Diff/1367984445), created in Diff/1325551997 with two AIREDCATS.

‑‑gurkubondinn 10:17, 29 September 2026 (UTC)

Yeah I didn't use an LLM to write it..?

As for the other two articles, if I recall I used an LLM to translate the Swedish version of the articles to English, but that should be it. AFeatherlessBipehead (talk) 10:21, 29 September 2026 (UTC)

If they are (how are you not sure?), then both fail WP:LLMT and are unattributed. The Skellefteå translation is inaccurate and pretty bad. No comment on the other article, because I haven't read the svwiki article. ‑‑gurkubondinn 10:34, 29 September 2026 (UTC)

Yeah I should have tagged it as an AI translation and given attributions, sorry about that. AFeatherlessBipehead (talk) 10:41, 29 September 2026 (UTC)

Then why have you been removing the {{ai-generated}} tag if they are AI-generated? Providing attribution is a legal requirement of the CC-BY-SA licence and a requirement of the WMF terms of service, this needs to be fixed. Attribution is not optional. ‑‑gurkubondinn 10:56, 29 September 2026 (UTC)

My reasoning was that, since I had only used AI to translate the articles, and since none of the information (that I saw) was false, that the tag was incorrect, in retrospect I obviously shouldn't have just removed them. AFeatherlessBipehead (talk) 11:03, 29 September 2026 (UTC)

The sources on the Skellefteå article are also all messed up in Diff/1325551997 (the svwiki wiki article has 19 sources, the enwiki article was created with 21 sources). The titles have also all be translated without providing the original titles in Swedish. One of those translations is kind of POV pushing, Högerextrema sajter sprider bild på misshandlad flicka became Far-right sites spread image of assaulted girl. The problem here is that "högerextremism" means "right-wing extremism" in English, "far-right" would be "extremhöger" in Swedish. It is a subtle difference, but it makes the wording a bit softer, and its particularly problematic because trans-title was not used so the original titles were not preserved in the title parameter like they should have been. ‑‑gurkubondinn 10:52, 29 September 2026 (UTC)

I see, I was a way less experienced editor when I created the article which is probably why those mistakes could pass over my head at the time... AFeatherlessBipehead (talk) 10:55, 29 September 2026 (UTC)

User:Help.Football.History

 Help.Football.History (talk · contribs · logs · block log) (29 September 2026)
Editor has been blocked from editing the mainspace (articles only) for AI/LLM violations. 

 Cleanup has been requested and is being tracked at /2026-09-30 Help.Football.History, 938 pages need review.

Help.Football.History (talk · contribs · deleted contribs · logs · filter log · block user · block log) is a prolific editor who has created 67 mainspace pages and appears to be using LLMs extensively. For example: On Draft:Iberia Tenerife, we have WP:AIREDCAT in the first draft and WP:AILEGACY (Iberia quickly became one of the leading clubs in Tenerife) unsupported by the source. On Draft:Sporting Club Tenerife, we have a non-existent URL, AIREDCAT, and repeated vague WP:AIATTR to "historical Tenerife sources". Since the user has already been warned twice ( , ) (and denied LLM use, a denial contradicted by all evidence), I blocked from mainspace to prevent further disruption, but their extensive work needs a closer look. Dclemens1971 (talk) 18:21, 29 September 2026 (UTC)

chatgpt params in September 2026, lots of non-existent templates , Special:Permalink/1312264784 (Sept 2025) had sources retrieved in 2022 (maybe copied from another article?) Kowal2701 (talk, contribs) 19:22, 29 September 2026 (UTC)

The editor's block has been extended by Yamla to sitewide following an AI-generated block appeal. Dclemens1971 (talk) 21:56, 29 September 2026 (UTC)

Tracker created, ignoring edits under +/- 35 bytes, which took it down from >3000 pages to 938. Honestly, from a quick check, I think a lot of the slightly larger edits are unnecessary too: lots of redirects created, and formatting fixes for tables of results that are easy to check and (imo) would be easier to do by hand than with an LLM. (But what do I know? There's very little it would occur to me to use an LLM for in the first place.)

This is very much not my topic area, so if someone else spots a good cut-off date or a better cut-off edit size, please feel free to update the tracker or ping me for it. Suðurhafsljósæta (talk) 09:51, 30 September 2026 (UTC)

Having read the repeated LLM block appeals, I feel even more out of my depth here... let's leave the smaller edits in for now. Suðurhafsljósæta (talk) 09:58, 30 September 2026 (UTC)

User:Sophiabennett

Cleanup efforts are ongoing and being tracked at /2026-09-29 Sophiabennett, 1 pages need review.

LLMPROD is not needed in this case

 Sophiabennett (talk · contribs · logs · block log) (29 September 2026)
Editor has been blocked from editing the mainspace (articles only) for AI/LLM violations. 

Paid editor persistently engaging in AI use at Draft:Kuo Zhang. They've engaged in a bunch of disclosed paid editing, probably all of it needs to be checked, especially anything in mainspace. I have already reported them to ANI at WP:ANI#Paid editor continuing to use AI after multiple warnings. Athanelar (talk) 19:08, 29 September 2026 (UTC)

Note: Partial block to mainspace was applied by Dclemens1971. —Tenryuu 🐲 ( 💬 • 📝 ) 19:30, 29 September 2026 (UTC)

• created tracker, tagged drafts Kowal2701 (talk, contribs) 23:33, 29 September 2026 (UTC)

  weirdly, a lot of their mainspace edits are saving declined AfC submissions, I could only tag those Kowal2701 (talk, contribs) 23:42, 29 September 2026 (UTC)

  stumbled across HRShami (talk · contribs) who works for a "Wikipedia consulting business" and does mostly paid editing AFAICT, one of their creations was tagged, loads of chatgpt params , worth looking into Kowal2701 (talk, contribs) 23:48, 29 September 2026 (UTC)

  yep that's another longtime ~2024ish AI editor Gnomingstuff (talk) 02:22, 30 September 2026 (UTC)

User:Vanderwaalforces (again)

(this was going to be an ANI thread but I decided to take it here instead because the situation does not make any sense, so sorry about the overexplaining below, it is written for an audience that doesn't do AI cleanup)

tl;dr: Vanderwaalforces added a lot of AI text to Wikipedia in 2023-2024 (which they have admitted to) and said that they stopped, but I don't actually think that they have stopped.

An AI noticeboard thread was opened earlier this month, here Wikipedia:AI_noticeboard/Archive_19#User:Vanderwaalforces. The noticeboard post speaks for itself, but I do want to reiterate this: we see a writing style that changes in lockstep with LLM style and capability. Their 2023 stuff (which they said was indeed AI) reads like 2023 AI output. Their 2024 stuff (which they also said was AI, at least some of it) reads like 2024 AI output, and so forth. There are some examples in the AINB thread, but here are a few more:

• 2023: Special:PermanentLink/1171654802: Blatant promotional tone (later LLM output is less blatant by default), superficial analyses (all those participials tacked onto everything), 2023-mid 2024 AIVOCAB (stuff like diverse cultural tapestry basically disappeared when GPT-4 did), what appears to be a knowledge cutoff disclaimer, etc. All straight down the line for 2023 LLM output and there are many more examples.
• 2024: Special:Diff/1253797894: More subtle but still promotional tone, same with the participials, 2024-mid 2025 AIVOCAB density, etc. Another example: Special:PermanentLink/1259216139, the rigid outline-like structure about challenges and future prospects that emerged around 2024. Again, all very typical and not unusual LLM output from the time.
• They didn't create much text in 2025, it's basically just one article, here it is. I'm actually not sure whether AI was used here or not (they have said it was not), but the takeaway is that the writing style is now quite different than either of the writing styles above.

Fast-forward to mid-2026. Around this time, LLM output suddenly became obsessed with providing disclaimers about how sources should be used, kind of a blurring of WP:AIDISCLAIMER and user metacommunication about the article that basically did not exist in AI-generated text until then. This stuff has absolutely popped off in the past few months. You can see some examples at Wikipedia_talk:Signs_of_AI_writing#2026_negative_parallelism/didactic_disclaimer/source_disclaimer_variant, and I encourage you to do that, because it'll become clear how rigid the formula is. There are two basic patterns (well, three that I've noticed in general, but two relevant here):

• "[Source/claim X] should be treated as [speculative/subjective thing] rather than [empirical/factual thing]"
• "...although [source/claim X] is not [some variation of documented in independent/reliable/verifiable sources]" (similar to WP:AIDISCLAIMER, except what's being said is "the claim can't be confirmed" instead of "the information is not available")

Again, this kind of disclaimer basically did not exist in AI-generated writing before mid-2026, and weren't common in human writing either (as usual, things similar to it existed, but not this specific thing at this volume). You also do not see anything like them in Vanderwaal's diffs above in any year, despite the subject matter being similar; and I haven't seen it in any of their other older diffs I've checked either. Come 2026, though...

• Special:Diff/1360004101: These accounts reflect Benin interpretations of political and ritual causation rather than a medically documented cause of death.; and The proposal rests on the stylistic discontinuity and an early European report of executions following Ogbebo's defeat, rather than on direct documentation concerning individual artists; and Such explanations belong to the oral and religious interpretation of the conflict and are not independently established by contemporary documentation.
• Special:Diff/1358323408: The dating of Ewuare's reign depends on oral tradition and later historical reconstruction rather than surviving contemporary records; and the narrative of a legitimate elder brother and a younger usurping brother may preserve claims about dynastic legitimacy rather than serve as a straightforward factual account.
• Special:Diff/1370236179: Academic treatments therefore generally approach the competing traditions as evidence requiring comparison and interpretation rather than as a single settled narrative.,
• Special:Diff/1366388901: although this remains an inference rather than a directly recorded fact, (I stopped reading here, you get the idea)

The only real difference in the pattern is that in these edits, the disclaimers are often attributed to sources. I don't know whether this is something the (hypothetical) LLM is doing -- in which case it'd be similar to what happened with WP:SUPERFICIAL -- or something that is being tweaked before the text is added, but it has the same verification issues where the source is only vaguely related to the text. For instance, Nzegwu maintains that Benin's past should be reconstructed by combining oral accounts, annual ceremonies, ritual performances, visual art, African conceptual frameworks and written sources rather than relying primarily on European documents is cited to pages 34-36 of this, which is making a much more specific argument.

There are other suspicious things, like some WP:AIREDCAT (e.g., Special:Diff/1357839320; "2008 art exhibitions" is a category now, but it wasn't at the time), various other mid-2026 AI text indicators also appearing (e.g., the "identifies X as Y"/"X is associated as Y" tic in Special:Diff/1356562287, which again is a new writing tic), filter log full of harv/sfn errors (not by themselves indicative of LLM use, but frequently accompanies it). I don't normally use AI detectors, but Pangram is consistently flagging these as AI-assisted: example 1, example 2, example 3, example 4. (Note that Pangram produces much more false negatives than false positives.)

So there are three basic possibilities here:

• They aren't using AI anymore and just coincidentally happened to pick up one of the top three tics of mid-2026 LLM output, at the same time that LLMs started to produce it en masse. I don't find the "I just write like that" explanation convincing because, well, they didn't write like that until 4 months ago.
• They're using some kind of Grammarly/DeepL editing tool and are not aware of the fact that it is LLM-based. I don't know how you get source disclaimers like those if you're just copyediting, but I'm trying to run through all the good-faith possibilities here.
• They're using AI again and lied about it. Which is actually worse than the 2023-24 stuff, because in 2023 and 2024 there was no guideline against AI-generated content on Wikipedia, and in 2026 there is -- and Vanderwaal is definitely aware that there is because they've reverted people for it. Also worse because it would mean we have non-compliant FAs and GAs.

Gnomingstuff (talk) 01:45, 30 September 2026 (UTC)

Noting that I’ve read the above and I’m both overwhelmed and will probably reply later if I’m able to. Gnomingstuff have always had a conviction that I’ve been using AI to write articles till date. They mention red cats as probably signs I write with AI whereas that is something I do almost every time.

I’m wondering though, what is AI-assisted? Have you found a source misrepresentation (which AI is fond of?), fwiw I’m not saying you won’t find errors in my articles.

I’m overwhelmed and not in the right state of mind to reply further. Vanderwaalforces (talk) 05:10, 30 September 2026 (UTC)

I’m wondering though, what is AI-assisted?

Why are you asking me this? You are the one who knows whether you did or did not use AI. Gnomingstuff (talk) 08:02, 30 September 2026 (UTC)

User:Hyderabad ki Awaaz News digital media

Y Cleanup has been completed.

This user has pasted what seems to be a AI generated summary of their company formatted as an article. They accidentally pasted the summary twice to their talk page which I undid and it's also on their user page. It follows common AI practices like dot points and bolding which makes me believe it is AI generated. TobyKohlhagen (talk) 08:39, 30 September 2026 (UTC)

Blocked as a sockpuppet, user page deleted. Less an AINB case than a run of the mill spam case. CMD (talk) 09:04, 30 September 2026 (UTC)

Thank you so much, apologies for the incorrect classification. It's my first time really ever reporting it so I'm a little bit rusty. In the future what page do I go to to report spam? TobyKohlhagen (talk) 09:15, 30 September 2026 (UTC)

It's not a problem. If you don't report here, you can report obvious spam to WP:AIV. Obviously promotional usernames can go to WP:UAA. CMD (talk) 10:56, 30 September 2026 (UTC)