Wikipedia talk:WikiProject Articles for creation

Main pageTalk pageSubmissions
Feed, Category, SortingShowcaseApply
Participants, By subjectReviewing instructions
Help deskBacklog
drives

Discuss the Articles for Creation project

AfC submissions Random submission

~5 weeks

2,638 pending submissions

Are you in the right place?

• If you want to ask a question about your draft submission, use the AfC Help desk.
• For questions on how to use or edit Wikipedia, use the Teahouse.
• Create an article using Article wizard or request an article at requested articles.

Start a new topic

• In addition to this page, you can give feedback about the AFCH helper script by creating a new ticket on GitHub.
• New to Wikipedia? Welcome! Ask questions, get answers.

New field for AI use disclosure at AfC Submit Wizard?

I've been thinking of how the AfC Submit Wizard could use additional field(s) to help submitters disclose their AI usage.

It would only help honest people be honest, but I think it would improve our understanding of how (new) editors use AI and help reviewer feedback be more specific than what we have currently.

I've drafted potential questions at Wikipedia:WikiProject Articles for creation/Submission wizard/AI use disclosure. I'm interested to see if this is a viable idea. Ca ^{talk to me!} 10:27, 1 July 2026 (UTC)

Thanks! I think this is worth a try. Newystats (talk) 12:11, 1 July 2026 (UTC)

I really like this idea! I suggest limiting to a single key question though, whatever the reviewer needs the most - in this case, just the first question. In general, fewer questions increases the chances that people will fill something out. Dreamyshade (talk) 23:12, 3 July 2026 (UTC)

I agree with knocking a question off or two. However, I don't think keeping only the first is a good idea. From my experience, most editors, when asked about their AI use, claim they used only used AI for copyediting or formatting when its clearly untrue. The latter two question are more evidence-based, rather than self-report-based. Ca ^{talk to me!} 13:48, 7 July 2026 (UTC)

I've reduced the number of question to just one. 👍 Ca ^{talk to me!} 10:54, 13 July 2026 (UTC)

mmm, the results would probably be useful, but it might give people the wrong impression that LLM use is allowed, and I'm not really comfortable w using newbies as guinea pigs like that. Idk if there's a way to address that? Kowal2701 (talk, contribs) 20:53, 4 July 2026 (UTC)

The WP:Article Wizard already informs participants that LLM generation is disallowed, so I'm not too worried about giving the wrong impression. Technically, AI use for "basic copyedits and translation" is currently allowed, so I tried to orient the example answers to follow that limitation.

I agree we shouldn't conduct systematic research with the disclosures without them being informed about it, but that wouldn't be its primary goal. I think its main use case would help reviewers distinguish between cases where drafts were wholly AI-generated and cases where it was just simply rewritten with AI. Both are disallowed by policy, but one is salvageable and one is not, helping reviewers provide more specific feedback.

I'd also add that frequent exposure to known AI-generated text has been shown to improve AI-detection abilities. Ca ^{talk to me!} 13:44, 7 July 2026 (UTC)

that sounds good then! Kowal2701 (talk, contribs) 21:46, 14 July 2026 (UTC)

Ready for implementation?

Tracked in github.com
Issue #10

Submission Wizard screenshot with additional AI disclosure field

Based on on- and off-wiki feedback (much appreciated), I've...

• reduced the question count from three to one
• replaced the request for prompts with one for transcripts (more comprehensive)
• and restyled the field to match rest of the wizard

I'm wondering if the community believes the current proposed field is ready for implementation. Ca ^{talk to me!} 09:32, 22 July 2026 (UTC)

Great! I support it. I'd just move "If you used another AI tool like Grammarly or Google Translate, or no longer have the transcript, simply describe how you used AI." before the input field instead of after, since it helps people understand what to put into the input field. Dreamyshade (talk) 09:55, 22 July 2026 (UTC)

Good idea— Done Ca ^{talk to me!} 02:56, 23 July 2026 (UTC)

This looks good to me! In solidarity, nil ^nz 03:59, 23 July 2026 (UTC)

I've opened a pull request on Github. Ca ^{talk to me!} 13:15, 25 July 2026 (UTC)

@SD0001, since you appear to be the maintainer, would you be interested, at your own leisure, in reviewing the PR? I think it would be worth trialing the new disclosure during the August backlog drive. Ca ^{talk to me!} 13:42, 31 July 2026 (UTC)

Hoping to see this PR reviewed and integrated! I'm really curious to see what people's prompts are. I think this will also help us detect LLM-generated articles more effectively as newer models come out, because we can re-run people's prompts with the new models and compare the outcomes. Dreamyshade (talk) 18:38, 17 August 2026 (UTC)

I don't think SD0001 is available right now. @Novem Linguae, would like to handle the review? Ca ^{talk to me!} 13:49, 19 August 2026 (UTC)

I'm also pretty busy right now. And this is not a repo I normally review so would take effort to get up to speed. Sorry. –Novem Linguae (talk) 01:13, 22 August 2026 (UTC)

Ditto, and my parents are in town this weekend so I'll try to take a look but no guarantees on if/when I get to it. Primefac (talk) 11:11, 22 August 2026 (UTC)

I would support this proposal. I don't see how this could do any harm, only possible good. Axolitl (talk | contribs) 22:58, 2 August 2026 (UTC)

I support this, as it would be a clear, explicit box for users who are potentially unaware of the policy to disclose their AI use. I believe that this is ready for implementation. BSH🌸 (talk) - (she/her) 14:15, 3 September 2026 (UTC)

This is live now. Check it out on WP:SUBMIT. – SD0001 (talk) 04:07, 29 September 2026 (UTC)

This is great, thank you. qcne (talk) 08:08, 29 September 2026 (UTC)

Support from me too. At least to see what happens, and it remains the case that many AFC reviews, at least on the Help Desk, claim to be unaware of the restriction. But let's give it a go and assess after a few weeks. Not sure what happens for those AI packages which directly create drafts "oven-ready" for AFC. ChrysGalley (talk) 06:25, 29 September 2026 (UTC)

• If anyone wants to see this new process in operation, here is a rather clear example:

Draft:Dr. Maulana Maqsood Ul Hasan Qasmi

I would ordinarily decline as AI, but it is also insufficient notability in terms of sourcing, and there are (fixable) NPOV issues too. ChrysGalley (talk) 14:33, 29 September 2026 (UTC)

Does it miss the typical new line break and horizontal line ----? See this example here for Draft:Shiv Sharda Public School. Courtesy ping @Novem Linguae and SD0001. Sorry if this was just an editor quirk, this is just the first time I've seen it. Bobby Cohn 🍁 (talk) 16:13, 1 October 2026 (UTC)

Now what?

Having seen a number of drafts with these AI disclosures, I now realise I don't quite know what I'm meant to do with them. Is the idea that the reviewer checks what AI has been used for and whether that's within the permitted use? And if it isn't, then what? For example in Draft:RE24, AI was used for pretty much everything, therefore... that's fine, and thanks for sharing? Straight to G15 speedy? Go over it with a fine-tooth-comb to check for hallucinations or other issues? In other words, what should I do, that I wouldn't do if that disclosure weren't there? --DoubleGrazing (talk) 07:12, 30 September 2026 (UTC)

Well I suppose it's more for the submitting editor. I declined the example above, but I did feel bad about doing so: here the editor was completely transparent, painfully so, so all I can do is decline (I've had so many G15 turned down that I no longer do that) but at least the editor can see (a) what they said and (b) my hopefully helpful response. ChrysGalley (talk) 07:50, 30 September 2026 (UTC)

Are people able to see transcripts without logging in? Tried the link in Draft:Recipe markup language and it sent me to a log-in page; I have no desire to make a Claude account. SomeoneDreaming (talk) 23:43, 1 October 2026 (UTC)

It depends on the platform and possibly on a user's individual settings. In this case, I couldn't access the session from my free Claude account, probably because it's a Claude Code (coding agent) session, not a plain chat. (The session link starts with https://claude.ai/code/, not https://claude.ai/share/.) —ClaudineChionh (she/her · talk · email) 23:54, 1 October 2026 (UTC)

Changing the AI decline notice

The current text is as follows;

This draft appears to contain text generated by a large language model (such as ChatGPT). You cannot use LLMs to generate article content.

LLM-generated pages with certain obvious signs of being machine generated may be deleted without notice.

These tools are prone to specific issues that violate our policies:

• hallucinations: they often invent false information and cite non-existent references.
• unencyclopedic tone: they tend to be vague, promotional, or essay-like, rather than neutral and factual.
• copyright issues: they may closely paraphrase existing text, leading to copyright violations.

Instead, only summarize in your own words a range of independent, reliable, published sources that discuss the subject.

See the advice page on large language models for more information.

The section beginning These tools are prone to specific issues that violate our policies is misleading, as it implies that LLM-generated text which does not contain these issues may be acceptable. In any case it's unnecessary; we don't need to give people a screed on why LLM generated text is bad, the only part which is relevant to their article being declined is that it's not allowed. Let's keep it short and sweet and minimise the room for misinterpretation (especially because AI chatbots love to misinterpret this kind of wiggle-room and whisper like Grima Wormtongue into their users' ears that AI usage is okay if it doesn't have these issues.) Athanelar (talk) 05:59, 27 July 2026 (UTC)

I agree. I have had pushback on that on my Talk page, since the LLM editor then says "there's nothing hallucinated here, I've checked everything." One editor wrote (paraphrasing), "OK, so yes I used LLM, but you are asking me to do a rewrite for a rewrite's sake, since all the sources and text wording are OK?" The answer is yes. But because the decline focuses on what could go wrong, a mini justification on our stance about LLM, it's actually not the point: the point is LLM is against Wikipedia's policies and guidelines, just like many other issues: copyright, non neutral text, unsourced. By all means give a one line link to the justification, but I don't see the point in deflecting off the core point, which is that LLM is not allowed. ChrysGalley (talk) 07:11, 27 July 2026 (UTC)

I've noticed that most editors who resubmit after an LLM decline just superficially reword the LLM output (random recent example: ). I have been adding a comment to preemptively discourage this (Please remove all text from the draft and start over from a blank slate without LLMs) but it would be better if the notice was rewritten to address this. Helpful Raccoon (talk) 04:36, 3 August 2026 (UTC)

I agree, that would be good wording to include Newystats (talk) 06:31, 17 August 2026 (UTC)

Proposed AI decline notice rewrite

I've wrote up a potential rewrite of the decline notice incorporating suggestions above as well as encouragement to WP:LLMDISCLOSE. Rationale are annotated with refs. Curious to see community opinion on this.

This draft appears to contain text generated by an AI model (such as ChatGPT). You cannot use AI to generate article content.

Instead, only summarize in your own words a range of independent, reliable, published sources that discuss the subject. If the draft is entirely AI-generated, you should start over from scratch.

If an AI chatbot assisted with this draft, it is recommended to share the chat transcript when resubmitting. You can do this by

• sharing a link to the chat (see the how-to guide), or
• by manually copying and pasting the transcript in this draft's talk page.

Reviewers can make mistakes. If you have not received any AI assistance, you may appeal at the AfC Help Desk.

See the advice page on large language models for more information.

Ca ^{talk to me!} 08:50, 7 September 2026 (UTC)

Courtesy pings: User:Athanelar, User:ChrysGalley, User:Helpful Raccoon, User:Newystats Ca ^{talk to me!} 08:51, 7 September 2026 (UTC)

I would remove the LLMDISCLOSE part. It risks implying that AI use is acceptable if disclosed. Athanelar (talk) 08:56, 7 September 2026 (UTC)

I agree LLMDISCLOSE is unhelpful here, indeed my concern with the proposed wording is that it seemingly gives off-ramps to LLM use, rather than just saying LLM use, apart from 2 narrow exceptions, is not allowed under P&G. ChrysGalley (talk) 09:00, 7 September 2026 (UTC)

The inevitable thing that's going to happen is that they'll resubmit an AI rewrite with the chat transcript included, then when they're told it's not allowed we'll get the "but the decline notice told me to do it!" and then we'll have to reply "yes, but it doesn't say it's allowed because you did it!" which just comes off more like a "neener-neener, got you on a technicality!" which is just going to create more frustrated belligerent AI users for us to deal with at the AFCHD.

This decline notice needs to be short and sweet; the only thing the submitter needs to know is that their draft seems to be AI generated and that that isn't allowed in 99% of cases, if they think we're wrong or they're covered by one of the exceptions then they should tell us. Athanelar (talk) 09:07, 7 September 2026 (UTC)

Support the above. ChrysGalley (talk) 09:15, 7 September 2026 (UTC)

Good point. The LLMDISCLOSE part would need wordsmithing, if not cut altogether if no desirable wording is found.

However, I believe including LLMDISCLOSE in some form has several beneficial side effects and is worth exploring:

a) Frequent exposure to known AI-generated text has been shown to improve one's detection ability. A lot of articles flagged at WP:AINB were actually accepted via AfC by experienced editors. I don't blame them since AI detection is an entirely orthogonal skillset to regular Wikipedia editing.

b) It will generate useful data on how people use AI on Wikipedia.

c) At the price of the proliferation of the scenario above, I think it will stop a separate category of people who claim they used AI only for "formatting/grammar/organization" (really??). My hope is that asking for a transcript will bring about some self-awareness.

Perhaps a solution would be to add a line like This will aid reviewers determine which parts need to be deleted? This will make clear, I believe, that LLMDISCLOSure is only to aid deletion and AI text is still unacceptable. Ca ^{talk to me!} 09:57, 7 September 2026 (UTC)

If you're set on it, I might suggest something to the effect of;

If you have used AI but you believe that your use is covered by one of the two exceptions (basic copyediting and translation between two languages in which you are fully fluent), please provide a transcript of the prompt(s) you used while writing your draft. Athanelar (talk) 10:05, 7 September 2026 (UTC)

@Athanelar I'm a bit late, but I really like that sentence "If you have used AI but you believe ...". David10244 (talk) 19:27, 12 September 2026 (UTC)

Sounds good to me! Ca ^{talk to me!} 07:54, 20 September 2026 (UTC)

My only issue with the citations is that none of the other decline notices include them, and so it looks a bit out of character Otherwise, I don't have an issue with the re-write. qcne (talk) 09:01, 7 September 2026 (UTC)

The refs are just my own meta-commentary. They will be deleted in the final version. Ca ^{talk to me!} 09:28, 7 September 2026 (UTC)

Aha, sorry. Then, Support! qcne (talk) 09:37, 7 September 2026 (UTC)

It is a rather nitpicky point, but I think it would be better to say "You must not use AI to generate article content" instead of cannot use AI. Cheers, SunloungerFrog (talk) 09:51, 7 September 2026 (UTC)

I've made an edit request at Template talk:AfC submission for the version without the disclosure instructions. Ca ^{talk to me!} 11:43, 10 September 2026 (UTC)

• I've just noticed this and disagree strongly about changing LLM to AI. Trained AI's for specific tasks do not have the issues we face with LLMs. LLM are specific AI models that, while very useful, have huge hallucination and accuracy issues. Most (All?) of our policies use the term LLM so this is just adding confusion. If you want to de-jargon LLM you should use "AI chatbot" or "LLM (chatbot)" not just AI. As an encyclopedia we should not dumb things down because a lot of people think LLM/Chatbots are that same as AI and add to the confusion, we should help people understand there is a difference. KylieTastic (talk) 19:06, 10 September 2026 (UTC)

  This brings the question if the community is ok with trained AIs creating drafts/articles. Does that fall under WP:BOT/WP:BRFA? How would even be able to tell one way or the other? S0091 (talk) 19:14, 10 September 2026 (UTC)

  We use non LLM AIs already such as ClueBot NG but I think the task of using trained AIs for writing encyclopaedic content is way to broad a task at the moment. For instance, while I am against the proposal to use LLMs for creation of Cites, I could be convinced an AI trained just to do that would surpass human accuracy. Any WP:BRFA for using AI for any task I think is likely to need a lot of evidence, but I think it will come at some point, but not for a while. But for this discussion I doubt anyone is trying to use non LLM AIs for AfC drafts, so our policies and messaging should stay targeting LLMs or at least "AI chatbots". KylieTastic (talk) 19:28, 10 September 2026 (UTC)

  My rationale is that a lot of people doesn't realize services like Grammarly and Google Translate use LLMs behind the scenes, since they simply market themselves as an AI product. I also doubt the community would be ok with AI generating articles, even if they are not LLMs. The AfC decline templates already contain a lot of white lies and simplifications. I don't think one more would hurt. Ca ^{talk to me!} 02:48, 11 September 2026 (UTC)

• I do see your point and I have struck "strongly" from my original post. I still don't like us using a different term than is used in the guidelines, but maybe on reflection the problem is that things like WP:NOLLM should be clearer to stop the "I did not use an LLM I used Grammarly" arguments. This may be just the engineer in me not liking sloppy use of terms, and not always being good at messaging for the average submitter.
  I also think the use of "entirely" would cause issues as it makes it sound like just making any small human change is enough to stop it being problematic. KylieTastic (talk) 09:28, 11 September 2026 (UTC)

  How about: "Please delete the portions which are AI-generated, and summarize in your own words a range of independent, reliable, published sources that discuss the subject." Ca ^{talk to me!} 15:15, 12 September 2026 (UTC)

  I would be happy with that, concur (without taking a position on terminology). ChrysGalley (talk) 15:17, 12 September 2026 (UTC)

Second proposed draft

Thanks for all the feedback above! I tried to incorporate all of them in this new rewrite:

This draft appears to contain text generated by an large language model (such as ChatGPT). You cannot use LLMs to generate, draft, or rewrite article content.

Please delete the portions which were LLM-generated, and summarize in your own words a range of independent, reliable, published sources that discuss the subject.

If you have used an LLM but you believe that your use is covered by one of the two exceptions (basic copyediting and translation between two languages in which you are fully fluent), please provide a transcript of the prompt(s) you used while writing your draft (see the how-to guide).

Reviewers can make mistakes. If you have not received any LLM assistance, you may seek a second opinion at the AfC Help Desk.

See the advice page on large language models for more information.

Ca ^{talk to me!} 00:18, 30 September 2026 (UTC)

Fine with me. KylieTastic (talk) 20:26, 2 October 2026 (UTC)

Great! I've re-opened the edit request at Template talk:AfC submission with the revised version. Ca ^{talk to me!} 03:13, 3 October 2026 (UTC)

References

Reviewer qualifications should include AI recognition

There is a lot of AI-generated content in draft space. And I have seen multiple reviewers approve it for publication, only for the article to be deleted, either because it meets {{db-g15}} speedy deletion criteria or AFD participants recognize it as AI slop. Or it gets moved back to draft space.

Generating article text with AI, or communicating on talk pages with AI, is generally prohibited on Wikipedia. That's the policy at WP:EPAI.

I suspect we have a population of reviewers who aren't aware of this policy (it's fairly new), or they aren't acclimated to the influx of AI-generated drafts. Experienced reviewers are probably still assuming that every draft was actually written by a real person, so it may be hard to switch the mental mode around to admit the possibility of LLM-generated content.

Therefore, I'd like to see the reviewer qualifications include some words about demonstrating an ability to recognize AI-generated content. It could be as simple as a declaration that a potential reviewer has read WP:AISIGNS, or something more involved such as participation on WP:AI noticeboard. ~Anachronist (who / me) (talk) 14:58, 14 September 2026 (UTC)

Requiring participation at AINB is a no-go. Personally, I learned to identify AI from doing AFC. If someone messes up and approves an AI-generated draft, let them know, so they can learn from it. But we can probably make some more of that clearer in the FAQ - @Qcne, thoughts? In solidarity, asilvering (talk) 15:29, 14 September 2026 (UTC)

Maybe something like this to the Reviewers instructions under the FAQ:

AI-generated or AI-assisted drafts

The use of large language models to generate or rewrite article content is generally prohibited under WP:Writing articles with large language models, including in draftspace.

Be alert to possible LLM generated content, particularly hallucinated citations and other signs described at WP:Signs of AI writing. Writing style alone is not proof that a draft was generated by an LLM, so don't rely just on your own impression or on an AI-detection tool.

Some drafts may meet the deletion criteria under WP:G15, but G15 has specific requirements and should not be applied just because a draft appears to be AI-generated. If you are unsure, ask for help rather than guessing.

And perhaps something like Ability to identify LLM-generated prose and hallucinated citations is beneficial to the Criteria? qcne (talk) 16:55, 14 September 2026 (UTC)

I'll be honest, I am finding it harder and harder to tell. There are several drafts I've reviewed that were previously declined as LLM and nothing is obvious to me even looking through Signs of AI but still come empty. So far I have not had anything I accepted G15'd (I starting reviewing again in the past week-ish) and have asked questions here or there. However, I do fear that I simply do not possess the necessary skills. S0091 (talk) 17:51, 14 September 2026 (UTC)

I do think though that some reviewers decline for LLM without a huge amount of evidence. qcne (talk) 18:02, 14 September 2026 (UTC)

I have wondered if it has become the new WP:V decline reason which was often problematic to use as the sole reason. S0091 (talk) 18:24, 14 September 2026 (UTC)

Yeah, falsely accusing an editor of using LLMs can be more damaging than failing to notice LLM-generated text. Helpful Raccoon (talk) 23:29, 16 September 2026 (UTC)

I certainly agree, and I believe AfC reviewers have a tendency to not notice AI-generated drafts enough (which is certainly me) or overuse the template. Sometimes, the use of AI is only noticeable when a user has created a dozen articles and there is a pattern to point to. I have seen some articles declined for AI usage because a parameter in a infobox is used incorrectly, despite it rendering properly and being impossible to notice if you have not seen the template in the source editor before. This level of granularity should not be the base expectation for reviewers when the rule of thumb is a 50/50 chance of a draft being taken for AfD. GGOTCC 00:42, 15 September 2026 (UTC)

Sounds great, thanks. In solidarity, asilvering (talk) 23:48, 14 September 2026 (UTC)

I've added that and tweaked the Criteria. qcne (talk) 20:14, 15 September 2026 (UTC)

I'm not sure we need to require reviewers to do anything, but we can make it clearer that continually approving AI generated drafts will be viewed as problematic. 331dot (talk) 16:40, 14 September 2026 (UTC)

That would be fine with me and would address my concern for now. I just want to be sure that any new reviewer be aware of the burden they are expected to take on when they request the reviewer right. And those who grant this right should also be aware of the fact that it's granting the applicant to review drafts that may be AI-generated. ~Anachronist (who / me) (talk) 22:58, 14 September 2026 (UTC)

I think requiring that reviewers acknowledge having read AISIGNS might be a good measure (?). I see so many LLM-generated articles that went through AfC, tbf some of them aren't obvious and I'm sure AfC stems the flood somewhat. I've considered courtesy pinging reviewers, but it seems a bit passive-aggressive no? It's the error rate that matters Kowal2701 (talk, contribs) 22:01, 14 September 2026 (UTC)

I support adding a bit about AISIGNS to the reviewer instructions (or other relevant section), and have no issue with politely telling someone that they accepted a draft which turned out to be created by an LLM. Mistakes happen, but I oppose required involvement in another wikiproject. GGOTCC 00:46, 15 September 2026 (UTC)

What I think would be helpful is to add an AfC sub-page with an example of an AI written draft that identifies multiple AI signs (or maybe two?) which can be linked to in the instructions. One of things I find difficult to parse reading through WP:AISIGN is it sectioned by each type. I understand why it is written that way but reviewers need to be able to look at drafts/articles holistically to determine it is written by AI with multiple tells. S0091 (talk) 18:26, 15 September 2026 (UTC)

@Gurkubondinn, @M kuhner, @Kowal2701 - if you recall such a draft that's replete with examples of AI signs, please reply. Even if the one you recall has been deleted, it can be restored for demo purposes. The one I recall was this first revision, which I took interest in and rewrote from scratch in mainspace as Star of Pure Land. ~Anachronist (who / me) (talk) 04:12, 16 September 2026 (UTC)

I went through my reviews from the backlog drive that were tagged for LLM. Draft:Church_of_Eleven_22 is pretty clear. Draft:MyNetDiary is a twofer: after the first decline as LLM they asked the LLM for a rewrite and got declined again. You can mine the whole list at . These are recent, and therefore probably a good model for what a new reviewer will see. Some signs, like massive bullet lists with boldface, are common in older articles but not in new ones.

If you take this as a tiny, unscientific data set, I saw quite a few declines for non-LLM reasons on clearly LLM articles. Good that they got declined--but of course I never saw the accepts.

I got nailed by a fellow editor for letting an LLM article through myself, 2025_California_surrogacy_controversy. That one was, at least for me, harder to see--I had qualms but convinced myself otherwise. In retrospect the other editor was right. M kuhner (talk) 04:29, 16 September 2026 (UTC)

To be fair, we only have two slots to decline an article. An article may be obviously AI, but also has no citations and reads like a resume. The most blatant are speedy deleted, but you can not blame a reviewer for ordering the rational in a different way than how you would have done it. GGOTCC 04:50, 16 September 2026 (UTC)

No blame, indeed. It's just interesting to see a multiple-declined article where the declines are LLM, notability, LLM, notability, promotion, LLM. (I can't put my fingers on that one right now, but it was quite a collection. Probably someone should have rejected.) M kuhner (talk) 04:59, 16 September 2026 (UTC)

There have been a few like that, but I certainly see what you mean! GGOTCC 05:11, 16 September 2026 (UTC)

we only have two slots to decline an article sure, and we don't need more than that. There are two "big" fails: notability and LLM (with V being a third, I suppose). If a draft fails either one of those, the other stuff doesn't matter; the decline rationale for the notability reasons also include things like "make sure there are sources". This is also why we allow reviewers to comment. Primefac (talk) 09:03, 16 September 2026 (UTC)

Well, yes... and no. If the draft is in, say, Japanese, it may well be AI-generated, but I'll decline it for not being in English. If it's a blatant advert, I'll decline it for that, whoever or whatever generated it. A hoax is a hoax, whether the author wrote it or got AI to do it.

Similarly, while I agree that notability is The Biggie, if a draft is in Japanese and has zero sources, hence zero evidence of notability, I'll still decline it as non-English, because that's the most obvious (to me, at least) reason. -- DoubleGrazing (talk) 09:20, 16 September 2026 (UTC)

That is a reasonable case/exception, though my comment was more regarding GGOTCC's hypothetical (though in re-reading it I can see how it made it seem like those are the "only" reasons). Primefac (talk) 22:04, 16 September 2026 (UTC)

I have been in many situations where I am 80% sure a draft was created by an LLM but 100% sure it lacks sources and is unencyclopedic. What is the harm in not declining an article for LLM use when there are other obvious issues? GGOTCC 00:44, 17 September 2026 (UTC)

There's no harm. It's just an iota of evidence about the tendency of AfC reviewers to not notice LLM. Real evidence would be a lot better. M kuhner (talk) 00:58, 17 September 2026 (UTC)

Hyperspecific and ultimately minor “harm” (if you can call it that) is I gather data on AI draft contents based on the presence of a tag, so if there’s no tag it doesn’t get tracked Gnomingstuff (talk) 00:20, 18 September 2026 (UTC)

could we integrate the tag into AfC's LLM decline message, and just make it invisible? Kowal2701 (talk, contribs) 11:36, 18 September 2026 (UTC)

Could maybe just add the category to the tag to the LLM decline message:

[[Category:Articles containing suspected AI-generated texts from {{subst:CURRENTMONTHNAME}} {{subst:CURRENTYEAR}}]]

But drafts declined for AI at AfC are already added to Category:AfC submissions declined as a large language model output, at least until the AfC templates are removed (until someone hopefully restores them).

Or adding the user talk page category to the message that gets sent to the user talk pages:

[[Category:User talk pages with large language model notices|{{PAGENAME}}]]

‑‑gurkubondinn 21:03, 18 September 2026 (UTC)

It could be worth looking through WP:AIPAGELIST for examples to annotate. —ClaudineChionh (she/her · talk · email) 08:53, 16 September 2026 (UTC)

Another option is to compose a fake article with examples of most of the AISIGNS including bogus sources, with refnotes explaining things. WP:GARAGE does that to good effect, I think, with an example "article" demonstrating the issues discussed in that essay.

You know, one could even have an AI examine WP:AISIGNS and have it deliberately compose an article full of blatant examples. ~Anachronist (who / me) (talk) 19:48, 16 September 2026 (UTC)

I can provide this, I have a lot of examples of drafts sorted by date

In general this stuff is much easier to grasp in aggregate than individually, the AI signs page is easy to misinterpret since the patterns are way more specific and literal than many people think they are, and the more examples of a thing you see the more you notice differences. (example - Bearcat mostly does category stuff and noticed an irregularity with categories popping up all of a sudden that was totally unheard of until recently; if you haven’t been doing category cleanup for ages then it might not seem noteworthy at all) Gnomingstuff (talk) 00:16, 18 September 2026 (UTC)

It's definitely not passive-aggressive to let someone know! Just don't be rude about it. In solidarity, asilvering (talk) 01:13, 15 September 2026 (UTC)

I think that some form of instruction to not only look at a draft in isolation, but to look at a user as a whole. There's a reason that AIN threads are usually per user and not per article. But that would be a big burden on AfC reviewers, almost as big as requiring AIN participation would be (the guides for granting should maybe note AIN participation as a plus though).

But I do agree with Anachronist's overall worry (and I said something very similar somewhere on AIN just today), most experienced reviewers are probably still operating on the base assumption that what they are reading was actually written by a real person. I have no idea how to solve this.

Sometimes I kind of wish that twinkle had a way to let someone know "hey just a heads up, you accepted an LLM-generated draft the other day" in a friendly and non-warningy way. Sometimes I let reviewers know, but not nearly always. When it is someone that I have a history of interacting with, I am more likely to let them know because then I am less worried that they will see it as blame or a warning. But that is a me problem for sure. ‑‑gurkubondinn 22:32, 14 September 2026 (UTC)

Yeah I only mention it if they’re actively praising and encouraging the stuff since that sends really mixed messages, “you’re doing amazing work” suddenly becoming “your work all needs to go,” and it isn’t fair to be on the receiving end of that Gnomingstuff (talk) 00:18, 18 September 2026 (UTC)

First, I understand the concerns.

Second, as AI gets better it is ever harder to identify. Detectors are not good enough - using Ai to catch AI seems fatuous to me - and one can argue that academic phrasing and Ai phrasing are so similar as to cause problems

Third, hallucinated references and text as a communication to the user seem useful

Fourth, The problem I see is that the human brain can start to recognise everything as AI, or nothing as AI.

All this is interesting and not that helpful, however. I know that it is ever harder to recognise AI slop. No-one lets it through on purpose. But giving reviewers an extra higher, ever higher, hurdle to leap will ensmallen the pool of reviewers.

Together with NPP we do ok. Add in community CSD/AFD if AFC and NMPP each fail to catch iyt, then it's pretty good. We could always do better, but AI will do ever better. 🇵🇸‍🇺🇦 Fiddle^Timtrent Faddle^Talk to me 🇺🇦‍🇵🇸 09:43, 16 September 2026 (UTC)

New decline-scrunching design

Main page: User:MrPersonHumanGuy/AFC decline redesign

After withdrawing an RfC that had gone on for over a month, I've settled on a single new proposal for how decline notices ought to look when the small parameter is used.

The notices are collapsed so they no longer take up so much vertical space, but they contain abbreviated decline reasons in their collapsed state so reviewers can still glean all the reasons instantly, and the full reasons can still be seen when the notices are expanded. Hopefully this will be more to everyone's liking. – MrPersonHumanGuy (talk) 20:09, 14 September 2026 (UTC)

Please sandbox your code, either in the existing sandboxes or as a fork (e.g. /sandbox2) so that your redesign page will actually look like it would on the live page without being scads of code. Primefac (talk) 22:30, 14 September 2026 (UTC)

Unfortunately, what I've presented is only an interactive visual representation of how I think the notices ought to look. {{AfC submission}} has multiple subpages, including /declined, /rejected and /comments, so if I were to try to sandbox a fully working model of an individual notice template by editing any of these pages, I would have no idea where to begin or what wikicode would have to be changed to make the small parameter return something similar. – MrPersonHumanGuy (talk) 14:51, 15 September 2026 (UTC)

@MrPersonHumanGuy Unpopular opinion but ChatGPT is now pretty good at converting stuff into Wikitext. qcne (talk) 09:08, 16 September 2026 (UTC)

WP:NOPROTOTYPE may be of interest. – MrPersonHumanGuy (talk) 21:23, 24 September 2026 (UTC)

If you can't or won't do it that's fine. I would prefer to have a semi-working version because if you do get consensus to implement your proposed changes, I don't really want it sitting here for years because no one has the time or inclination to figure out how to make it work. I have no opinions on the proposal and I am certainly not opposing because there isn't a coded proof of concept. Primefac (talk) 13:47, 26 September 2026 (UTC)

If we were to go ahead with this and try to make the necessary edits to the relevant sandboxes, we should be able to get something like this at the testcases page:

Visual representation of hypothetical testcases

{{AfC submission/declined|nn|reason2=rs|decliner=Example|declinets=20171101032705|small=yes}}

{{AFC submission/declined}}

{{AFC submission/declined/sandbox}} (I don't know how to accurately replicate the gap between the reason boxes)

Submission declined on 1 November 2017 by Example (talk). Fails Wikipedia's criteria for inclusion / Inadequately supported by reliable sources

Declined by Example 8 years ago.

{{AfC submission|d|j|decliner=Example|declinets=20171101032705|reject=yes|small=yes}}

{{AFC submission}}

{{AFC submission/sandbox}}

On which template sandbox subpage should the abbreviated declines be put in order to achieve something like this? – MrPersonHumanGuy (talk) 13:28, 16 September 2026 (UTC)

Any that would be relevant to the proposed redesign; I don't think there's anything currently being sandboxed right now so you could overhaul all of the primary /sandbox pages with no issues. Primefac (talk) 11:41, 20 September 2026 (UTC)

Although these changes may be doable in theory, I don't think I would be able to figure out how to properly alter the formatting of a set of templates this complicated. If anyone wants to try to edit the sandbox versions so their small forms would resemble my designs (including the reason shorthands, see the full list at this page), they can do so. I'm sure that the wikicode for the real thing would need to be very different from the wikicode I typed up. – MrPersonHumanGuy (talk) 12:55, 20 September 2026 (UTC)

Well, this is probably one of the few times where we should have proof-of-concept before actually deciding to do anything, because if the desired end result cannot (easily) be obtained there's not much point in discussing it. To be completely honest this is not a high priority for me so I will probably not be investing time but I'm happy to consult if there are issues. Primefac (talk) 14:07, 20 September 2026 (UTC)

MrPersonHumanGuy, the sandbox is the place to try :).

If you haven't made any progress or found someone interested in two weeks, ping me and I may or may not have some time then. LittlePuppers (talk) 20:06, 25 September 2026 (UTC)

Consensus? (no RFC this time)

Would anyone support or oppose my new proposed redesign of the small form of {{AfC submission}} as illustrated within the discussion just above this subsection? – MrPersonHumanGuy (talk) 21:35, 24 September 2026 (UTC)

Looks great to me. Thank you for your work! Ca ^{talk to me!} 00:21, 30 September 2026 (UTC)

Submissions by US TAs, Wikipedia:Sockpuppet investigations/Ylevental

So there's a sockmaster that uses TAs to submit drafts through AfC, see WP:AINB#Aregwaerf (the SPI was initially filed at that username). There were about 30 known TAs that succeeded in putting ~50 LLM-generated drafts through AfC (at most, any one TA did a handful). The TAs appear to geolocate all over the US, and the drafts IIRC were often BLPs. If someone more familiar w SPI and TA socking could narrow the profile down, that'd be grand, otherwise can I just ask that people be on the lookout for such submissions? Kowal2701 (talk, contribs) 22:09, 14 September 2026 (UTC)

cc @Ethmostigmus, Sulfurboy, FuzzyMagma, and Cosmic840: I've previously messaged y'all about this sockmaster... have y'all seen any patterns in their articles? Ed ^{[talk] [OMT]} 00:53, 15 September 2026 (UTC)

Cheers for the ping The ed17 - looking at my AfC log, I've only directly interacted with this sockmaster once through a TA, ~2026-40969-38. TAIV fetches 12 associated TAs editing on a range of topics, primarily BLPs and articles about locations, mostly related to the United States. I see lots of US BLPs, articles relating to the states of Maryland, Virginia, West Virginia, and Washington, and a few articles relating to Judaism, trains/railways, and scientific concepts. I haven't personally come across any more articles from this sockmaster at AfC since then, and I don't see an obvious overarching pattern besides the connection to the US (which is not especially unique given our large American userbase). Ethmostigmus 🌿 (talk | contribs) 03:13, 15 September 2026 (UTC)

Basically same conclusion from me, although this is a lot to sift through, but I'm not seeing an overarching pattern. Sulfurboy (talk) 06:12, 15 September 2026 (UTC)

Can someone walk through Draft:Night Owl Video and point out the AI signs as it is not obvious to me? S0091 (talk) 16:15, 15 September 2026 (UTC)

I'm hardly an expert at detecting LLM text compared to some of the AI cleanup folks, and shorter articles like this are more difficult to spot AI signs in because there's less text to analyse, but the thing that stands out to me is the repeated attribution: According to The Guardian, In interviews, Hamel and Mills said, The Guardian reported that. These three phrases and the statements they are attached to are all in reference to the same single interview. Humans also do this sometimes, but it happening three times in such a short article, all in reference to the same source, in an article by a user already known to use LLMs... it's a little suspect. Hamel began his career is another common bit of vague puffery often seen in LLM outputs. If you haven't already given it a look, WP:AISIGNS has a lot of useful examples of potential indicators. Ethmostigmus 🌿 (talk | contribs) 02:40, 16 September 2026 (UTC)

Also not liking the vague connector words like "participates in." And the structure of the paragraph which starts out talking about the two founders and then seems to forget one of them. LLM text is often a bit disjointed, losing track of the main idea and wandering off. M kuhner (talk) 04:44, 16 September 2026 (UTC)

@Ethmostigmus I have read through AISIGNS multiple times while reviewing articles trying to determine if they are LLM. Can you point to which sign that exemplifies what you describe? I don't see anything about repeated attributions. What I see there are examples of "vague attentions". Citing a specific source is not vague and stating what it says is not vague. @Kowal2701 the reason it does mention the other founder is because the interviews/statements were from Hamel so Mills is only briefly mentioned in the sources. Mind you, I am not all saying this was not LLM. Iif these truly are signs then AISIGNS may need to be updated. S0091 (talk) 20:36, 16 September 2026 (UTC)

@S0091, what Ethmostigmus means, I believe, is that those three attribution statements were all about the same article. In solidarity, asilvering (talk) 20:39, 16 September 2026 (UTC)

It is not at all uncommon to use one source for to support multiple claims. My question goes back to which WP:AISIGN is this? S0091 (talk) 20:51, 16 September 2026 (UTC)

WP:AIATTR (over-attribution intext), but that's only one sign. The rest of it is plausibly AI, and experienced AfC reviewers can trust their intuition, but personally I'd need to run a passage through Pangram and do a couple spotchecks to determine that this article was AI-generated in isolation (I usually don't deal with marginal articles, I look through someone's contribs for the more obvious ones). Kowal2701 (talk, contribs) 21:06, 16 September 2026 (UTC)

Thanks you @Kowal2701. If intuition is fine to use I will because most new drafts I see have LLM vibes but I can't necessarily point to a specific WP:AISIGN. I will say that if WP:AIATTR is meant to include specific attribution rather than the examples given of what is still vague attribution (i.e. media outlets or listing several), then it needs to updated. S0091 (talk) 21:25, 16 September 2026 (UTC)

yeah it's both, emphasis on general coverage in sources as well. Tbh I have seen one declined draft that I thought was human-written, so there are almost certainly false positives wrt intuition, but for the time-being idk what other options we have Kowal2701 (talk, contribs) 21:28, 16 September 2026 (UTC)

In case investigation we do rely very heavily on looking at multiple articles. I worked on one high-profile case where I did detailed source/text checks. The first two articles I checked were...maybe? Don't like this sourcing but...? The third was clearcut, and this suggests the first two were also LLM.

Confronted with a page of articles to check, I go for big edits and page creations, and as recent as possible; signs are easier to see without intervening edits. Often other editors will have some sense that things are not right, and will remove some of the signs without actually identifying LLM as the cause. If the big edits and creations are bad, I start reverting medium-sized ones such as source additions, even though I can't necessarily prove they were LLM. There is some sacrifice of accuracy for speed, because the problem is fairly overwhelming. We have started omitting very small edits from the tracker: there is a non-zero chance that they are big edits with balanced addition and subtraction, but there's just no time. (Also cases over about 1200 articles break the tracker.)

One thing that would really help is a better way to find the LLM start date. I have approximately 20 cases in my personal spreadsheet that need a start date and tracker, but it's painful hand work to figure out....

Another thing on the wish list is a clean way to check everything in a massive sockpuppet case, especially ones where SPI has not provided a list of all of the socks. M kuhner (talk) 22:02, 16 September 2026 (UTC)

@M kuhner, can you explain more about that last sentence? In solidarity, asilvering (talk) 22:18, 16 September 2026 (UTC)

I think she was referring to Wikipedia:Sockpuppet investigations/Ylevental (correct me if wrong) where it's very difficult (impossible?) to exhaustively find all (even most) of this person's TAs, let alone cleanup all their contribs. There's got to be a way to make finding related TAs easier, including old TAs Kowal2701 (talk, contribs) 22:24, 16 September 2026 (UTC)

I was actually thinking of WP:AINB#Aregwaerf where we seem to be giving up as there are too many socks.

There are two problems, one of which is out of my ken. The first is finding all the members of a sockfarm. The second is organizing the cleanup. Our tracker tool is superb for a single account case but not so good for a sockfarm. It feels like better tools are possible here though I don't know exactly what is best.

We have a different priority than SPI. At SPI they don't care about old TAs and inactive accounts, as they likely will not post in future. At AINB we have to clean up after them, so we do care.

(As long as I'm asking for the moon--I want a version of Who Wrote That? which lets you enter an editor name and jump to text from that editor, if any exists.) M kuhner (talk) 22:36, 16 September 2026 (UTC)

(that's the same one btw, the SPI was moved to an older account) Kowal2701 (talk, contribs) 22:38, 16 September 2026 (UTC)

Unfortunately @M kuhner that is kind of asking for the moon. We limit the availability of IP data, for example, past 90 days. In solidarity, asilvering (talk) 22:51, 16 September 2026 (UTC)

I figured.

A way to put a list of editors into an interface and get a workable consolidated tracker for all of them wouldn't be bad, though. I keep looking at Aregwaerf and just wincing and doing something else. Still leaves the problem of getting the list, but at least once I get it, I could use it. (I'll talk to User:DVRTed, our LLM cleanup tools guru.) M kuhner (talk) 23:34, 16 September 2026 (UTC)

Check out WT:AIC § Potentially useful userscript, it's pretty close and rather useful. Still super fiddly to clean up an article, but it's better than staring at diffs and trying to figure it out that way. ‑‑gurkubondinn 23:40, 16 September 2026 (UTC)

The Aregwaerf AIN case and the Ylevental SPI are the same user, just discovered under different account names. I don't remember how it ended up in two SPI cases though, but they've been merged now: Diff/1373584151. ‑‑gurkubondinn 23:45, 16 September 2026 (UTC)

Citing a source, including multiple times, is not a sign of anything. However, also providing a text attribution (in addition to the citation) is generally reserved, by humans, for statements where the identity of the source is important to interpretation of the claim. So a human will write "The movie came out in 1983 (1)" and not "The New York Times said the movie came out in 1983.(1)" However, a human will write "Roger Ebert said it was a terrible movie.(2)" because this is an opinion or judgment, and those should be attached to a person.

LLMs frequently give in-text attributions which a human would regard as unnecessary. In this article, According to The Guardian, it was the first freestanding video store to open in New York City since the closures of Kim's Video and Videology particularly stands out. This is not an opinion, and knowing it was The Guardian adds little to our understanding; an inline cite would be plenty.

Also, In interviews, Hammer and Mills said is typical LLM. Why do we need "In interviews" when we are giving in-text attribution and a citation--what does this add? (I'd also immediately be suspicious of the plural--for some reason LLMs love to talk about multiple interviews, critics, sources, etc. even when there is only one.)

A more detailed analysis would show that "In interviews, Hammer and Mills said" is actually false. It's attributed to two sources. In one, Hammer and Mills said something like this in a statement--not an interview. In the other, the person saying something like this is not Hammer or Mills but one of their customers. This isn't useful for snap diagnosis but when I found it I would feel 100% sure this was LLM.

M kuhner (talk) 21:16, 16 September 2026 (UTC)

@M kuhner so helpful! Thank you! S0091 (talk) 21:34, 16 September 2026 (UTC)

A few tweaks to the AfC Tabs, Reviewer Criteria, Help desk header - plus a protected edit request

I've made a few small changes intended to make the project navigation and signposting a little clearer:

AfC tabs

• Reordered the links under Submissions to Feed > Category > Sorting, so the live feed is the first option.
• Moved Apply to what seemed a more natural position, with Participants and By subject grouped beneath it.

Reviewer applications

• Moved the reviewer criteria to the page where editors actually apply to become reviewers. They were previously on the list of reviewers, which seemed an unintuitive place for them.
• Expanded the criteria with a short Helpful experience section and some guidance on how editors who do not yet meet the requirements can gain relevant experience.

Help desk header

• Made a few wording changes to make the signposting clearer.

Protected edit request: Wikipedia:WikiProject Articles for creation/Participants

Since this page is fully protected, could an administrator please remove the old header and application button? These have now been moved to the reviewer application page at Wikipedia talk:WikiProject Articles for creation/Participants.

Please replace:

{{pp-protected|reason=Per previous full protections, and to curb people adding themselves to the list who do not meet requirements|small=yes}}
{{/header}}
{{shortcut||WP:AFC/P|WP:WPAFC/P}}
If you want to participate in this WikiProject, please click the button below, and an admin will review your request to see if you meet the qualifications.

{{center|1={{Clickable button|{{Space|5}}Apply to become an AFC reviewer{{Space|5}}|url={{fullurl:Wikipedia_talk:WikiProject_Articles_for_creation/Participants|action=edit&section=new&nosummary=true&preload=Wikipedia:Articles_for_creation/Preload&preloadparams%5b%5d={{urlencode:{{lc:{{#titleparts:AFC helper script}}}}}}}}|class=mw-ui-constructive}}}}

Participants who have not carried out any reviews in more than 6 months, or have not edited at all in the last two months, are considered inactive and may be moved to the inactive reviewers list.

==Active reviewers==

with:

{{pp-protected|reason=Per previous full protections, and to curb people adding themselves to the list who do not meet requirements|small=yes}}
{{/header}}
{{shortcut||WP:AFC/P|WP:WPAFC/P}}

Participants who have not carried out any reviews in more than 6 months, or have not edited at all in the last two months, are considered inactive and may be moved to the inactive reviewers list.

==Active reviewers==

qcne (talk) 08:48, 16 September 2026 (UTC)

 Done Sohom (talk) 17:37, 16 September 2026 (UTC)

@Sohom Datta thanks! I stupid kept in the header line in my edit request:

{{/header}}

Any chance of removing that too? qcne (talk) 17:43, 16 September 2026 (UTC)

I disagree with this change; there's nothing wrong with having an intro to being a participant and the guidelines for being one on the main Participants page. Primefac (talk) 22:08, 16 September 2026 (UTC)

@Primefac Happy to revert it to include the Criteria header - I just thought it was a bit more tidy to have one page that's just the list of participants and one page that is just the criteria and AfC requests. It's why I've re-ordered them in the tabs to make the hierarchy clearer. Less duplication that way. Thoughts? qcne (talk) 10:34, 17 September 2026 (UTC)

Yes, I would like to keep the headers; people will expect the main page to have the criteria. Primefac (talk) 10:48, 19 September 2026 (UTC)

Happy for you to re-add! qcne (talk) 12:13, 19 September 2026 (UTC)

Also, on the revert of the COI section on the Reviewer instructions... I very much had Greghenderson in mind when I wrote that. Not that he was a reviewer, but if a situation arises where a reviewer does have a CoI with a topic like Carmel on Sea as Greg did, it would be good to explain what to do. Happy to have a think about wording? What are your thoughts. qcne (talk) 10:38, 17 September 2026 (UTC)

That was about disruptive editing; the COI was part of the issue but not the only reason for the block(s). At most I would suggest a sentence along the lines of "if you have a COI and cannot be objective, leave the review for someone else". We can also (obviously) keep the paid editing note. Primefac (talk) 10:55, 19 September 2026 (UTC)

I'll add that :) qcne (talk) 12:11, 19 September 2026 (UTC)

Next steps for Article Guidance feature in AfC

Hi all! Following a Discord call last month where we shared the Article Guidance feature (https://www.mediawiki.org/wiki/Article_guidance) and collected initial feedback (see previous discussion), I'd like to continue the conversation and take it a step further.

I think the AfC context is one of the most promising places to test and develop this further, as the feature would help junior editors arrive with a better first version of an article. I'd love to find a couple of volunteers from this community to work on next steps.

Apologies for directly pinging some reviewers I see are very active in the review process: ChrysGalley, EaglesFan37, DoubleGrazing, Lynch44. Would you be interested in helping, or do you know others you would recommend?

Thank you in advance! GGalofre-WMF (talk) 14:36, 19 September 2026 (UTC)

Hi @GGalofre-WMF: without knowing exactly what you'd want me to do, or whether and how I could add value to this, I'm certainly happy to get involved, at least for the parts that happen on en.wiki (I'm not too keen on Discord). This sounds like it could potentially have a big positive impact on AfC, NPP, AfD, and pretty much the entire project, so I'd be silly not to want to help if I can. (Plus, I'm always looking for ways to spend more time here, because I'm not quite up to 24/7 yet!) -- DoubleGrazing (talk) 14:55, 19 September 2026 (UTC)

As messaged on Discord, also happy to help. qcne (talk) 18:15, 19 September 2026 (UTC)

I can try and help where possible, although I don't have Discord. Wasn't familiar with this but after a quick read this certainly seems like a net positive to the AfC process. Lynch44 20:07, 19 September 2026 (UTC)

@GGalofre-WMF - yes happy to help. Particularly if it's in a few weeks, just undergoing medical stuff at the moment. ChrysGalley (talk) 08:26, 20 September 2026 (UTC)

DoubleGrazing, Qcne, Lynch44, ChrysGalley Thank you all for the quick responses and willingness to help, really appreciated!

For coordination, I'd suggest using this talk page as the main space so everyone can follow along regardless of whether they're on Discord. I'll keep both channels updated.

As a first step, I'd love for everyone to try the feature directly. Some of you already have. It can be accessed in Simple English Wikipedia: https://simple.wikipedia.org/wiki/Special:NewArticle.

The project page has a breakdown of the workflow steps (https://www.mediawiki.org/wiki/Article_guidance) and how guidance can be customized via outlines (https://www.mediawiki.org/wiki/Article_guidance/How_to_create_an_outline). I am also happy to do another virtual more hands-on session to go through it if there is interest.

Once you've had a chance to explore it, please share your initial reactions here on what works, what doesn't, or what has potential but it's not quite there yet. Eventually I'd like to understand what would be a must before you'd consider enabling this for junior editors submitting to AfC and testing it for real.

From there we can figure out the best way to continue the conversation, whether that's here with me asking specific questions, testing specific proposals on prototypes, a more interactive call, etc. GGalofre-WMF (talk) 14:52, 20 September 2026 (UTC)

I'm not entirely sure how whelmed to be here. I do like that you get a little pre-filled template to insert text in the appropriate places (I checked a few of the "type" options and verified they have different presets). On that note, though, Subject unavailable? Pick a type instead makes no sense and I only clicked the "Pick a type" because it happened to be blue (maybe consider changing "type" to something more descriptive?).

Only other thought I have at the moment is that I disagree that the Reference checks actually does anything (I put in m.com and it said nothing).

Neat tool, though, will be interested to see it progress. Primefac (talk) 15:09, 20 September 2026 (UTC)

Thanks Primefac for the feedback!

Agreed on the wording on "Subject unavailable? Pick a type instead". We're currently working on alternatives to make that clearer. The idea is that if you don't see your topic in the search results you can still indicate what type of article you're writing, which is needed to show appropriate guidance.

A related solution we're also exploring is a "generic guidance" path as a default set of guidance for cases where there is no outline available for your topic. Less specific than topic-tailored guidance, but potentially better than a dead end.

On the reference check, you're right that it's highly dependent on editors adding discouraged sources to each outline. If nothing is listed, there's nothing to check against. Connecting this to existing community lists like the perennial sources list is possible, so it doesn't rely entirely on per-outline manual input.

Are there existing lists of discouraged sources that would be safe enough to add to any or all outlines by default? GGalofre-WMF (talk) 15:30, 20 September 2026 (UTC)

WP:RSP is a non-exhaustive list of the most common sources; a link to that might be helpful if/when this gets moved into the enWiki space. I'm not sure if other languages have such lists, though I wouldn't be surprised if they did. Primefac (talk) 15:52, 20 September 2026 (UTC)

@PrimefacThanks for that. Making it easy to link an existing source list to an outline is something we can and will work on. I'm slowly going through all the feedback here and I'll share a list of all collected feedback with how likely we are to implement each piece. GGalofre-WMF (talk) 17:18, 23 September 2026 (UTC)

I tried a few names from my todo list. It took so long for the first check I thought it was broken. Then it told me twice I could not continue as subjects have to exist on wikidata. I found one that was on wikidata, but then said it had to exist on another wiki. So I then deliberately picked one from another that was on wikidata and another language wiki: It took 2 mins 10secs to look up. Finally I tried with one I wrote here Alexander H. Curtis just to try to get somewhere. I was surprised it just asks you to add sources and does not even link to the existing article as a good place to start looking. Nor does it link to searches to help. So far it's just worked as a huge block to starting. I added some sources, and found another issue: the sources have to be urls and with no explanation that offline sources are valid for articles just not this tool. I finally got to the very basic outline and frankly that was a lot of effort for not a lot. I like the idea of doing pre-checks and giving an outline to help new users, but in it's current state if I had tried that as a new user I'm pretty sure I would have just walked away not just from creating the article but from editing in general. KylieTastic (talk) 15:51, 20 September 2026 (UTC)

Quick side comment about the WD thing, I never got that as a "roadblock" and I made up some pretty weird titles just to avoid existing pages. Primefac (talk) 15:53, 20 September 2026 (UTC)

We would for sure disable the WD check and the other-wikis on en-wiki, so don't worry too much about that. In solidarity, asilvering (talk) 16:22, 20 September 2026 (UTC)

Although I agree that would make sense for us, it does not change the feedback validity for others. As this was on simple I also considered it from a user of that site (even if my participation there has been a pitiful 82 edits). KylieTastic (talk) 21:48, 20 September 2026 (UTC)

The WD issue appears to depend on subject "type". Actor, Politician, Person, Museum, Conflict, Religion etc gets "You can create this type of article only for subjects with a Wikidata entry." Business gets the slightly different "To create an article about this subject, it must already exist in other Wikipedias." Topics such as Landform, Plant, Animal etc allow you to proceed. I can see why some wikis would want to put more controls on some subjects but I think especially for new users they should have a clear explanation why. Such as and very much assuming here: "You cannot create this type of article for subjects without a Wikidata entry. This is because most people are not notable enough for a Wikipedia article so to avoid disappointment creation of such article is not currently possible via this tool." Also It would be good if for subject types like 'Politician' it actually had follow up questions to allow some through: "Is this person elected or just a candidate?" "Are they elected to a international, national, or state/province-wide public office?". If a wiki was going to stop people from proceeding via a tool, I assume will be promoted to new users, it should be really clear why they cannot proceed. Just telling people "NOPE" is very negative, telling someone "sorry because...." is just disappointing.

Lastly maybe it's just me but "You can create this type of article only for subjects with a Wikidata entry." reads bad: it starts positive "You can create this type of article" then slams the door. I'm sure I've read you should not do this. I would have put "Sorry, but you cannot create articles with this subject type that do not yet have a Wikidata entry. ... reason..." Maybe this is just a demo config issue though. KylieTastic (talk) 21:42, 20 September 2026 (UTC)

KylieTastic Thank you for this thorough run and sorry for my late reply, I was out of office for some days. I'll first address your comments here and later on I'll consolidate all feedback received into a post indicating what we're taking on and what we're not prioritizing (and why).

It took so long for the first check I thought it was broken and It took 2 mins 10secs to look up

We know in some cases it does take too long and we are working on it from two directions.

The first is technical: changing how and in what order the "Wikidata check" and the "outline matching" happen. We think by tweaking these we can get better loading times.

The second is a fallback: rather than leaving someone waiting, cut the check short and move them into the workflow with a generic guidance flow. The generic guidance option would allow for communities to create a generic outline. This could be used when loading times exceed a specific threshold as well as for topics that have no matching outline yet.

the sources have to be urls and with no explanation that offline sources are valid for articles just not this tool

Taking this one. The range of sources we accept and the formats we handle has come up a few times. I think that even before we implement offline reference intakes we could improve this by explaining that for now only URLS are accepted in this tool but that books can be used later on in the article, for example.

Do you see newcomers using offline sources often? I ask because validating them is harder than validating a URL, so there may be value in simply accepting them at that step first, even if we cannot yet tell the editor whether the source is encouraged or discouraged.

I can see why some wikis would want to put more controls on some subjects

Correct, and that choice is for each wiki to make. The Wikidata and crosswiki checks are optional and are set per article type and editor experience, which is also why Primefac never hit them at all.

For en.wiki that decision would be yours, made when creating the outline for a type: whether either restriction genuinely prevents bad articles for that type, or just gets in the way. And if in doing that you spot a restriction we have not thought of that would make a real difference, we are open to building it.

it should be really clear why they cannot proceed

Agreed, and I am taking your copy suggestion. Two things I want to carry beyond the wording: the reason has to be visible rather than just the refusal (this is something easy we can change), and a blocked editor could also be offered somewhere to go instead of a dead end (this is a bit more complex but also valuable feedback to implement).

Is this person elected or just a candidate?

This is an interesting idea. I agree a step like this could be the deciding factor between blocking or allowing someone to continue. We also want to think about a balance between too many steps and too few. We're comparing Article Guidance to "no guidance at all", a blank page, so we're conservative about adding steps: the more we add, the more likely someone drops. That said, your case is the exception to that logic, since for someone who is currently blocked outright, one extra question is not a cost, it's the difference between an article and nothing, or between learning what is notable or walking away without knowing.

does not even link to the existing article as a good place to start looking.

Linking to an existing article sounds like a good option for the case of Simple English and English, as they share the language, but may not be that useful in other language combinations. I do agree that sharing available sources from the same article in other langauges could be a useful addition in the source step.

that was a lot of effort for not a lot and I would have just walked away not just from creating the article but from editing in general

Noted, I think this is a useful read that aligns with our priorities for the next month or two: making each step flow better, with lower loading times, better copy and less drop due to unclarity or bad UX. Following with some of the improvements flagged here and in other comments. GGalofre-WMF (talk) 09:24, 29 September 2026 (UTC)

Hey GGalofre-WMF, no worries on the delay we all need breaks. To answer do we see newcomers using offline sources often?, not as much as online but three main cases come to mind: translating from another language, using a source from another article, referencing newspapers or magazines for certain subjects. I think as long as they are informed it is a tool limitation not a policy limitation it is fine. Also the idea of not gust given editors reasons over refusals, but also a place to go next is helpful as dead ends make too many just give up. If the process leads people to either create an article, or informs them and gives them a place to go next it will be a net positive, we just need it not to be a frustrating dead end. Cheers KylieTastic (talk) 09:42, 29 September 2026 (UTC)

I don't really understand why this tool doesn't make use of citoid. Newbies should be able to drop in an ISBN and have that citation accepted by this tool. (Yes, some books aren't reliable sources, but the vast majority of books people try to cite are going to be just fine.) In solidarity, asilvering (talk) 14:56, 29 September 2026 (UTC)

I agree with both KylieTastic and asilvering. Anything that can be automatically generated in VE (ISBNs, DOIs, etc.) should flow through the process but other offline sources are used often enough, there needs to be some kind of accommodation/guidance for them. S0091 (talk) 15:25, 29 September 2026 (UTC)

From a brief glance:

• The wait after entering a title ("checking for matching subject") was way too long (minutes?), and sometimes gave errors
• Agreed that "pick a type" is unclear
• Maybe a search for the "what is this" list, especially if you're expecting it to get longer? Or subcategories, i.e. select "person" and then have more specific options to choose from?
• Is there no "other" option? (Or is that just a simplewiki thing?)
• The descriptions (are they just the short descriptions?) for the categories/templates are not great... "Any single member of Homo sapiens, unique extant species of the genus Homo", "Non-tangible executable component of a computer"

That's a far as I'm getting for today. LittlePuppers (talk) 16:44, 20 September 2026 (UTC)

LittlePuppers Thanks, that is a useful list even from a brief glance.

The wait after entering a title ("checking for matching subject") was way too long (minutes?), and sometimes gave errors

On the wait, see my reply to KylieTastic above: we are working on it from the technical side, and separately looking at cutting the check short and moving people into a generic guidance flow rather than leaving them waiting.

I would like to separate the "errors" from the slowness. If you remember what the error said, or roughly what you had typed when it appeared, that would help. A failure and a timeout look the same from the outside but are different problems for us to fix. An error is also different than "no results". So more context there would help to look into this.

Agreed that "pick a type" is unclear

This is actively being worked on, thanks!

Maybe a search for the "what is this" list, especially if you're expecting it to get longer? Or subcategories

Agreed on both, and taken. That list only grows as communities add outlines, so it needs structure rather than more scrolling. Grouping under broader types, with a search, is the direction we're taking.

Is there no "other" option? (Or is that just a simplewiki thing?)

Not a simplewiki thing, there is no "other" anywhere at the moment, which is the dead end a few people have hit in this thread. What we are working on is a generic guidance path: a community-written outline not tied to any specific topic, used when nothing matches, and possibly also when the matching step is taking too long. That is the nearest thing to "other", and I would like to put it in front of this group once we implement it.

The descriptions (are they just the short descriptions?) for the categories/templates are not great

Yes, they are the Wikidata descriptions used exactly as they come, which is how you end up with "Any single member of Homo sapiens". They cannot be edited today. Adding a description field to the outline itself, so communities can write something readable, looks straightforward and I will check what it would take. Feedback taken. GGalofre-WMF (talk) 09:32, 29 September 2026 (UTC)

@GGalofre-WMF my attempt was also stuck by the Wikidata block. In my case it was for the Irish architect Christine Casey, who 101% is notable, but just didn't have a Wikidata entry. But there was an oblique reference to a "researcher" with ORCID information (actually an unrelated immunologist) so I would have been tempted to plug on with that entry instead making the block here counterproductive in terms of data mapping, though in reality I didn't. And, um, to make a point this is probably firmly Women in Red territory. So I think that Wikidata issue is pretty significant. Looked nice though! ChrysGalley (talk) 07:07, 21 September 2026 (UTC)

@ChrysGalley thanks for trying it!

there was an oblique reference to a "researcher" with ORCID information (actually an unrelated immunologist) so I would have been tempted to plug on with that entry instead

This is the part of your comment I find most useful, and I would like to pull on it, because I think there are two problems tangled together here: the results themselves, and what we give you to judge them by.

On the second, two questions:

Would the same results have been less confusing with better labelling? In your case Christine Casey the Irish architect is not in Wikidata, but other people with the same or a similar name are. Each result carries a visible descriptor tag, "researcher" or "sportsperson" and so on, is that confusing or not visible enough to tell you these are different people?

And would a more explicit way forward have helped? Something along the lines of "Not listed? Start with a new subject", making clear that a search failing to find your subject is not the end of the road and you can still go on to write the article. Is that roughly the right shape, or would you word it differently?

More generally we are looking at the first matching step as a whole: loading times, what the results show, and what happens when they are not what the editor expected. Your case is a useful test of all three. GGalofre-WMF (talk) 10:34, 29 September 2026 (UTC)

@GGalofre-WMF, the issue with the same-named other person isn't that the article writer can't tell they're different people. It's that they're prevented from creating an article unless they pass this gate, so when they're frustrated by something blocking them, they're going to pick "option that lets me clear this gate", whether that's correct or not. It's what I'd do, too. In solidarity, asilvering (talk) 14:50, 29 September 2026 (UTC)

Only now getting around to trying this. For the first subject I attempted, there wasn't a suitable type; I wanted 'organisation' (as distinct from business or eg. university), so I've now suggested that. I then tried to proceed by selecting 'business' instead, but that ground to a halt with To create an article about this subject, it must already exist in other Wikipedias. My proposed title was "UK Research and Innovation", and an article does exist at that title here in en.wiki at UK Research and Innovation (and, in case the "other Wikipedias" plural is taken literally, the Turkish and Norwegian wikis have a corresponding article also).

I'll try a different subject, so I can go through the whole journey, hopefully. -- DoubleGrazing (talk) 08:56, 23 September 2026 (UTC)

DoubleGrazing Thanks for trying it!

For the first subject I attempted, there wasn't a suitable type; I wanted 'organisation'

Noted, and thank you for suggesting it rather than abandoning the attempt.

More generally this is what the "generic guidance" path is meant to solve: a community-written outline that is not tied to any particular topic type, used when nothing in the outline list matches. Less specific than a tailored outline, but better than no option that fits. It is in progress and I would like to share it to this group once it exists.

it must already exist in other Wikipedias

"Other Wikipedias" in this restriction option currently means at least five, not two. UK Research and Innovation exists on three, so the check failed.

That is the same problem KylieTastic raised above: these restrictions right now block a user without explaining why. A message saying "this subject needs an article on at least five other Wikipedias" would at least have let you understand the outcome better.

Whether five is the right number is a fair thing to argue about and it can evolve.

I'll try a different subject, so I can go through the whole journey, hopefully

Worth knowing before you do: these restrictions are set per article type and per wiki, and for Simple English they have stayed fairly strict. Types such as Landform, Plant and Animal carry no restrictions at all in that wiki, so if you want to see the whole flow without fighting the gates, those will get you through.

What I would ask is that you read the restrictions themselves as choices Simple English made rather than as how the feature has to behave. For en.wiki those choices would be yours, made per article type when writing outlines. Whether each restriction genuinely prevents bad articles for that type, or just gets in the way.

If you have different restrictions in mind that would be more efficient, please feel free to share them.

Thank you! GGalofre-WMF (talk) 10:54, 29 September 2026 (UTC)

Went ahead and gave it a try. I didn't find the process of getting there to be that confusing, although from a perspective of a new editor I can definitely see how it could be. I agree with LittlePuppers above that some of the category descriptions are bizarre, and with Primefac that the "subject unavailable" note is poorly worded. The thing that confused me the most was the Wikidata requirement; I found it weird that there are no direct links to the Wikidata source material from the script. I also noticed that the script requires two sources, at least for here I'd make it three per WP:42, with a note encouraging editors to add more if possible. The source and notability guides were helpful. Lynch44 01:39, 21 September 2026 (UTC)

Lynch44 Thanks, and useful to know which parts held up.

some of the category descriptions are bizarre and the "subject unavailable" note is poorly worded

Agreed on both, and both are on our radar. Fuller answers in my replies to LittlePuppers and Primefac above.

I found it weird that there are no direct links to the Wikidata source material from the script

We ask you to accept or reject a match without giving you any way to go and check it, that's a good point. ChrysGalley hit the consequence above: they were offered an unrelated person and could reasonably have accepted it, and a link through to the item would have made the mismatch obvious in seconds. Taking this one.

On the Wikidata requirement itself, that is one of the optional restrictions rather than something built into the feature. It can be applied to an article type or left off, so for en.wiki outlines that would be your call. We are also open to hearing which other restrictions would be feasible and make more sense than the ones we currently offer.

I'd make it three per WP:42, with a note encouraging editors to add more if possible

Fair, and the number should not really be ours to pick. The obvious answer is to make it configurable per outline, so a wiki can set two, three, five or whatever suits that article type, with the encouragement note alongside. Adding it to the list.

The source and notability guides were helpful

Good to hear, and worth knowing, since most of this thread has been about the steps before anyone reaches them. GGalofre-WMF (talk) 11:00, 29 September 2026 (UTC)

I have not tried this out but I think paste-check needs to be included in the process which I don't think is currently enabled in draft space. S0091 (talk) 20:02, 21 September 2026 (UTC)

S0091 Good point, thank you!

On the connection between the two, the edit checks and suggestions that appear in the visual editor will also appear for articles created through Article Guidance. They are switched off at the moment, and deliberately so: Article Guidance pre-loads a structure into the editor, and the checks were firing on that skeleton rather than on anything the editor had written, which is noise at best and misleading at worst. Getting them to fire on the editor's own content and not on ours is active work we're doing at the moment collaborating with the Editing team. GGalofre-WMF (talk) 11:03, 29 September 2026 (UTC)

New decline template reason: too detailed

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.

---

I've noticed a gap in the existing AfC decline reasons for drafts which are not necessarily essays, spam, or have npov issues - but are just way too detailed and indiscriminate to work as encyclopaedic article.

I'm thinking of drafts which document basically everything that can be sourced about a subject rather than being a concise summary of reliable secondary sources. At the moment, essay and npov come closest, but neither really describes this problem.

Would there be support for adding a separate decline reason, maybe detail, like the below:

This draft contains too much detail and does not provide a concise, encyclopaedic summary. Wikipedia articles should summarise the most important information from reliable sources and give each aspect of a topic avoid giving too much space to minor details. Rewrite the draft to:

• summarize secondary sources: the main body should summarise what reliable, secondary sources state about the topic;
• remove excessive detail: omit trivia, routine information, exhaustive lists, and minor events that are not important to understanding it.

Instead, only summarize in your own words a range of independent, reliable, published sources that discuss the topic.

qcne (talk) 18:31, 19 September 2026 (UTC)

Being too detailed is not a reason to fail a draft, in my opinion. If it's indiscriminate and/or just a random collection of too much detail, then it would probably trip not. Primefac (talk) 19:49, 19 September 2026 (UTC)

I agree. If a draft is skirting it, I suggest tagging with {{Overly detailed}} and leaving a comment with the hope they will clean it up some. If they don't but it otherwise meets notability, then it should be accepted.. S0091 (talk) 20:19, 19 September 2026 (UTC)

If that is a genuine problem with the draft, the reviewer should either remove some of the obviously excess content or tag it. I wouldn't !vote delete on an article with too much detail. The reviewing instructions state, If a problem would ordinarily be dealt with in mainspace through normal editing or a maintenance template, it usually does not justify an AfC decline. 🌀Hurricane Wind and Fire, why'd you decline my draft? (talk, contribs)🔥 22:15, 19 September 2026 (UTC)

I find myself agreeing with Primefac here, if a draft is good enough that having too much detail is its only problem, I think it can go into mainspace (with a tag). squawk7700 (talk) 10:13, 20 September 2026 (UTC)

Being "too detailed" is not an acceptable reason to decline. If it's notable and no other stopping issues, accept and tag. It will then be likely to be improved by others, while many notable subject declined just get abandoned and deleted. I always like to consider things from the readers point of view: If you look up a subject what is better, no article or a "too detailed" article? Frankly IMHO a bigger issue is "too little detail", we have way more very short stubs than we have overly detailed articles. KylieTastic (talk) 10:39, 20 September 2026 (UTC)

The thing is, an article doesn't have to be in perfect shape^ to be accepted for AFC. Drafts are generally written by new users and generally do not know about Wikipedia's manual of style (MOS) and may not understand policies like the neutral point of view (NPOV) and due/undue weight. It's likely that they are not intending to write a good article (as defined under WP:GA). As long as the subject is notable, it should be accepted even if it requires a lot of work.

^I'm defining that as an article that would not be quickfailed at GAN JuniperChill (talk) 20:43, 20 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.

What to do here?

I was scrolling through the CAT:AFC0 queue and noticed two very similarly named drafts: Draft:Artificial intelligence investment and Draft:AI(Artificial Intelligence) Investment. I declined the latter as dup. @Sulfurboy: declined the former three times ( and ) and rejected here, stating Tendentious resubmitting without improvement. Intentional removal of previous declines as well. Disruptive editing such as this is not tolerated on Wikipedia. I don't have any strong feelings on the reject, but it appears the user created the second draft to potentially bypass the rejection despite resubmitting the initial draft. Should I reject again and warn the user or review based on the draft's merits? 🌀Hurricane Wind and Fire, why'd you decline my draft? (talk, contribs)🔥 20:16, 20 September 2026 (UTC)

(edit conflict) Drafts are identical, so yes a decline of dup is appropriate. Keep ec'ing so I'll reply more later. Primefac (talk) 20:32, 20 September 2026 (UTC)

@Hurricane Wind and Fire while I'm not extremely experienced, I'd discuss your concern with the user, and wait for a reply before reviewing. In solidarity, Dafootballguy (Want to talk?) 20:23, 20 September 2026 (UTC)

Good luck, they're not really much for interacting based on their talk page. Primefac (talk) 20:32, 20 September 2026 (UTC)

Yeah they're clearly WP:NOTHERE for any reason other than to push this article through. Continual removal of previous declines, tendentious resubmission, and now resubmitting a rejected article. I just issued their third disruptive editing warning. Sulfurboy (talk) 22:10, 20 September 2026 (UTC)

Often the best thing to do with problematic editors who are making inappropriate actions, not being communicative, being promotional and being generally NOTHERE is to make sure they get escalating warnings. Once they get to a final warning and continue it is much easier for admins to act. Also I have seen many people over the years start to interact only once they get a several warnings, or a final warning, and sometimes issues are then resolved in a positive way. KylieTastic (talk) 22:31, 20 September 2026 (UTC)

I'm going to consider the latest warning by sulfurboy to be a final, I'll temp block if they continue into tomorrow. Primefac (talk) 22:51, 20 September 2026 (UTC)

@Primefac They resubmitted Draft:AI(Artificial Intelligence) Investment again after Sulfurboy's last warning, I went ahead and rejected. I think it's time for a temp block. 🌀Hurricane Wind and Fire, why'd you decline my draft? (talk, contribs)🔥 23:13, 20 September 2026 (UTC)

They've asked for help at AFCHD, so I think inroads have been made. Might be slow going but it's better than nothing. Primefac (talk) 09:19, 21 September 2026 (UTC)

AFC/R reviewing instructions

I've noticed that the AFC/R reviewing instructions could probably be improved quite a bit. I think I'd like to work on that, and I wanted to see if any other regulars wanted to help out.

I was also surprised that there's no specific talk page for AFC/R; it's largely a separate process. Would people be opposed to me making one? Skarmory (talk • contribs) 10:40, 21 September 2026 (UTC)

Excellent idea, I think itd be worth adding the AfC tabs to it too so it's all cohesive. I just re-wrote the AfC reviewer instructions so feel free to steal any ideas/formatting. qcne (talk) 16:57, 21 September 2026 (UTC)

Capitalization

Why is this capitalized as "Articles for creation" and not "articles for creation" or "Articles for Creation"? Doesn't make a lot of sense to me. Szxq (talk) 22:34, 23 September 2026 (UTC)

Wikipedians Love Sentence Case Whenever Possible Pennecaster (Chat with Senne) 10:16, 24 September 2026 (UTC)

Wikipedia titles by default are first-letter-capitalised, so the title of the page (regardless of whether you refer to it as articles or Articles) the displayed page name is still the same. Primefac (talk) 13:50, 26 September 2026 (UTC)

Potential plan for addressing the severe AfC backlog

If you aren't aware of this yet, there are currently about 2,220 pending AfC submissions awaiting review. Compared with last week, when there were roughly 1,900–2,000, this is a significant increase.

I propose lowering the undeleted article-edit requirement for AfC reviewers from 500 to 450 while the backlog remains this high. This doesn't have to be permanent. The requirement could return to 500 when the AFC submissions dip below 2,150

The reason behind 450 is because 250 and 300 are too low and would cause another crisis because even though there would be more people helping, but most would be inexperienced. Historian Felix (talk) 21:16, 26 September 2026 (UTC)

Are there large numbers of people with 450 edits clamoring to be reviewers? 331dot (talk) 21:18, 26 September 2026 (UTC)

I honestly have no idea. I dont know how to. If we could get that info, this would become much more clearer. Historian Felix (talk) 21:24, 26 September 2026 (UTC)

And, if it should be 450 now, it should be 450 all the time. As noted on the Teahouse, the backlog has been much higher in the past. 331dot (talk) 21:29, 26 September 2026 (UTC)

YTD there have been ~12 people who have failed the minimum criteria, most falling well short of 500 edits; I see little benefit in dropping the requirements. Primefac (talk) 21:34, 26 September 2026 (UTC)

is it only 12? If so I guess 450 edits wouldn't be an option Historian Felix (talk) 21:36, 26 September 2026 (UTC)

I thank you for offering an idea, but I don't think this will be effective. 331dot (talk) 21:40, 26 September 2026 (UTC)

alright I get that, sorry for the inconvenience Historian Felix (talk) 21:42, 26 September 2026 (UTC)

I really want to help with AfC and contribute more to Wikipedia, but I'm currently short of the required edits. Is there any way I could help before I reach 500 edits? Historian Felix (talk) 21:43, 26 September 2026 (UTC)

You can participate at WP:AFD, which is chronically starved for participants. This will also give you the experience needed to be an AFC reviewer. In solidarity, asilvering (talk) 22:14, 26 September 2026 (UTC)

thanks Historian Felix (talk) 22:16, 26 September 2026 (UTC)

also could substantial demonstrated AFC/AFD experience be considered as an alternative route for getting the 500 undeleted edits I think I have 300 undeleted edits I just want to know because i really want to help. I know what its like to get a decline for your article I've been declined like that more than 5 times Historian Felix (talk) 22:27, 26 September 2026 (UTC)

500 edits is a hard requirement. Also, if you've already had 5 draft articles be declined, that's actually a good indication you probably shouldn't be an AfC reviewer right now. If you aren't successfully applying the requirements to your own drafts, how could you determine if somebody else is? Athanelar (talk) 22:53, 26 September 2026 (UTC)

some of my articles have been accepted so I fyoure saying I don't have experience, I think that's not the best statement. What you said hurt me a little and I'm deciding if I even want to be an AFC viewer anymore. I just wanted to know, not cause drama Historian Felix (talk) 23:18, 26 September 2026 (UTC)

It's not wrong to say you don't have experience; 400-ish edits is nothing in the grand scheme of things. It's also not a bad thing, and not something you should be offended by. I'm past 7000 edits now and still have a lot to learn. Athanelar (talk) 23:46, 26 September 2026 (UTC)

its alright I just wanted to know because I really do want to help AFC drafts but if I cant I guess I could keep working to gain your guy's trust to be an AFC reviewer Historian Felix (talk) 23:48, 26 September 2026 (UTC)

Back log drive awards...

Hello! I don't know where else to ask, but when are the awards being given out. I don't mean to seem selfish or anything, but I realized I had enough reviews for one. Could someone tell me please. Thank you, happy editing! Robloxguest3 (talk) 01:11, 27 September 2026 (UTC)

Robloxguest3, looking at the most recent backlog drive page I see 9.5 points for you and it looks like the first award is at 15 points. LittlePuppers (talk) 01:16, 27 September 2026 (UTC)

Interesting, i've done over 75 reviews during the backlog drive. Perhaps maybe they reviewed my reviews ^[Humor] early. Thanks. Robloxguest3 (talk) 01:22, 27 September 2026 (UTC)

Hi @Robloxguest3, the backlog drive only counts reviews performed during the drive, which in this case was for the month of August. Looking at your logs, whilst you have done 75 reviews, only 8 of those took place in August.

Even though your July and September reviews didn't count towards points in the backlog drive, please don't feel that your efforts are unappreciated – they most definitely still are! nil ^nz 02:56, 27 September 2026 (UTC)

FYI: edit check enabled in draft namespace

Me and Pppery have enabled all edit checks in the draft namespace. This means new editors now get a pop-up whenever they forget a reference, or more importantly for reviewers, when they paste text from elsewhere. If an edit check gets shown, a tag gets added to the diff.

• Note that the LLM paste check v1 cannot detect all LLMs pastes. It's based on meta-data, which is ofen absent, not a statistical analysis of burstiness etc that AI detectors usually use. So if it gets flagged, it means somebody is highly likely to have used an LLM. If the edit is tagged with a normal paste check instead, there is still a very good chance that the edits are from LLMs
• SuggestionMode is also enabled, so drafters are now more encouraged to keep on improving their drafts. In solidarity, —Femke (talk) 🐦 07:19, 28 September 2026 (UTC)

 Courtesy link: Wikipedia:Edit check... for anyone else wondering WTH 'edit check' is. (No? Just me, then.) Yours cluelessly, -- DoubleGrazing (talk) 07:36, 28 September 2026 (UTC)

Reviving WikiProject Notability?

I wondered if anyone was interested in reviving WikiProject Notability in the hope that we might be able to start clearing the backlog of 55,000+ articles that are tagged for questionable {{notability}}. I asked a while ago on the WikiProject's talk page, but the silence was deafening. So I'm widening the search. If you are interested, huge thanks, and please add your name to the list of members so that I can gauge support. And do publicise the notion elsewhere if you think it would be useful. Posted to Wikipedia talk:Notability, Wikipedia talk:WikiProject Articles for creation, Wikipedia talk:New pages patrol/Reviewers, Wikipedia talk:WikiProject Unreferenced articles. Cheers, SunloungerFrog (talk) 12:09, 29 September 2026 (UTC)

Replies to declines embedding in the decline template

We have had this issue before and I think it was a simple fix. When one clicks "Reply", it embeds in the template rather than below it. See User talk:RadioStPete#Your submission at Articles for creation: Radio St. Pete (September 29) for an example but I have seen this many times. S0091 (talk) 17:54, 29 September 2026 (UTC)

This seems to have happened in this example because they replied to your AfC comment. I assume this is because there is a signature there, so DiscussionTools automatically decides it should be reply-able. Not sure how to address this. Axolitl (talk | contribs) 21:27, 30 September 2026 (UTC)

Userscript to check for hallucinated sources

I've made (vibecoded) a userscript which checks for hallucinated and nonexistent sources at User:Fermiboson/sourcecheck. Questions, suggestions and advice are welcome. Fermiboson (talk) 04:40, 30 September 2026 (UTC)

Just a cheeky comment, but I love the irony of using AI-generated code to find AI-generated code. Primefac (talk) 14:34, 3 October 2026 (UTC)

This is not lost on me and by no means am I claiming that vibecoding is a replacement for actual development… but we needed these tools six months ago. Hopefully this need will be increasingly obviated as the WMF starts paying more attention. Fermiboson (talk) 15:23, 3 October 2026 (UTC)

Oh, I get it; simple tools are what AI is good at (provided they're still checked/verified by someone breathing) and I cannot find a ton of fault in that sort of thing. Primefac (talk) 15:30, 3 October 2026 (UTC)

New comment by submitters on AI use

Hey folks, just flagging this for people. I just went through a few drafts in CAT:AFC0, and several drafts have a comment by the submitting editor disclosing their AI use and how they used it for the draft. Don't know how we are judging this per the AI policy, but it's probably a case-by-case basis.

Examples: Draft:Mumbo Jumbo (YouTuber) Draft:"Frien-do" Draft:Peter Zhmutski

Keep up the good work y'all! Bkissin (talk) 00:55, 2 October 2026 (UTC)

@Bkissin I would decline all those drafts as AI. In solidarity, Dafootballguy (Want to talk?) 01:16, 2 October 2026 (UTC)

See the above discussion Wikipedia talk:WikiProject Articles for creation#h-New field for AI use disclosure at AfC Submit Wizard?-20260701102700; editors are now being asked to disclose their AI usage when they submit a draft. Helpful Raccoon (talk) 01:18, 2 October 2026 (UTC)

Ok, but when we see (eg) {{AfC comment|1=In preparing this draft, I disclose that AI assistance was used as follows: <nowiki>I used an AI tool to help translate and format my original historical documents and family archives into Wikipedia's standard markup. <!--Comment automatically added by the Submission Wizard--> [[User:Example|Example]] ([[User talk:Example|talk]]) 13:35, 3 October 2026 (UTC)}} what are we expected to do about it. AI is AI, isn't it? 🇵🇸‍🇺🇦 Fiddle^Timtrent Faddle^Talk to me 🇺🇦‍🇵🇸 16:16, 3 October 2026 (UTC)

I view it almost as self-incrimination, which is fine by me! Bkissin (talk) 18:42, 3 October 2026 (UTC)

That is precisely how I am treating it.

I used the wizard at WP:AFC to see when and how this was presented to me, and I failed to find it. I just declined a submission in Marathi language where AI use was declared. 🇵🇸‍🇺🇦 Fiddle^Timtrent Faddle^Talk to me 🇺🇦‍🇵🇸 18:51, 3 October 2026 (UTC)

Perfect, thanks all! I figured it might have been an additional layer of security added to the wizard, but wasn't sure. Bkissin (talk) 13:40, 2 October 2026 (UTC)

Odd help desk behaviour

There's something wrong with the most recent sections at AFCHD, Wikipedia:WikiProject_Articles_for_creation/Help_desk#06:52,_2_October_2026_review_of_submission_by_Assekoas and Wikipedia:WikiProject_Articles_for_creation/Help_desk#07:25,_2_October_2026_review_of_submission_by_Coopersworld100. Some comments are going into wrong section. When you try to edit the source of the earlier section, it opens the latter instead, and when you try the latter it says 'Cannot find section'. I tried to fix things by editing the source for the date, but that only shows today's first thread for some reason. And when I looked at the source for the whole page, I can't see anything wrong with it. Any ideas? -- DoubleGrazing (talk) 08:21, 2 October 2026 (UTC)

A template with typo in closing a template meant that = October 1 = was treated as text not a header. Should be fixed now. Primefac (talk) 14:41, 3 October 2026 (UTC)

Huh. Thanks. My old eyes couldn't spot it. -- DoubleGrazing (talk) 14:50, 3 October 2026 (UTC)

Heh, still took me longer than I'd like to admit; even with syntax highlighting on in the source editor, I still had to check line-by-line 'til I found it... though finding the section itself was just a case of finding the off-by-one edit link and going one higher. Primefac (talk) 15:05, 3 October 2026 (UTC)