Wikipedia:Village pump (technical)

 Policy Technical Proposals Idea lab WMF Miscellaneous 

• Table of contents
• First discussion
• Jump to bottom
• New topic
• All village pumps

The technical section of the village pump is used to discuss technical issues about Wikipedia. Bug reports and feature requests should be made in Phabricator (see how to report a bug). Bugs with security implications should be reported differently (see how to report security bugs).

If you want to report a JavaScript error, please follow this how-to page. Questions about MediaWiki in general should be posted at the MediaWiki support desk. Discussions are automatically archived after remaining inactive for 5 days.

« Archives, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 231, 232

Several AMPOL pages are unable to load on mobile

Since shortly after the 2026 White House Correspondents' Dinner shooting, the wikipedia articles for that page, the Charlie Kirk page, Assassination of Charlie Kirk, and Attempted assassination of Donald Trump in Pennsylvania have all been unable to load on mobile, at least on my iphone. Whenever I open any of these four pages, it says "a problem occurred repeatedly" after being open for a couple seconds. I'm not sure what is causing this issue; I suspected it was length that was the problem, but Donald Trump's own wikipedia page loads just fine on mobile so I'm not sure. This issue persists whether logged out or in, and adblocker on or off. Unnamed anon (talk) 08:17, 13 September 2026 (UTC)

Which browser do you use? Could you try another one? Could you try clicking Special:Random until you find a few other pages that crash? Could help in finding out what the pages have in common that could cause the problem. — Chrisahn (talk) 11:24, 13 September 2026 (UTC)

I use Safari on ios 17. I do not currently have another browser on my phone, though I have not yet noticed major issues on Chrome on my PC. I clicked random many times and did not find any other pages that crashed. Unnamed anon (talk) 17:26, 13 September 2026 (UTC)

I have tested something by going onto desktop mode on my iphone. Initially, it appeared that it fixed the lag issue. However, the pages all experienced some slight lag, and crashed after about a minute (a significant improvement over crashing in 5 seconds though). Unnamed anon (talk) 17:41, 13 September 2026 (UTC)

The three pages you mention there are all too long and need to be reduced. Your phone is probably just choking on their length. Izno (talk) 18:44, 13 September 2026 (UTC)

If anybody starts a split suggestion for Tyler Robinson, I would absolutely support it, partially on the basis of letting the assassination page load properly, partially because Robinson has enough notability. Unnamed anon (talk) 22:18, 13 September 2026 (UTC)

Sounds reasonable and that's what the OP suspected, but OP also said the page Donald Trump loads fine, although it's much longer than 2026 White House Correspondents' Dinner shooting. — Chrisahn (talk) 19:36, 13 September 2026 (UTC)

Yeah, that is weird. Izno (talk) 20:18, 13 September 2026 (UTC)

That is interesting—and it's not like only one aspect of the page is longer, there's more text, more images, more references, etc. Perhaps the only difference I notice is that Donald Trump, unlike the others linked, has no videos? Do other pages with videos crash for you? LittlePuppers (talk) 21:13, 13 September 2026 (UTC)

I had a sneaking suspicion that might be related. Let's try pinging the resident expert @TheDJ. Izno (talk) 21:34, 13 September 2026 (UTC)

None of them crash for me, but I’m on iOS 26. Could be some sort of iOS CSS measuring bug of course, but I don't have that browser version anywhere any longer :( —TheDJ (talk • contribs) 21:56, 13 September 2026 (UTC)

@LittlePuppers: @Izno: @Chrisahn: I think the pages having multiple videos (in combination with the length) might be the issue. Killing of Iryna Zarutska also crashes, and that article also has 2 videos. Kirk's page has 3 videos, the page on the murder of Kirk has 4 videos, and the pages on the first and third attempts on Trump's life have 3 videos each. Interestingly, Attempted assassination of Donald Trump in Florida loads properly in my experience, and only has one video. Unnamed anon (talk) 22:19, 13 September 2026 (UTC)

That definitely makes sense. Although I would think that (at least ideally) an unloaded video shouldn't take many more resources than an image. LittlePuppers (talk) 01:38, 18 September 2026 (UTC)

@LittlePuppers: @TheDJ: Add another mobile crashing AMPOL page to the list: I just found out that 2025 Brown University shooting also crashes on mobile. I currently only have access to my iphone; does that page also have a video? Unnamed anon (talk) 07:06, 20 September 2026 (UTC)

Yes, that page has one video. In any case, iOS 17 is quite old. Are you able to update to a newer iOS version or try another browser on your phone? Some1 (talk) 11:57, 20 September 2026 (UTC)

I updated to ios 26 yesterday (not specifically due to this wikipedia issue, mostly because the apps for my medical care and work now require at least ios 18), and so far, this appears to have fixed the AMPOL page crashes. From testing, Kirk's page, the page about his murder, and the page on Trump shooting #3 all load properly on my phone now. If they crash again, I'll come back to let you know. Unnamed anon (talk) 06:44, 29 September 2026 (UTC)

@Unnamed anon So you have had this issue since April ? —TheDJ (talk • contribs) 21:53, 13 September 2026 (UTC)

Around April, May, or June. I remember that in 2024 and 2025, the first Trump Assassination page loaded just fine. Then, the third attempt happened, and that page loaded fine for only about a month before multiple AMPOL pages started crashing. Unnamed anon (talk) 22:18, 13 September 2026 (UTC)

Enabling iFrame graphics from Our World in Data on English Wikipedia

Note: Moved to Wikipedia:Village pump (proposals) per multiple requests.

Loss of session data

Tracked in Phabricator
Task T423206

I am getting a lot of Sorry! We could not process your edit due to a loss of session data. Please try saving your changes again. If it still does not work, try logging out and logging back in error messages, with seemingly no rhyme or reason to which edits they apply - any guidance? GiantSnowman 18:42, 19 September 2026 (UTC)

Annoyingly all the time. See phab T423206. KylieTastic (talk) 18:49, 19 September 2026 (UTC)

Eurgh, great, thanks! GiantSnowman 19:00, 19 September 2026 (UTC)

Me too and for me it has been going on for a while, like a a year. It is getting to the point I don't edit anymore because I constantly have to long back in. S0091 (talk) 19:46, 19 September 2026 (UTC)

I don't have to log back in, I just have to try to publish two or more times till it works. Mostly it's when editing on multiple tabs at once. It's not just this issue though, in general editing is just slower and more problematic and keeps making me stop trying. It used to be the limit on the editing was me the human in the loop, but now it is the slow and buggy tech. KylieTastic (talk) 20:37, 19 September 2026 (UTC)

So damn annoying and yes, It's most often when I have multiple tabs open. Had one page last night that took well over 5 submits to take. Firefox (up to date), Windows 11. Zinnober9 (talk) 15:30, 21 September 2026 (UTC)

I've no guidance, but it's not just you. It's been particularly bad for me today, with about half my save attempts getting that error. Certes (talk) 20:00, 19 September 2026 (UTC)

Same here Οἶδα (talk) 02:59, 20 September 2026 (UTC)

I'm having the same issue today, for the first time ever. -- Brad (talk) 04:43, 21 September 2026 (UTC)

FYI

 – Merged consecutive sections —ClaudineChionh (she/her · talk · email)

I am constantly getting the error "Sorry! We could not process your edit due to a loss of session data. Please try saving your changes again. If it still does not work, try logging out and logging back in." I have tried logging out and back in and nothing doing... Anyone seeing this issue? Any suggestions? Zackmann (^{Talk to me}/_{What I been doing}) 03:29, 20 September 2026 (UTC)

Yes, in the thread immediately above yours (I'll merge them now to keep this tidy). —ClaudineChionh (she/her · talk · email) 03:38, 20 September 2026 (UTC)

Facepalm wow... Way for me to not read, at all.... Thanks ClaudineChionh! Zackmann (^{Talk to me}/_{What I been doing}) 03:41, 20 September 2026 (UTC)

As I've commented in threads before, I've had occasional log-outs of the "page loads not logged in, with a popup saying 'you are now logged in, refresh to fix it' variety. That's been stepping up the pace yesterday and today (a "normal" pace was once per day; it's happened twice in the last hour), and yesterday I was also getting the mentioned 'loss of session data' a few times while trying to save edits. Good to know it's not just me. - The Bushranger _{One ping only} 07:34, 20 September 2026 (UTC)

This has not been a regular/constant issue for me, just the last few days. GiantSnowman 12:09, 20 September 2026 (UTC)

Had it happen fourfivesix times over the last hour (once on Commons, once on Wikidata, twicethricefour times on en.Wiki) where loading a page had me logged out. - The Bushranger _{One ping only} 23:00, 20 September 2026 (UTC)

...and again. This is becoming functionally unusable. - The Bushranger _{One ping only} 23:08, 20 September 2026 (UTC)

Just adding my own report to the list. It happened very frequently to me yesterday. Suðurhafsljósæta (talk) 05:46, 21 September 2026 (UTC)

Agree Wikipedia is becoming functionally unusable that this point and now for past couple days I am getting logged out even navigating from one page to another. I have now been logged out over 10 times in the last hour (and logged out trying to post this). S0091 (talk) 18:14, 21 September 2026 (UTC)

Tracked in Phabricator
Task T439139

As it's now at the point where I'm being logged out every third or fourth edit, I am no longer interested in volunteering my time blocking vandals, working SPIs as a checkuser or oversighting material when the website is working in such a pitiful way. I'll hope for better tomorrow, but am doubtful.-- Ponyo^{bons mots} 19:12, 24 September 2026 (UTC)

@Ponyo: That's curious, because for everyone else, it seems to have cleared up. - The Bushranger _{One ping only} 23:31, 24 September 2026 (UTC)

I also have been experiencing it today, and I heard from a few others also the same. Izno (talk) 23:33, 24 September 2026 (UTC)

I did have it happen once earlier, but 'once per day' is par for the course for me. I did notice the charts on the phab pages did show a tick up today a bit. - The Bushranger _{One ping only} 02:36, 25 September 2026 (UTC)

(Also as a note: from that second phab ticket, some users lost their local session but not their auth.wikimedia.org session is exactly what happens to me in 99% of the cases. - The Bushranger _{One ping only} 02:38, 25 September 2026 (UTC))

Same, same. Carlstak (talk) 02:45, 25 September 2026 (UTC)

Exactly. I'm logged out locally but not centrally and I have to refresh a time or two in order to regain my local login. This is less than ideal when oversighting material or doing multi-sock blocks. Given the potential to accidentally expose private data I generally back out altogether and redo the "paperwork" from scratch. -- Ponyo^{bons mots} 15:22, 25 September 2026 (UTC)

My account is automatically logout

On the Recent Changes Page and Pending Changes page, I'm automatically logged out and then logged back in a few seconds. Is there an issue my account? Thanks. ~🌀Ampil ^{「💬 / 📝」} 08:37, 20 September 2026 (UTC)

merged thread NightWolf1223 <Howl at me•My hunts> 14:32, 20 September 2026 (UTC)

Is this happening with any browser? ~2026-51340-95 (talk) 12:43, 24 September 2026 (UTC)

New TA randomly created in between one edit and another?

I was editing as ~2026-50825-19 and upon undoing a seemingly hoax edit Wikipedia randomly created a new Temporary Account (this current one).

Is it normal behaviour for a new TA to be created even when you already have one assigned? ~2026-50785-44 (talk) 23:31, 20 September 2026 (UTC)

Probably the same thing as the other session loss issues discussed above. Anomie⚔ 23:42, 20 September 2026 (UTC)

This is still occurring. I have been logged out 5 or so times today. sometimes while editing and sometimes not. S0091 (talk) 21:31, 28 September 2026 (UTC)

Redlinked categories, again

The latest run of Special:WantedCategories again features a small cluster of redlinks I cannot fix myself.

• Category:"+f+" and Category:"+x+", both on a user's .js settings page I cannot edit to remove them due to lacking the necessary privileges.
• Category:Cite Q - cites a work with an expression of concern notice, a category that's present only on private template testing pages and cannot be created for that purpose, and couldn't be named that way even if its creation were justifiable anyway, but which is being smuggled in via a template or module I can't find. (I found a template edit I thought was responsible for it, because it was made within the past three days by the same editor whose testing pages are popping up in the category, and reverted that only to find that the reversion failed to make the redlink go away, and then I got completely lost in the woods trying to find where else the problem lies.)

So could somebody with more knowledge and/or privilege in such matters clear these redlinks? Thanks. Bearcat (talk) 07:23, 22 September 2026 (UTC)

The first two are only on User:Asukite/scripts/NPPTools.js and Asukite is an active administrator who can fix it. The last has now been created. Special:PrefixIndex/Category:Cite Q - shows eight categories (two are category redirects) so maybe Template:Cite Q#Tracking categories should be updated. PrimeHunter (talk) 14:43, 22 September 2026 (UTC)

Category:Cite Q - cites a work with an expression of concern notice is due to all four rows at Template:Cite Q/testcases#Expression of concern notices. This must be due to some recent edits in Template:Cite Q. --Redrose64 🌹 (talk) 20:06, 22 September 2026 (UTC)

Thanks for the ping, I've been inactive the last couple days. I just blanked the JS for now, I have a few other issues with that script, least of which is the fact that I never added any functionality to look for session loss before saving (JS is not a strength of mine and I've been trying to learn) - the issue wasn't present in my original source, that was an error introduced by a minifier I used. ASUKITE 23:32, 22 September 2026 (UTC)

Late to this but @Asukite, I think adding

// <nowiki>

before the first line of the script and

// </nowiki>

after the last line ought to solve that particular issue. Accessedgrant (Epicgenius mobile alt) (talk) 02:18, 28 September 2026 (UTC)

Thanks! I left that on the back burner as I've been working on something else. The original source used string concatenation to avoid this issue, and naturally the minifier just undid that. ASUKITE 16:37, 28 September 2026 (UTC)

LLM Paste Check - live later this week

On Thursday this week, LLM Paste Check will be enabled here. It is an Edit check that will appear when someone pastes text into Wikipedia that has been copied directly from an external generative AI chatbot. It can be used to point them directly towards the local policy on AI use.

The initial implementation will activate when both of the following conditions are met:

1. Someone pastes ≥50 characters or at least 100 characters are entered at the same instant, consisting of at least 10 words
2. The pasted text is accompanied by metadata popular chatbots are known to add

Any edits that trigger an LLM Paste Check to be shown will apply an Edit tag to the revision (editcheck-llm-paste-shown), to enable community monitoring.

The wording shown in the description, and the default-link it uses, can and should be locally configured. The default wording and link can be found (after September 24) and overridden at MediaWiki:editcheck-copyvio-llm-description and MediaWiki:editcheck-copyvio-llm-policylink, and the defaults can be previewed at testwiki.

As with all edit checks, admins can configure who sees it (currently based on min/max editcounts) and where the Check is and is not shown. The default is showing it to all editors.

Feedback here or mw:Talk:Edit check is welcome and appreciated. Quiddity (WMF) (talk) 22:44, 22 September 2026 (UTC)

@Quiddity (WMF) does this work in all namespaces or only article? Asking because as an AfC reviewer it would be very helpful in Draft space. S0091 (talk) 17:33, 24 September 2026 (UTC)

Yes it can! As of later this week, communities can enable any Edit Check or Suggestion in additional namespaces. It'll be announced in next week's Tech News. Quiddity (WMF) (talk) 17:39, 24 September 2026 (UTC)

@Quiddity (WMF) Awesome! Another question. why do you guys announce this stuff here rather than VPWMF page?. Things like this impact editors across the board, not just tech editors. S0091 (talk) 19:20, 24 September 2026 (UTC)

Partially because it's a technical change, partially because the feature requires local technical configuration for optimal usage, partially because this page has more watchlisters, plus it's simply where I'm accustomed to reading and posting about technical changes for 20+ years. Quiddity (WMF) (talk) 21:30, 24 September 2026 (UTC)

I've made fairly substantial configuration changes to try to bring the wishy-washy wording more in line with enwiki's strict AI policies. In particular I reworded MediaWiki:Editcheck-copyvio-llm-keep-generated which defaults to "I used AI to generate this text and I have confirmed it complies with this wiki's AI policy." to point only to the specific LLM-assisted translation exception, since we only allow AI for that and minor spelling changes which already have their own exceptions. I would also suggest deleting the "none of the above" option entirely; allowing "other" reasons is contrary to enwiki's rules entirely and I hope that denying them the opportunity to weasel out like that is more likely to make people rethink what they are doing. * Pppery * _{it has begun...} 04:45, 26 September 2026 (UTC)

Also why do we have "I wrote this text myself. It is not AI-generated" at all. Can someone think of how this LLM markup-paste check could produce false positives? I can't. * Pppery * _{it has begun...} 04:46, 26 September 2026 (UTC)

As mentioned at Wikipedia talk:WikiProject AI Cleanup#c-DLynch (WMF)-20260916171100-Kowal2701-20260916142100, “it’s a little fuzzy”. Pastes from ChatGPT are probably higher confidence than the pastes from MS Word. I think the idea is to try the paste check and review how it does. Dw31415 (talk) 08:15, 26 September 2026 (UTC)

Okay, that's fair. I still would suggest dropping "none of the above" though; either you didn't use AI, you used it in a way that fits in the spelling exception, you used it in a way that fits in the translation exception, or you are violating enwiki's policies. There's no fifth option. * Pppery * _{it has begun...} 15:46, 26 September 2026 (UTC)

agreed Kowal2701 (talk, contribs) 18:49, 26 September 2026 (UTC)

The default text was deliberately crafted and discussed. Maybe revert your bold change and open a new discussion on it. Dw31415 (talk) 08:21, 26 September 2026 (UTC)

It's okay, the default text was targeted at all wikis, including ones that permit LLM-generated content Kowal2701 (talk, contribs) 12:36, 26 September 2026 (UTC)

I suggested a similar text to Pppery's on phab (phab:T438521), which hadn't made it over to production yet (?). The new text gives more clarity about expectations, and will therefore be less likely to lead to subsequent biting. In solidarity, —Femke (talk) 🐦 10:27, 27 September 2026 (UTC)

So what are the differences between this and the existing paste check feature? Different messages; will this catch more? LittlePuppers (talk) 04:58, 26 September 2026 (UTC)

It will catch more use. It’s now detecting meta-data as mentioned above. Dw31415 (talk) 07:55, 26 September 2026 (UTC)

What is the purpose of #editcheck-llm-paste (hidden tag)? I mean, I can filter on it in recent changes but because it is hidden there is no record so you do not know an a edit was flagged by looking at someone's contributions or a page's history. S0091 (talk) 16:41, 26 September 2026 (UTC)

Just a guess: Since there may be false positives, it’s better to get some experience with it before making it more visible. Dw31415 (talk) 18:40, 26 September 2026 (UTC)

If so, it would be helpful to know that so we can feedback. S0091 (talk) 18:57, 26 September 2026 (UTC)

Looking at recent changes, I don't understand as clearly as I thought I did. The editcheck-llm-paste-shown is visible in the edit summary, while editcheck-llm-paste is not. I don't have a chance just now to reread the docs. Dw31415 (talk) 19:19, 26 September 2026 (UTC)

The docs at mw:Edit check/LLM Paste Check#Edit_tags are now updated to clarify when/why the hidden tag might be added.

Sidenote: As you can see from the results of the hidden tag, it's also detecting when the Check might be shown in other namespaces. You all/An admin may wish to add "llm-paste": {"extraNamespaces": {"draft": true}} (with appropriate commas) into the local json config (MediaWiki:Editcheck-config.json). HTH. Quiddity (WMF) (talk) 02:25, 27 September 2026 (UTC)

Thanks. That explains the two tags very well! Dw31415 (talk) 02:50, 27 September 2026 (UTC)

Done that and also enabled it in user namespace. * Pppery * _{it has begun...} 02:56, 27 September 2026 (UTC)

I've also enabled it for the generic paste check. Is there a simple way to enable all edit checks in draft? Or should I add this to all checks individually? In solidarity, —Femke (talk) 🐦 10:29, 27 September 2026 (UTC)

Should we enable this paste check in all talk namespaces, per WP:AITALK? And since Wikipedia space often hosts discussions, should we enable it there? What about Template space, where DYK discussions still live? – Jonesey95 (talk) 13:37, 27 September 2026 (UTC)

I don't think we can yet. The highest priority for me is to have it enabled as part of discussiontools on talk pages (phab:T429515). The normal paste check wouldn't work there, as it's normal for people to paste long quotes. Maybe a paste check with a much longer minimum length can work, as well as the normal llm-paste-check. In solidarity, —Femke (talk) 🐦 14:30, 27 September 2026 (UTC)

There is! If you check out the Edit_check/Configuration#Defaults_for_all_checks entry, there's a magic check-name * in the config. Anything you set on it gets slotted in for all checks when we build the config (which for each check is the sum of: base defaults + check's defaults + all-check config + specific-check config + specific-rule config for textmatch).

(The main thing we're missing in this config system is more-complicated composition. I.e. there's not currently a way for you to use set the all-check config are, then just add/remove a namespace to one check, unless you completely spell out the whole set of namespaces on that one check's config. The namespace-config is already implemented halfway to letting you set different config values in different namespaces, and I suspect fancier composition would be a prerequisite for finishing that up...) DLynch (WMF) (talk) 23:27, 27 September 2026 (UTC)

It's maybe useful to consider that editcheck-llm-paste-shown without editcheck-llm-paste is the "this was a successful educational intervention" case. It means someone was shown the check, and decided to remove the content as a result. DLynch (WMF) (talk) 17:14, 30 September 2026 (UTC)

That does require the patroller to understand that. Would it make more sense to have a single tag that's either "edit-check-llm-retained" and "edit-check-llm-removed"? Difficult to think of a good educational intervention to ensure patrollers understand and avoid biting. In solidarity, —Femke (talk) 🐦 17:52, 30 September 2026 (UTC)

Hmm. You lost me there. I understand the paste tag won’t be recorded if the metadata is removed, but do we know the generated text was removed? Dw31415 (talk) 18:27, 30 September 2026 (UTC)

DLynch (WMF), I posted something similar on Discord yesterday, but I'm noticing that neither check (paste or paste-llm) got tagged in a couple of WP:G15s (pages deleted for unambiguous LLM content). They were all created after we enabled these two checks, and with the VE tag.

• Draft:NJ_Morris had a separate LLM tag of 'possible AI generated citations, but not the paste check
• Draft:Sainik Kshatriya also had the VE tag, but no paste check
• User:SOMDUTT JI same
• Can you see what is going on there? I've not looked at G15s in mainspace to see if there is a similar pattern. None of the 4 articles in draft and user space created in VE I checked had the tag, so I don't think this is a matter of copying to a word processor first. In solidarity, —Femke (talk) 🐦 19:23, 30 September 2026 (UTC)

AFAIK it's only when people accidentally copy over stuff beyond the text Kowal2701 (talk, contribs) 20:24, 30 September 2026 (UTC)

I believe it's the paragraph breaks that can give this metadata? But anyway, in these edits there was also no normal paste-check tag, which there should have been if we were correct in our G15s. On enwiki, we've adjusted the wording for the paste check to also explain LLM policies to users. If we ever get a foolproof llm paste edit check, we'll have to adjust that text, but for now, both help us against llm misuse. In solidarity, —Femke (talk) 🐦 20:48, 30 September 2026 (UTC)

Note that the current detection only extends to those five specific chatbots. Plus (as DLynch notes below) it does also rely on the detectable-strings being within the content, and it's not always contained within short amounts of text; Two paragraphs worth of content seems to be a consistent way to trigger it. HTH. Quiddity (WMF) (talk) 20:10, 1 October 2026 (UTC)

I tried to get it to fire for copy/paste from the LLM that I subscribe to (need it for work), but none of the ~4 copies I tried triggered the check. Looking at my raw clipboard data, there was no good meta data in there. Used the following to inspect my clipboard: swift -e 'import Cocoa; if let d = NSPasteboard.general.data(forType: .html) { print(String(data: d, encoding: .utf8) ?? "Could not decode HTML") }'. I can provide more details by email or on Phab if that's helpful to WMF folks. Dw31415 (talk) 20:58, 30 September 2026 (UTC)

I think this site might capture the metadata on paste. I haven’t confirmed yet: https://wysiwyghtml.com/ Dw31415 (talk) 16:49, 1 October 2026 (UTC)

This site doesn’t capture everything. There’s much more data when I check the clipboard with the swift command above. Dw31415 (talk) 17:49, 1 October 2026 (UTC)

For what it's worth, I just loaded this to toolforge. https://link-remover.toolforge.org/paste-html.html It allows you to view the raw html in your clipboard. I'm goint to ask the one user to recreate the paste from "MS Word". Let's see what that yields. Dw31415 (talk) 01:33, 2 October 2026 (UTC)

The tricky part is that it's really dependent on exactly what you copy. We look for markers, but there's plenty of ways to get through without the markers ever being exposed to us, either because they weren't included in what was copied in the first place, or because the user pasted in a way that stripped them out before we could see them. (And you'll get differences depending on e.g. whether you clicked a chatbot's "copy" button in its output, or manually selected content and pressed ctrl+c...) DLynch (WMF) (talk) 17:44, 1 October 2026 (UTC)

Figured out why it didn't flag. I accidentality turned paste check off while enabling the other edit checks. Should now work with paste(llm) enabled in user+draft, and all the others just in draft. In solidarity, —Femke (talk) 🐦 18:42, 1 October 2026 (UTC)

I'm not sure what to do with this edit and then talk page discussion. The user claims to have edited in MS Word and then pasted back into editor. Is there a place they can paste so we can see the html? I reviewed the edit and didn't see signs, but I'm newer at it. Dw31415 (talk) 20:28, 30 September 2026 (UTC)

that user's also added chatgpt parameters , and AFAIK Paste Check doesn't trigger for MS Word (or Google Docs). People regularly lie about LLM use, to incredible lengths, I don't get it Kowal2701 (talk, contribs) 20:36, 30 September 2026 (UTC)

Paste-check is intended not to trigger for Word (if I'm reading the code correctly) but that relies on some other metadata being present. I'm curious to see if the copilot metadata just starts showing up in Word pastes and tripping the LLM. Dw31415 (talk) 20:53, 30 September 2026 (UTC)

<math> visual issue

i don't know where else to put this but the "<math>" tag is incredibly hard to read when in dark mode. I don't know if this is the right place to put this but idk where else I would. I would prefer it be a readable white color instead of literally dark black blending in with the background. Caleb's World11 (talk) 17:59, 25 September 2026 (UTC)

@Caleb's World11 How are you entering dark mode? The built-in Dark Mode in Vector 2022 and Minerva works just fine with the <math> tags. If you're using a 3rd-party gadget, you'd have to contact the author of that gadget. See mw:Manual:Dark_mode. --Ahecht (TALK
PAGE) 19:33, 25 September 2026 (UTC)

Sorry for the late reply, I am using the built in method on right of the screen, I think its either a chrome os issue or a chrome extension issue, so its probbily nothing to worry about honestly. Caleb's World11 (talk) 14:20, 28 September 2026 (UTC)

yeah this is an issue with my extentions, you can ignore this section. I would delete this section but can't obviously. Caleb's World11 (talk) 22:02, 28 September 2026 (UTC)

2026 United States elections ref issues

Since September 26th, this article is using Template:excerpt in 3 of its sections. The excerpt template has ported over lead section content from the 3 other articles but the complete references did not follow. Right now there are 47 Harv errors that seemingly stand in the way of easy WP:Verifiability. Tracing the sources stops at the excerpt code. If a reader does happen to click on each individual "excerpt from" phrase, they are then taken to the original content and can possibly find the sources that way. Is there an easy/elegant way to perhaps fix these truncated referencing issues? It took me a while to figure out what was happening so I thought I'd ask here. - Shearonink (talk) 02:24, 28 September 2026 (UTC)

The editor who added the excerpt should have copied over the required cites, they're not optional. I've copied over all the ones that were needed. There's no technical solution for this other than copying them over. -- LCU ActivelyDisinterested «@» °∆t° 16:52, 28 September 2026 (UTC)

Thank you. I came upon the previous iteration late in my day and was flummoxed on how to fix it. Thx again. - Shearonink (talk) 17:20, 28 September 2026 (UTC)

Tech News: 2026-40

Latest tech news from the Wikimedia technical community. Please tell other users about these changes. Not all changes will affect you. Translations are available.

Updates for editors

• Users of default skins on desktop and mobile web (Vector 2022 and Minerva respectively) will now be informed if the search suggestions displayed as they type into Search come from page redirects. Previously, users could get confused when the top search suggestions did not match exactly what they had typed. The new redirect notice clarifies that these suggestions are not random, but related to the search. The suggested pages may be redirects to the most relevant destination articles being searched for.
• Logged-out Wikipedia users on mobile web will begin to see a bookmark button for saving articles to their reading lists instead of the current Watchstar button. When clicked, the bookmark button prompts users to log in or create an account to save an article for later. An earlier experiment found that the bookmark icon was four times more likely than the Watchstar button to encourage logged-out users to create an account. This change is part of efforts to encourage more readers to become account holders and use features that help them save and return to content.
• An experiment will begin on September 29 on Spanish, Arabic, English, and French Wikipedias to simplify the experience users encounter immediately after account creation. The test will explore whether removing friction from the current Welcome Survey helps new users understand what to do next more easily. The experiment will run through the end of October.

• The Product and Technology Advisory Council has put out a call for the community to suggest topics that it should consider and discuss with the Wikimedia Foundation. Editors and technical contributors are invited to add topics to the council's talk page.
• Later this week, it will be possible for communities to configure Edit Checks and Suggestions so that they are shown within additional namespaces, such as a Draft: namespace. Feedback is welcome.
• A bug in the Global Watchlist is causing the page to fail to load for users that have unseen changes on Wikimedia Commons. Developers are working on a fix. Until then, affected users can follow the workarounds described in the relevant help section.
• New magic words {{CATEGORYSORT:TIMESTAMP}} and {{CATEGORYSORT:RTIMESTAMP}} are now available for use. They are able to change the default sorting of categories to be based on timestamp of categorization.
• View all 20 community-submitted tasks that were resolved last week. For example, the issue where Wikistats for Wikipedias was not reachable has now been fixed.

Updates for technical contributors

• The datacenter switchover, scheduled on September 23, has been postponed. The equinox exercise revealed capacity issues, preventing the switch of all services to the other datacenter. The process has been paused, prioritizing investigating that issue, to ensure that we continue to be able to serve our users reliably. A new exercise will be scheduled. Meanwhile, all traffic and edits continue to work as usual.
• On September 23, a power loss briefly affected all wikis, creating intermittent issues on both reading and editing modes. This also affected Gerrit. This is unrelated to the datacenter switchover postponement.
• Produnto has been deployed to mediawiki.org and some Indic language wikis. We want to hear your feedback on this experimental product. Produnto allows users to use Lua modules hosted in Wikimedia's GitLab. It simplifies the sharing of Lua code between wikis. Instead of copying each individual Lua module from one wiki to another, users will be able to easily share packages across all wikis.
• Detailed code updates later this week: MediaWiki

Meetings and events

• A Cross-wiki code collaboration workshop will be held online on October 2 at 12:30 UTC, bringing together South Asian technical contributors to learn about Produnto and begin piloting it on Hindi, Punjabi, Odia, Telugu, and Malayalam Wikipedias. Produnto is a package manager for deploying Lua modules hosted on GitLab to Wikimedia wikis, and was recently deployed on some pilot wikis.

Tech news prepared by Tech News writers and posted by bot • Contribute • Translate • Get help • Give feedback • Subscribe or unsubscribe.

MediaWiki message delivery 10:46, 28 September 2026 (UTC)

The {{CATEGORYSORT:TIMESTAMP}} keyword does not work in preview... Christian75 (talk) 14:56, 28 September 2026 (UTC)

Christian75, adding ?cldsort=timestamp to the url works. — Qwerfjkltalk 17:03, 28 September 2026 (UTC)

Hello all! and Christian75! and Qwerfjkl!

I have created Template:Display order, to dynamically choose the display order of elements in the category. You can see its three buttons in the description of Category:Open Wikipedia bot requests for approval. But:

• This is for maintenance categories (encyclopedic categories should not need this).
• There are limitations (see Phabricator:T433768 or the summary in the template's doc).

Regards --NicoScribe (talk) 14:00, 29 September 2026 (UTC)

@NicoScribe I hope you don't mind, but I modified that template to allow setting a default value which takes advantage of the new magic words. --Ahecht (TALK
PAGE) 19:15, 29 September 2026 (UTC)

I have documented the feature in a new section Help:Category#Sort order. PrimeHunter (talk) 01:48, 30 September 2026 (UTC)

Music members vs. Hired members

i think actual web pages like "band" pages should have a strict "members" mentioned section with the actual band and song writers mentioned and any and all hired musicians a different section, hired musicians are just hired musicians and are not a contributor towards the band or it's self image or success what so ever.

Should be as follows:

Members

Past members

Hired members

Contributors/collaborations ~2026-52135-56 (talk) 14:19, 28 September 2026 (UTC)

This already is the case, as far as I'm aware; most bands' articles that I recall already do separate full-fledged members of the band from session musicians and touring musicians. Are there articles where you feel this approach isn't currently being followed? ModernDayTrilobite (talk • contribs) 18:17, 28 September 2026 (UTC)

This is the village pump for technical issues. Your request would be better at WP:VPIL or at WP:VPR. Izno (talk) 18:51, 28 September 2026 (UTC)

Wikipedia talk:Manual of Style/Music sounds better for something so topic specific. PrimeHunter (talk) 19:44, 28 September 2026 (UTC)

It's actually been discussed at Wikipedia talk:WikiProject Musicians several times, see the archives. --Redrose64 🌹 (talk) 21:26, 28 September 2026 (UTC)

Internet Archive pdfs

Hi, regarding the problem that many internet archive pdfs only show the first page on ios ipads and ios mobiles while displaying perfectly on Android. I asked Google AI about it and they said adding if_ to the archive url in the position shown below would solve the problem so that the pdf displays in full on ios and android and it seems to work. Here is an example: https://web.archive.org/web/20121123165203if_/http://go-ahead.com/~/media/Files/G/Go-Ahead/ir/presentations/archive_pres/1996pres/ar1996.pdf shows the pdf in full whereas https://web.archive.org/web/20121123165203/http://go-ahead.com/~/media/Files/G/Go-Ahead/ir/presentations/archive_pres/1996pres/ar1996.pdf only shows the first page on ios. Apologies if this is common knowledge, regards Atlantic306 (talk) 15:26, 28 September 2026 (UTC)

This is probably something that should be reported to Archive.org or Apple. Editors are just going to copy the URL they have when creating articles. -- LCU ActivelyDisinterested «@» °∆t° 16:57, 28 September 2026 (UTC)

connected content in VE?

If I edit Special:Permalink/1377329084 in VE and click on the infobox, I get a You are currently editing a template and one or more pieces of connected content (wikitext and/or additional templates) warning. Deleting the {{use dmy dates}} template (i.e. Special:Diff/1377329165) makes the warning go away. What's going on here? I've read the VE explanation about this but it doesn't seem to apply here. Have I just found a VE and/or parsoid bug?

I first noticed this in Chimpanzee and managed to get it down to a minimal case in my sandbox using the time-honored technique of just deleting stuff until the problem goes away :-) -- RoySmith (talk) 23:33, 28 September 2026 (UTC)

Looks like the output HTML begins like

<p about="#mwt2" typeof="mw:Transclusion" class="mw-empty-elt" id="mwAg" data-mw='{"parts":[{"template":{"target":{"wt":"Use dmy dates","href":"./Template:Use_dmy_dates"},"params":{"date":{"wt":"March 2025"}},"i":0}},"\n",{"template":{"target":{"wt":"Speciesbox\n","href":"./Template:Speciesbox"},"params":{"status":{"wt":"EN"},"status_system":{"wt":"IUCN3.1"},"image":{"wt":"015 Chimpanzee at Kibale forest National Park Photo by Giles Laurent.jpg"}},"i":1}}]}'>
</p><table class="infobox biota" style="text-align: left; width: 200px; font-size: 100%" about="#mwt2">

The metadata there indicates that, for whatever reason, Parsoid decides that the <p> covers both the {{use dmy dates}} and the start of the {{Speciesbox}}. Probably that's why VE talks about connected content. Anomie⚔ 23:56, 28 September 2026 (UTC)

Thanks. Sounds like a parsoid bug, then.

Tracked in Phabricator
Task T439492

-- RoySmith (talk) 00:22, 29 September 2026 (UTC)

Bug with external links

I was viewing the project page for the village stocks, which has many external links to Wikipedia action logs and rdiffs. However, for some reason, the literal for these links is seemingly broken. Here is an example:

Expected: On 1 August 2005, Ed Poor, one of Wikipedia's most experienced editors, boldly decided to delete the entire VfD deletion process...

Actual result: On 1 August 2005, Ed Poor, one of Wikipedia's most experienced editors, boldly decided to wikipedia.org/w/index.php?title=Special%3ALog&type=delete&user=&page=Wikipedia%3AVotes+for+deletion delete the entire VfD deletion process...

Please fix this bug as soon as possible. For those interested, I am currently using Google Chrome version 144.0.7559.262, on ChromeOS. AndyShow1000000 (talk) 12:27, 29 September 2026 (UTC)

FWIW, it looks as expected to me, using Vivaldi 8.2.4133.68, which is based on Chrome 152.0.7977.137 (and Linux Mint 22.1 as OS). I have no idea whether/how the browser version might affect that display, but if you can update, maybe that's worth a try. Suðurhafsljósæta (talk) 13:25, 29 September 2026 (UTC)

They look fine to me (latest Firefox on Mac OS, Vector 2022) – Jonesey95 (talk) 13:29, 29 September 2026 (UTC)

@AndyShow1000000: It's caused by importing User:EpochFail/wikignome.js in User:AndyShow1000000/common.js. Maybe the script cannot handle that https: is omitted in [//en.wikipedia.org/w/index.php?title=Special%3ALog&type=delete&user=&page=Wikipedia%3AVotes+for+deletion delete] EpochFail only has one edit since 2023. PrimeHunter (talk) 15:48, 29 September 2026 (UTC)

I have removed the user script, and things are working just fine now. Thank you for all of your support! AndyShow1000000 (talk) 16:28, 1 October 2026 (UTC)

What is going on with my infobox?

I was working on an infobox at my user sandbox User:Shocksingularity/sandbox and I've been having trouble getting it to work with parser functions. I've isolated the problem to between a few comment blocks I've already added. Could anyone help me?

(Just ignore the mass of curly brackets and the parameters under them, that will be fixed once I uncomment stuff (because some of the opening brackets are now inside comments). You can also ignore the giant wall of text that is data9 and anything before that, I left a large line break after it so it'll be easier to tell when to stop scrolling past.)

Thank you to anyone who can help! Shocksingularity (talk) 21:32, 29 September 2026 (UTC)

You had three { characters where just two { were needed. Three is for parameters, and two is for templates and parser functions. I recommend enabling the Syntax Highlighter gadget in your Preferences; it made these problems pretty easy to locate. – Jonesey95 (talk) 21:37, 29 September 2026 (UTC)

I fixed the { but unfortunately it is still not working Shocksingularity (talk) 23:22, 29 September 2026 (UTC)

It looks like it is working at least at a basic level. I have posted on your User talk page. – Jonesey95 (talk) 23:41, 29 September 2026 (UTC)

Listening to Wikipedia Android app test

Further information: mw:Readers/Listening to Wikipedia/Phase 1

An Android screen recording of the beta version of spoken articles

Hi all! The Readers team is exploring creating a feature to allow readers to listen to Wikipedia articles using text-to-speech software, responding to m:Community Wishlist/W304 and user feedback we've received on the apps. This work is early-stage, so to begin we have developed a beta version that presents spoken article leads in the existing "For you" feed of the Android app. We would like to conduct a monthlong A/B test of this beta version to determine whether or not it boosts reader retention, which will inform our thinking about whether to invest in developing the feature further. We estimate that around 1.5% of Android app users viewing their feed in English will see the test.

You can read more about the planned feature at its project page, and about this experiment at the Phase 1 page. As always, we're happy to answer questions and we'd value your input. Does this seem like a useful feature? Does it raise any particular concerns we should have in mind? What would you like to see if we develop it further? Let us know either here (for the A/B test) or on the MediaWiki talk page (for broader thoughts). Cheers, Sdkb‑WMF ^talk 23:47, 29 September 2026 (UTC)

Replying here because I was notified about this effort by Sdkb as a member of WikiProject Spoken Wikipedia. I appreciate notifying the project in general, although I somehow doubt that a community of people who enjoy creating audio versions of articles with real, human voices, will respond positively to using generative AI to replace their efforts. I am aware SPOKEN is a small community with limited output ability. Recording audio takes time. Recording it well takes even longer, and not everyone who participates has access to the best setups. That limits who can participate. GenAI can produce audio versions exponentially faster than we meatbags can and if you just want an audio version (not a good audio version, mind you) of article intros for an experiment, you can hardly beat GenAI. Call me a fogey if you like, I still don't care for it and still think it goes against the "human-first" ideal the Foundation has been espousing for the last year and change.

Rather than this make yet another referendum about Foundation staff trying to jam generative AI into Wikipedia (that went super well last year), let's address the core questions being asked here.

Does it seem useful? Eh, not in my use cases. As a reader, I use Wikipedia to quickly check on things I am curious about. I don't scroll around it like it's a social media feed. Others might. They might find it useful.

Does it raise any particular concerns? Yes.

• If the goal as alluded to on the project page is to have a "natural, human-sounding voice" reading these intros, the experiment has already failed. The example provided does not sound remotely human. It sounds like generative AI. It sounds less robotic than a screen reader, but it still sounds like a machine.
• If the goal of Wikipedia is to impart good, credible information to readers, that should extend to pronunciation. AI does not know how to pronounce a great many things and if it mispronounces something, it can't easily be corrected. Loyal-SOAK (from what I can tell, should be Loyal-SOCK, though a PA native may correct me, ~:20 in the video), 56.7 M (a human would know to read that as 56.7 METERS, that would take a listener right out of the recording), ROTCH (should be Rohsh, 1:05), headQUARters (just weird emphasis, 1:14ish), four quick examples in the 2-minute clip here. This kind of pronunciation research is exactly what SPOKEN members are supposed to be doing during their recording processes and exactly what cannot be done with a TTS GenAI clip.
• In general, I don't think we should be looking to incorporate GenAI in this manner into any part of Wikimedia. If we would not allow GenAI to write an article, why would we allow it to record one?

What would you like to see if we develop it further? I'd rather you didn't develop it further, at least not in this form. If this is something the Foundation wants to do, it should invest in SPOKEN and human editors. Encouraging humans to record these intros would be vastly superior and I'd be willing to bet real people reading these would give the Foundation a more accurate test. That investment might come in the form of a contest, a grant for SPOKEN to incentivize recordings, or something similar. The Foundation can't continue to claim to be human-focused and then try to cut the humans who want to do this kind of work out of the process. Pick one or the other. No having your cake and eating it too. M4V3R1CK32 (talk) 03:13, 30 September 2026 (UTC)

Thank you for taking the time to write out this feedback @M4V3R1CK32. My name is Haley, and I'm a product manger on the Mobile Apps team, working on this experiment. You are correct that one of our goals is to have a natural sounding voice and nothing is more natural than a human reading out loud, but the other goal is developing infrastructure to keep recordings current so that readers can follow-along with the text of the article, and provide audio for articles at scale so that readers can regularly listen to them. There are currently only 1,937 spoken articles on en-wiki, which is 0.03% of all articles. It would take an incredible amount of volunteer time and resources to record and rerecord articles with every edit. Part of the foundation's AI strategy is to “...use AI to build features that remove technical barriers to allow the humans at the core of Wikipedia to spend their valuable time on what they want to accomplish, and not on how to technically achieve it.” We are starting with the test as-is so we can see whether readers are interested in listening to articles to begin with. If we're able to prove this and are ready to spend more time building, we can consider making more options available for the articles themselves. In the meantime, we can start brainstorming for various options here. I'm curious to know what you think of the existing process to create recordings using a text-to-speech model. Which parts of human oversight in the process are most important vs. where are the opportunities to enable volunteers to create article recordings more efficiently?

Thanks for highlighting those pronunciation issues. This is an early experiment, and the model we're using now isn't necessarily the one we'd use at scale. Before any wider rollout, we'd work on various improvements like voice quality, and build workflows to handle issues like "M" being read as something other than "meters". At the moment, the model is highly configurable, enabling improvements like text normalization (teaching the model how to handle certain types of content) and whitelisting (setting a specific pronunciation for a given word or phrase) during the development process — but because we're testing a static set of audio files, we're not able to make improvements during the course of this initial experiment. It's possible that down the line, we'll have a scenario where editors are able to inform the pronunciation of certain words by contributing to whitelisting or through another mechanism. I'm curious to know what your thoughts are on that idea? HNordeen (WMF) (talk) 22:11, 1 October 2026 (UTC)

the other goal is developing infrastructure to keep recordings current so that readers can follow-along with the text of the article, and provide audio for articles at scale so that readers can regularly listen to them.

Totally understood. There are 7.2 million English articles alone. Creating recordings of just the start-class articles and higher would be a massive undertaking. And forget about keeping them up to date. A team of 1,000 narrators working full-time on nothing but updating recordings could never keep up with all the changes, but AI can. I think keeping things truly current would probably be a big burden on Commons as new recordings are created. I don't know what discussions have been had about how often recordings should be updated based on changes made or if old recordings should be deleted but I imagine that is something for down the road.

And I totally get using AI for the purposes of the experiment here. I also get that members of SPOKEN want to use their valuable time to accomplish using their voices to record articles. So if Part of the foundation's AI strategy is to “...use AI to build features that remove technical barriers to allow the humans at the core of Wikipedia to spend their valuable time on what they want to accomplish, and not on how to technically achieve it., using AI to remove the technical barrier of recording articles from members of SPOKEN is pulling a fish from water and expecting it to breathe. The technical part is the point with recording articles. So again, the WMF cannot have its cake and eat it too.

And look, SPOKEN is a real small group. I don't claim to represent any of it, and I don't know that it's a big enough group that my concerns related to SPOKEN should do anything to derail this experiment. And SPOKEN is fairly intermittent in its activity, I get that. I know it has been quite some time since I have recorded any articles myself. A big part of that is concern that projects like this will use my voice to power a generative AI model, which is a big problem for someone who also uses their voice to create content and make money outside of Wikipedia. I don't want someone using an AI version of my voice to create recordings of me saying abhorrent things or promote products I would never promote, and creating spoken articles gives everyone a free voice print to use as they see fit (more or less) because it's licensed under Creative Commons. I imagine that is a concern of at least some other people and probably a part of the reason SPOKEN is not as active as it could be in the AI age. There are no protections for us in the Wikimedia ecosystem and it is something we have to balance as content creators. That's beyond the scope of this discussion, apologies for digressing.

It would take an incredible amount of volunteer time and resources to record and rerecord articles with every edit.

Absolutely agreed. It's a big problem and a big challenge. A quality recording of an article can be made with as little as a cell phone and a decently full closet. These don't need to be studio quality, but even with those fairly minimal requirements that still doesn't eliminate the time it takes to actually record the words and do some basic editing. Heck, even just recording 8,000 article introductions for this experiment would be a huge undertaking. I do think volunteers would be vastly more incentivized to record if they knew their voice would be more easily heard. I used Wikipedia for almost two decades before I knew SPOKEN was a thing. Recordings are buried and hard to access. SPOKEN is terrible at advertising itself. Some of that is technical stuff the WMF can help with, some of it is SPOKEN's own problems. All of it is beyond the scope of this discussion, so, again, apologies for the digression.

I'm curious to know what you think of the existing process to create recordings using a text-to-speech model.

I can't speak to most of the technical stuff and I'm not going to download any GenAI software to my system and try to follow these instructions, but here are a few thoughts:

• In the intro, "blind people" --> "visually impaired"
• Standardize the file type. OGG is preferred, that's the standard used by SPOKEN, let's just use that, remove potential issues with MP3s
• SPOKEN reads the captions of media files. That's important for accessibility. Usually I say something like, "This section contains 1 image, with the caption TKTKTK"
• There is nothing in there that says "check the work of the AI and make sure it actually did a good reading". It's implied, but that pronunciation bit is a super important part of recording any text and it should be prominent in the instructions. It also mentions issues with audio being wonky, that all should be checked and that should be an explicit requirement

Which feeds in well to Which parts of human oversight in the process are most important. I think the most important parts are quality control for the audio itself (e.g. consistency of volume within the recording, is it at the right DB for comfortable listening, peaking/distortion, ensuring the full text is actually recorded, etc.) and quality control for pronunciation. Recording a spoken article can require a shocking amount of additional research and people using AI to make recordings of articles should be putting in that same effort. I don't know how you check that stuff without actually listening back to the article, which is the part that would take the longest and could not be automated, because AI doesn't know how to pronounce things and could not catch if there was a mispronunciation. Moreover, would someone listening back even realize that a word was pronounced incorrectly? I check pronunciations because I stumble in my recordings, but if a reviewer is listening to an AI confidently say the wrong thing, would they even realize?

By where are the opportunities to enable volunteers to create article recordings more efficiently? I presume you mean using AI to create recordings. Can't help you there. If you mean humans creating recordings with their own voices, SPOKEN has a guide that could probably use some updates. In particular, there are things that can be done with macros in Audacity that can make the editing process much faster. But that's outside the scope of this discussion.

Before any wider rollout, we'd...build workflows to handle issues like "M" being read as something other than "meters".

There are contexts when it would be, though. Like "million". Maybe programming that context is not difficult, I don't know. But there are an absolute ton of weird contextual things like that on Wikipedia. Maybe that isn't a big issue though, I don't know.

It's possible that down the line, we'll have a scenario where editors are able to inform the pronunciation of certain words by contributing to whitelisting or through another mechanism. I'm curious to know what your thoughts are on that idea?

Sure, sounds great. But how many people would get misinformed before the error is brought to attention and corrected, and how big of an issue is that? These recordings will get fed into AI training data, and the misinformation will rapidly propagate. How big of a problem is that? Genuine questions, and ones that should probably be decided by a consensus rather than one cranky dude who wants AI to go back in its hole and never come out. M4V3R1CK32 (talk) 05:10, 2 October 2026 (UTC)

@HNordeen (WMF) curious about your response to this from Bernadette Meehan: "I don't think you will ever see AI writing or generating Wikipedia articles," [Meehan] said.. That is from today, Oct. 2, 2026. Is this experiment not AI generating articles? Like Kowal2701 pointed out with alt-text, I don't see a distinction between generating a recording and generating the text of an article. So which cake do we get to have and which do we get to eat? M4V3R1CK32 (talk) 15:34, 2 October 2026 (UTC)

Hi @M4V3R1CK32, thanks for the note. We're thinking about this experiment as testing out an AI-powered tool to provide a different way of consuming an article's content, without altering that content at all. All of these things are so new that we will definitely have philosophical questions to tease out together. And definitely agreed that these things should be figured out by consensus here onwiki. It's already been hugely helpful to see people's thinking around Spoken Wikipedia (this came up a lot at WikiCon North America last weekend, and we had some really productive conversations with contributors there who have worked closely with it). HNordeen (WMF) (talk) 21:06, 2 October 2026 (UTC)

i implore the wmf to not subject our readers to the creepiest "voices" ever conceived of. ltbdl (talk) 02:38, 2 October 2026 (UTC)

Well, I'm coming at this from a perspective of one of the many people who does listen to podcasts/audio works -- though, I concede, I'm probably on the very high end of that number. (Nobody ask to see my lifetime Audible stats or Spotify Wrapped; it's very scary). So maybe I should be treated as an outlier.

Podcasts and audio-only formats do not quite map to spoken Wikipedia articles; if I'm going to listen to a podcast, audiobook, audiodrama, recorded lecture series, then the voice has to be pleasant. AI voices, even the human sounding one in the example.... not that pleasant? They also have a narrative structure; compare a Teaching Company course to an encyclopedia article, and you'll see they have very different structures and goals. We're a reference work; they aren't. I will listen to Wikipedia articles on history subjects, when there's a Spoken article available, but if I want to learn history I just... put on a history podcast or video. If I want to find information quickly, I consult a reference book.

Now let's look at more practical, less subjective metrics. This feature, it's in strict competition with browser's TTS function. And, at first glance, the inbuilt TTS service in Chrome looks much better. I can skip around either by sentence, or by highlighting a paragraph later in the page. I have a much more visible little highlight above the currently being read word, so I can quickly find my place in a page if I wish to consult the written text (perhaps for something more data-heavy, or to see where an image is). (And, for texts not in my native English, so I can follow the text more easily. So why would I use this feature when my browser comes with something better?. There's also a matter of speed; when listening for work, not pleasure, I like being able to put something on 2x speed.

I understand that developing these features may not be a priority unless there's an appetite for this feature; however, without functionality that appears on other platforms (even Overdrive or, in terms of the basic "fast-forward" functionality, a Playaway). I can't see there being as much appetite for this feature. GreenLipstickLesbian💌🧸 06:16, 2 October 2026 (UTC)

FWIW, there was rough consensus against allowing AI-generated alt text for images at AINB, I struggle to see how this is much different in principle Kowal2701 (talk, contribs) 12:31, 2 October 2026 (UTC)

I do like the idea of showing a human read-aloud section prominently and can see a niche for us given the interest in podcasts and similar passive media consumption especially on app-based services. However, I don't think "brainrot Tiktok voices reading out Wikipedia ledes" is a good way to capture that market segment and I definitely do not think the WMF should be moving further in this direction. Sohom (talk) 15:58, 2 October 2026 (UTC)

idk whether WikiProject Spoken Wikipedia would have been more successful had it been launched as it's own wiki rather than just an obscure WikiProject (ik none of the history here) Kowal2701 (talk, contribs) 16:25, 2 October 2026 (UTC)

Hi all, we're seeing a lot of good comments and questions here. The Apps team is mostly signed off for the weekend, so we will loop back here next week to consider these points together. Thanks! EBlackorby-WMF (talk) 22:02, 2 October 2026 (UTC)

Notified: Wikipedia talk:WikiProject Spoken Wikipedia, meta:Talk:Wikispeech and meta:Talk:Community Wishlist/W304. Sdkb‑WMF ^talk 23:53, 29 September 2026 (UTC)

Why are such things from way, way down on the wishlist () with only 3 supports after nearly two years, prioritized by the WMF? It seems as if there are plenty of more popular wishes to choose from, like "Show categories on mobile" which has 47 supports but is not being worked on apparently. Fram (talk) 16:19, 2 October 2026 (UTC)

And why is the WMF reinventing the wheel? As GreenLipstickLesbian notes, text-to-speech is already available from browsers - created by people with far more experience and resources. People who use such things regularly (because of a visual impairment, or for any other reason) are likely to want to stick to what they know. This looks like pointless duplication. AndyTheGrump (talk) 16:26, 2 October 2026 (UTC)

@Fram, Why are such things from way, way down on the wishlist ([14]) with only 3 supports after nearly two years, prioritized by the WMF? I can speak to some of that. This task was prioritized during the previous iteration of the wishlist where WMF was treating the wishlist more as a idea dump to supplement their existing product direction dictated by the Annual Plan rather than a a "wishlist". I think both multiple folks have raised concerns about this kind of prioritization during the discussion surrounding the disbanding of the Community Tech team (including this particular wish) which has led to the Foundation hiring a new manager for the Community Wishlist process from the community and committing to revamping the process to be similar to the "old" wishlist where wishes were prioritized based on votes. Sohom (talk) 16:36, 2 October 2026 (UTC)

Then perhaps they shouldn´t use "responding to the wishlist" as an excuse for this. Fram (talk) 17:33, 2 October 2026 (UTC)

What is going on with Special:WantedPages?

Special:WantedPages is full of pages that supposedly have tens of thousands of incoming links. But none of these links seem to actually exist, or have ever existed. This has been the case for as long as I've been aware of the list, and it makes the list essentially useless. What is happening? –CopperyMarrow15 (talk · edits) 09:36, 30 September 2026 (UTC)

Please give examples. Johnuniq (talk) 09:53, 30 September 2026 (UTC)

Do you not see what I'm seeing? Here's a screenshot. –CopperyMarrow15 (talk · edits) 10:11, 30 September 2026 (UTC)

The names with about 32,000 links are listed at Wikipedia:WikiProject Israel/to do2, which is transcluded onto thousands of talk pages as part of the Wikipedia:WikiProject Israel talk page banner. -- John of Reading (talk) 11:03, 30 September 2026 (UTC)

I think that I have cleared the WikiProject Spam entries. It may take a few days for them to disappear from the report. It looks like most of the entries on the first page are either on WikiProject To-Do lists (many of which may be stale and unused, and all of which are hidden when they are transcluded on talk pages, so it is unlikely that editors are seeing them) or on navbox templates (e.g. {{Serotonin receptor modulators}}, which has 1,500 transclusions and multiple red links). It looks like Wikipedia:Most-linked-to redlinks tries to make some sense out of this report. – Jonesey95 (talk) 13:17, 30 September 2026 (UTC)

I see. Would it be possible to add a feature that lets you filter by the namespace of the incoming links? I feel like that would make it a lot more useful. I'm bringing this up because there is a button at the top of my Special:Contributions that says "New page" that is supposed to encourage you to make a new page, and it just links to Special:WantedPages, which doesn't seem helpful. –CopperyMarrow15 (talk · edits) 05:19, 1 October 2026 (UTC)

I once made a proposal at Wikipedia talk:WikiProject Council/Archive 21#Proposal: Disallow transcluded to-do lists but didn't follow up. There was support but some thought every relevant WikiProject should be notified. PrimeHunter (talk) 13:57, 30 September 2026 (UTC)

Probably the best thing to do is to make a database report (or find an existing one) using a SQL query. Someone skilled in SQL could probably make a query that somehow excludes links from outside of article space. – Jonesey95 (talk) 15:03, 30 September 2026 (UTC)

As an example, So Long Davey! (58,777 links) was recently deleted and has hidden links on thousands of talk pages due to {{WikiProject Video games}} listing current and recent AfDs on the topic. It's the very opposite of a Wanted Page. Certes (talk) 17:31, 1 October 2026 (UTC)

That particular issue looks like a symptom of a different problem. That page was listed on the page Wikipedia:WikiProject Video games/Article alerts/AfD, but the link to it was removed on 23 September, more than a week ago, after the article was deleted. A week later, What links here for that article still (incorrectly) shows thousands of links. Null-editing any of the Talk pages in that What links here list will remove it from the list. The root problem is that page caches are not cleared frequently enough, leading to a current stale article backlog of 40-plus days just in article space. That means that Special:WantedPages can take 40-plus days, and sometimes a lot longer, to reflect changes that are made today. More details are at the ten-year-old bug reports / feature requests at T135964 and T157670. Anyone who can contribute to helping pages refresh more frequently will fix a lot of categorization, reporting, and other issues related to out-of-date page links. – Jonesey95 (talk) 18:00, 1 October 2026 (UTC)

@Jonesey95 There's always User:Ahecht/Scripts/refresh.js as a "hackish and silly" way to forcibly purge (a null edit isn't needed, just a forcerecursivelinksupdate purge) all the pages in What Links Here, which is better than waiting for Bot1058 to get around to it. As I recall, the bot attempts to do articles every 40 days or so, but talk pages only get touched every few months. --Ahecht (TALK
PAGE) 19:14, 1 October 2026 (UTC)

Thanks. I know how to do it, and I have that script (or some variant of it) loaded, but refreshing 43,000 pages while trying to do other stuff sometimes results in my browser getting excessive activity warnings, or failing to save edits, especially since they tightened the screws a couple of months ago. The Wikimedia job queue should do a better job of actually queueing the pages that transclude pages that change, IMO. We invented machines to relieve us of drudge work like this. – Jonesey95 (talk) 19:21, 1 October 2026 (UTC)

Several notification popups with zero actual notifications

Twice now today I have had the number beside the bell icon turn red, and when I click on it no new notification shows up and I cannot find where it could've come from, even on other wikis. Not sure if this is a known issue but it happens periodically. Eoline 17:36, 1 October 2026 (UTC)

Chances are that the pages from which the notifications come were deleted. NguoiDungKhongDinhDanh 17:48, 1 October 2026 (UTC)

@Eoline Thanks for reporting. A few followup questions before I file a task:

Is the number beside the bell icon showing as "0", or as a "1" or more?

When you click on it, does the badge go back to black/white or stay red, and does the number go away? If not, can/did you make the badge reset, by some other method, afterwards?

[EC] +1 to NguoiDungKhongDinhDanh's speculation, and: Do you think that's a likely explanation - I.e. if this has occurred periodically, do you often edit/watchlist/subscribe to pages which are later deleted, e.g. Drafts/Newpages? Quiddity (WMF) (talk) 17:50, 1 October 2026 (UTC)

@Quiddity (WMF): The bell icon (last time it did this) stayed at 28. It grays out when you click it but no notification pops up. I rarely subscribe to deleted pages and I doubt I would have two notifications from deleted pages on the same day, unless a page being deleted causes a notification that I'm unaware of. Eoline 17:54, 1 October 2026 (UTC)

I was referring to the scenario in which someone creates a page with a ping to you, then that page goes on to be deleted. I get a lot of these ghost notifications from vandals. That might be the case for you as well (or not). NguoiDungKhongDinhDanh 19:18, 1 October 2026 (UTC)

Confused

Aren't categories supposed to show up in mobile? One other day I was using my phone to scroll through Wikipedia when i tried to find the categories and I could not find them... FoxOutOfRange 18:21, 1 October 2026 (UTC)

To see categories on mobile website, go to Special:MobileOptions and turn on "Advanced mode" near the bottom of the page. —⁠andrybak (talk) 18:36, 1 October 2026 (UTC)

The editing toolbar overriding both the Ctrl+E and Ctrl+K keyboard shortcuts is annoying.

The fact that the editing toolbar (2010 wikitext editor) now overrides the Ctrl+E (on Windows) keyboard shortcut, which is the default shortcut key for web search on most browsers, for the insertion of <math display="inline"></math> is an annoyance that I'm having trouble getting used to working around. (The other browser web search shortcut, Ctrl+K, brings up the insert link dialog box.) Disabling access keys in user preferences does not disable these keystrokes; only disabling the editing toolbar does.

I know I can get the same thing by using Ctrl+L or Alt+D to go to the address bar and then type ? to trigger web search, but that's two extra keystrokes, and it's an ingrained habit that will take a lot of effort to relearn, which users shouldn't be forced to do. Is there a way to disable these shortcuts without losing access to the Reftoolbar? Is it a local setting or something built into MediaWiki? --Paul_012 (talk) 19:40, 1 October 2026 (UTC)

Does your browser not have an omnibox (search or URL in the same box)? If you have it set that way, Ctrl+L does what you want.

I'm not aware of a way to override these shortcuts. They are not access keys either, which is why the gadget in user preferences doesn't work. I can say that conflicts with keyboard shortcuts are commonplace and often hinder development. The <math>...</math> shortcut for example used to be Ctrl+m, but that conflicted with mw:Extension:UniversalLanguageSelector. I have made a note at phab:T95148 (which might replace the existing toolbar in the coming months) to make keyboard shortcuts configurable. — MusikAnimal ^talk 21:52, 1 October 2026 (UTC)

It does have an omnibox, but there's a difference in behaviour between when one types directly in the address bar and when one specifically calls it using the search shortcut (or precedes the query with a question mark). Auto-complete misfirings and other errors happen often enough for it to be an annoyance. --Paul_012 (talk) 14:07, 2 October 2026 (UTC)

Comment: Ctrl+K has always been insert link for me on a computer. Axolitl (talk | contribs) 22:52, 1 October 2026 (UTC)

SVG Dark Mode Question

I had a question about SVG appearance in dark mode. Suppose you go to the article English-speaking world on a PC displaying Wikipedia in dark mode. The image File:English language distribution.svg appears with what appears to be light blue for "majority native speakers" and dark blue for "official or administrative language...". Then, when you go to the Wikipedia image page for that image, it looks reversed and the oceans are now colored white. Why is that? And why do the oceans appear white in the image page, but dark (or transparent) in articlespace? -- Veggies (talk) 20:37, 1 October 2026 (UTC)

It's because of "class=skin-invert-image" in the file transclusion code in the article. IKhitron (talk) 20:41, 1 October 2026 (UTC)

NBSP

Hello, please remove that weird NBSP in the File history section, for example, see this file: File:2026 PanAm Aquatics U19 Water Polo Championships logo.png. There is a non-breaking space between "T." and "(talk)", causing my nickname to be split across two lines. Thanks, Maiō T. (talk) 22:28, 1 October 2026 (UTC)

What makes you think that there is a non-breaking space near your username in that section? When I view source, I see this HTML for the table cell that contains your username and the talk/contribs links:

<td><span class="ext-checkuser-userinfocard-button-wrapper"><button type="button" aria-label="Open user info card" aria-haspopover="dialog" class="ext-checkuser-userinfocard-button cdx-button cdx-button--action-default cdx-button--weight-quiet cdx-button--icon-only" data-username="Maiō T."><span class="cdx-button__icon ext-checkuser-userinfocard-button__icon ext-checkuser-userinfocard-button__icon--userAvatar"></span></button><a href="/wiki/User:Mai%C5%8D_T." class="mw-userlink" title="User:Maiō T."><bdi>Maiō T.</bdi></a></span><span style="white-space: nowrap;"> <span class="mw-usertoollinks">(<a href="/wiki/User_talk:Mai%C5%8D_T." class="mw-usertoollinks-talk" title="User talk:Maiō T.">talk</a> | <a href="/wiki/Special:Contributions/Mai%C5%8D_T." class="mw-usertoollinks-contribs" title="Special:Contributions/Maiō T.">contribs</a>)</span></span></td>

I don't see an nbsp in there. I see that your username and the (talk|contribs) block are in two separate span tags, the second of which has the nowrap style, but no explicit nbsp. – Jonesey95 (talk) 22:36, 1 October 2026 (UTC)

It's not a nbsp character but the space is inside <span style="white-space: nowrap;"> ...</span> so the effect is the same. Maybe the space should be moved outside that. PrimeHunter (talk) 23:14, 1 October 2026 (UTC)

I think we would need to see a screen shot to understand the OP's actual issue. They say There is a non-breaking space between "T." and "(talk)", causing my nickname to be split across two lines. That would not be caused by an nbsp; an nbsp would cause the T. and (talk to stick together, not wrap. If their username is being split across two lines (Maiō on one line, T. on the next line), then the fix for that would be to add a nowrap style around the username, I think, perhaps in the bdi tag. Adding nowrap to text of unpredictable length can cause other problems, though. User names can be long (e.g. Vanished user ewfisn2348tui2f8n2fio2utjfeoi210r39jf, 52 characters, is in the top 3,000 users by number of edits); if they are nowrapped, tables might flow off of the page when they would otherwise fit. – Jonesey95 (talk) 23:37, 1 October 2026 (UTC)

Logged out it renders as this for me because wrapping is disallowed at the space after the username:

Maiō
T. (talk | contribs)

I think this would be better:

Maiō T.
(talk | contribs)

User names can be long so I don't think they should be nowrapped but I don't see a problem in allowing a wrap right after the username. PrimeHunter (talk) 00:53, 2 October 2026 (UTC)

Looks like a good idea to me. I was not aware that the two adjacent spans would automatically not break. – Jonesey95 (talk) 02:28, 2 October 2026 (UTC)

I see the wrap between "Maiō" and "T. (talk ..." when logged in, provided I zoom in sufficiently. It is indeed because the space before "(talk" is after the <span style="white-space: nowrap;">...</span> instead of before it. --Redrose64 🌹 (talk) 18:44, 2 October 2026 (UTC)

Created T440082 to request removal of style="white-space: nowrap;" It looks better without it (by removing in Chrome dev tools). Dw31415 (talk) 12:56, 3 October 2026 (UTC)

Passcode grid.

Hi,<br>

I prefer the recovery codes to a third party authentication app.

Revenue Canada gives a passcode grid somewhat similar to the recovery codes. The grid doesn't expire after eight authentications. Is a passcode grid feasible for Wikimedia?

Thanks, ... PeterEasthope (talk) 03:40, 2 October 2026 (UTC)

From a technical standpoint, it's definitely something that could be done. You'll need to start a discussion in Phabricator to talk to the developers and security people about whether it's something they want to do and consider secure enough to be offered. Anomie⚔ 12:35, 2 October 2026 (UTC)

When will {{Rfc}} be moved?

A requested move at the talk page of {{Rfc}} found consensus to move to {{Request for comment}} in February 2022. However, implementation quickly stalled, because of needed updates to bots to make sure the move would go smoothly. Is there any future where this template is moved? Pinging @Headbomb, @Hellknowz, @kanashimi, and @Legoktm who operate bots which the template is relevant to. Axolitl (talk | contribs) 03:45, 2 October 2026 (UTC)

For Alerts bot, it doesn't matter. If it's changed before the move, it will break. It if doesn't change after the move, it will break. So I just normally fix it afterwards when it breaks because I cannot predict when exactly on-Wiki change will happen. It's also aware of redirects, so it will mostly still work regardless. So no issue here. I'm aware of this Rfc move, but there isn't anything I can do ahead of time in this case. —  HELLKNOWZ ∣ TALK 08:39, 2 October 2026 (UTC)

Adding User:SodiumBot and User:Sohom Datta for Wikipedia:Feedback request service Dw31415 (talk) 13:04, 2 October 2026 (UTC)

Image loading issues

On 25 September 2026, someone uploaded a version of File:Scene (2026 film).jpg in high res, but the bot reduced it with the image displaying an older poster looking distorted. Clear cache attempts do not render the latest poster. Not just this, I see this in a lot of posters now. What to do? Kailash29792 (talk) 04:34, 3 October 2026 (UTC)

You refer to File:Scene (2026 film).jpg#filehistory where two of the images - those uploaded by DatBot - are clearly a different poster than the other three. I would imagine that DatBot has somehow retained an old copy from the last time that it carried out a reduction, and assumed that since the file names are the same, it must be the same image. Have you left a note at User talk:DatBot (where I see that some similar issues have been raised), or contacted DatGuy (talk · contribs), the bot operator? --Redrose64 🌹 (talk) 09:26, 3 October 2026 (UTC)

I see someone else already has raised the issue. I circumvented it on Scene by uploading a manually resized version of the desired image. Kailash29792 (talk) 09:33, 3 October 2026 (UTC)

Also, administrators can view the deleted revisions no matter how old they are. There are two clearly different posters involved, one with the "SCENE" lettering in yellow, one with lettering in silver. Let's refer to them by colour. The original version, of 22:00, 11 September 2026, plus all three DatBot uploads, are "Yellow". The others - of 14:27, 14 September 2026; 18:04, 25 September 2026; 04:35, 3 October 2026; and 04:36, 3 October 2026, are "Silver". So DatBot made the mistake twice. --Redrose64 🌹 (talk) 09:39, 3 October 2026 (UTC)

What links here issue

What links here appears to have an error where its including pages that do not seem to link to the page, such as showing Mister Venezuela under what links here for Mister International, despite not appearing to have a link. Does anyone know why/how to avoid it? Thanks, PeriodicEditor (talk) 11:05, 3 October 2026 (UTC)

Mister Venezuala has a link to Mister International at the top here. Another? -- zzuuzz ^(talk) 11:25, 3 October 2026 (UTC)

Oh, thanks, I didn't see it, I was looking in the texts body, not at that part. PeriodicEditor (talk) 14:42, 3 October 2026 (UTC)