Wikipedia:Village pump (WMF)

 Policy Technical Proposals Idea lab WMF Miscellaneous 

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

The Wikimedia Foundation (WMF) section of the village pump is a community-managed page. Editors or Wikimedia Foundation staff may post and discuss information, proposals, feedback requests, or other matters of significance to both the community and the Foundation. It is intended to aid communication, understanding, and coordination between the community and the foundation, though Wikimedia Foundation currently does not consider this page to be a communication venue.

• Discussions of proposals which do not require significant foundation attention or involvement belong at Village pump (proposals)
• Discussions of bugs and routine technical issues belong at Village pump (technical).
• Consider developing new ideas at the Village pump (idea lab).
• This page is not a place to appeal decisions about article content, which the WMF does not control (except in very rare cases); see Dispute resolution for that.
• Issues that do not require project-wide attention should often be handled through Wikipedia:Contact us instead of here.
• This board is not the place to report emergencies; go to Wikipedia:Emergency for that.

Threads may be automatically archived after 14 days of inactivity.

Behaviour on this page: This page is for engaging with and discussing the Wikimedia Foundation. Editors commenting here are required to act with appropriate decorum. While grievances, complaints, or criticism of the foundation are frequently posted here, you are expected to present them without being rude or hostile. Comments that are uncivil may be removed without warning. Personal attacks against other users, including employees of the Wikimedia Foundation, will be met with sanctions.

« Archives, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17

Source Verification Suggestion

Hi y'all – one outcome of the recent AI-generated edit suggestions discussions was reinforcing the need for us to be in touch with you all early in the process of exploring new inference-based suggestions. This way, we can discuss the experimental suggestion's risks and decide whether en.wiki might be a good place to evaluate its reliability before considering a wider deployment.

We're now at this point with a new experimental suggestion that we'd value your feedback on.

Inspired by the volunteer-authored AI Source Verifier, and how helpful many experienced editors have found it to be in identifying cases where a citation might not support the claim it's attached to, the Foundation's Research, Machine Learning, and Editing teams are working with Alaexis to try integrating this script as a "suggestion", only visible to experienced volunteers who have opted into seeing experimental suggestions in Suggestion Mode.

Below, you will find more information about how this proof of concept will work and the input we are needing from you all. Before that, a note on why we're prioritizing work on this suggestion right now…

We are prioritizing this exploratory source verification work in response to hearing from volunteers:

1. How tedious it can be to find potentially unverified claims within an article (e.g. read the claim, click on each source, ensure you can access each source, etc.)
2. How source verification work is becoming even more important and prevalent as AI increases both A) the ease with which people can add new content to Wikipedia and B) the risk that said content is not supported by the sources it's accompanied by (i.e. contains hallucinated references).

While experienced editors will remain responsible for evaluating whether a source verifies its associated claim, we are seeking to learn whether a tool like this could make the mechanical parts of this wiki work less toilsome.

And in case you're curious, we did a bit of digging to put some numbers to all of this:

• English Wikipedia has more than 70 million citations.
• In one study, annotators worked through a few hundred claims and the web pages cited for them, and judged 12% not supported by the source at all.
• A separate study found that at least 3.3% of facts on English Wikipedia contradict another fact elsewhere on the project.

Note: Neither of those sets represents the encyclopedia as a whole, so the real number is not currently known to us.

How it works

Further information: User:Alaexis/AI Source Verification/How it works

This experimental suggestion will:

1. Identify all inline URL-based citation(s) in the article
2. Extract the text that precedes said citation(s)
3. Retrieve the content of each cited source (and indicate if it's unsuccessful in doing so)
4. Use an open weight Qwen3.6-27B model, hosted on Wikimedia's LiftWing infrastructure, to compare the claim against the source's contents
5. Flag claims that the model has deemed to be only partially supported, not supported by omission, or not supported by contradiction.
   1. Note: The model will accompany each conclusion with the passage(s) from the source it's based on, except where the conclusion is "not supported by omission"

Note: This suggestion would not make edits to Wikipedia and can be configured, like other Edit Checks and Suggestions, to show/not show based on a variety of conditions.

We need your input

We will soon be ready to generate an initial experimental dataset that volunteers can use to offer feedback about the usefulness and reliability of this suggestion. First, we need to decide which wikis/languages to include in this dataset.

This leads us to wonder:

1. Would any of you all be interested in evaluating a batch of these experimental suggestions for en.wiki articles? Note: The suggestions would be made available to you all in a spreadsheet and as a suggestion within Suggestion Mode, visible only to volunteers who have published ≥100 edits and have opted into experimental suggestions.
2. If so, what types of articles do you think would be helpful for us to include within this dataset? E.g. new articles, articles of a certain quality, etc.

For anyone interested in seeing a demo and talking about this in a voice call, we will be hosting a meeting in the Wikimedia Community Discord on 14 Sep 2026 from 17:00 - 18:00 UTC. We'll of course be responsive here as well.

In the meantime, you can get a sense for how the suggestion works by installing the user script that User:Alaexis and User:LuisVilla have been maintaining.

References

FAQ

Based on some of the questions that volunteers raised in the previous discussion about LLM-backed suggestions.

Expand to read the FAQ

1. What kind of communication with communities has there been on this project so far?

This is the first message we've put on-wiki about this work. Note: we've mentioned this work in previous discussions [1][2][3] and have been discussing implementation details in Phabricator.

2. How did we / do we QA lists of suggestions like these?

Quality assurance of this suggestion will happen in two phases, both of which will start once we know what wiki(s) would like to participate in this pilot:

• Internal review: Before any dataset is shared with volunteers, the development team will review a subset of the suggestions and confirm no basic issues are present. We will attempt to address issues we find, and in cases where we can't, will share them with you all to consider as part of the evaluation you will be doing, should this work move forward at en.wiki.
• Volunteer review: Once the internal review is complete, we will make the dataset available for y'all to review in two forms:
  • A spreadsheet so you can identify and categorize error patterns in bulk, rather than one-at-a-time. Note: A spreadsheet is likely not a long-term solution. Though, we think it's a useful first step.
  • Within Suggestion Mode (as an experimental suggestion) so that experienced editors (you all) can see how the suggestions would look and work in practice.

How does this sound to you? What (if anything) about the above do you think we could make clearer? Might there be step(s) you think we ought to consider adding?

3. What determines whether suggestions are good enough to go beyond testing?

Two inputs will determine whether a suggestion is good enough to go beyond testing:

1. Qualitative feedback volunteers will share from reviewing the dataset (spreadsheet)
2. Quantitative feedback about how experienced editors who opted-in to seeing the experimental suggestion engage with it (i.e. the rate at which they engage with and dismiss the suggestion). This information will be tracked using data logging we've implemented within Suggestion Mode.

Note: Prior to beginning the volunteer evaluation, we will share, and invite your feedback about, a concrete proposal for what thresholds we think need to be met for us to collectively consider the suggestion good enough to move forward.

Might there be existing gadgets/scripts that have gone through similar types of volunteer testing that you think would be useful for us to learn from?

4. Is source verification appropriate to point an LLM at, given its nuance and complexity?

Partly. 10 months of volunteer testing of the AI source verification script is leading us to think that LLMs are well suited for assisting volunteers with the mechanical work involved with identifying claims that may need their attention. Things like:

• Identifying all inline URL-based citation(s) in the article
• Extracting the text that precedes said citation(s)
• Retrieving the content of each cited source (and indicating if it's unsuccessful in doing so)
• Comparing the claim against the source's contents
• Flagging claims that the model has deemed to be only partially supported, not supported by omission, or not supported by contradiction.

Through this mechanical work, we believe an LLM can help editors hone in on the claims that need their scrutiny. We do not think an LLM is appropriate for making actual determinations. We see this as nuanced and experience-dependent wiki work that volunteers need to continue being responsible for.

What (if any) of this thinking doesnot align with how you're thinking about this?

5. Does it make sense for the message to the world to be "Knowledge is Human", but we are also using AI for certain things?

We think so. Reason being, we see a core tenet of the "Knowledge is Human" message being the fact that the information included within Wikipedia is grounded in sources written by people, deemed reliable by people, and verified to support the claims they are attached to by people.

Volunteers' time is a finite resource, and editing is a time-intensive process. Our goal is to use AI specifically to reduce the toil that doesn't require human judgement — like information retrieval and pattern matching.

We do not see this suggestion as doing anything to change and/or minimize peoples' role in making any of these determinations. Rather, we see it as helping volunteers to decide what source verification wiki work to prioritize.

We'd be curious to know how (if at all) this thinking lands with y'all….

6. What are the known shortcomings of this approach?

One of the major shortcomings of this approach is that it can only check inline, URL-based citations. This means there are entire categories of suggestions (e.g. paywalled articles, books, etc.) that are beyond the scope of this proof of concept.

Should this initial, URL-dependent approach prove viable, we will explore the feasibility of expanding coverage to include more source types.

False positives are another potential shortcoming; the model might flag claims that, upon volunteer review, turn out to be supported by its source. We understand this to be similar to existing on-wiki tools, like Earwig's Copyvio Detector and ORES. Accordingly, we're understanding the key concern is that this suggestion actually saves volunteers time and effort! Said another way: we wouldn't want the false positive rate to be so high that volunteers spend more time finding genuinely unverified claims than fixing them.

How does this sound to you? Might there be other shortcomings you can see with this approach?

Community Feedback

Thank you for reading and thinking critically about this work! PPelberg (WMF) (talk) 19:09, 10 September 2026 (UTC)

1. Yes. 2. Is it possible to grab the datasets from those two studies (or any other similar) and cross-reference the pages by article-page categories, talk-page categories, and/or talk-page templates (like WP:CTOP templates), to see if any categories or CTOPs are more error-prone than others? That might help prioritize review of any identified problem areas. I would think WP:BLPs, which are all in Category:Living people, would be a top priority for source verification. Also: articles that are new, haven't been edited for a long time, or are tagged with relevant maintenance templates, like {{dubious}} or {{failed verification}}. Levivich (talk) 19:24, 10 September 2026 (UTC)

Is it possible to grab the datasets from those two studies (or any other similar) and cross-reference the pages by article-page categories, talk-page categories, and/or talk-page templates (like WP:CTOP templates), to see if any categories or CTOPs are more error-prone than others? That might help prioritize review of any identified problem areas.

Oh, I think this is a great question/idea. @Alaexis: do you know if we're able to access the datasets used in the studies we referenced so that we could do the comparison @Levivich is describing?

I would think WP:BLPs, which are all in Category:Living people, would be a top priority for source verification.

This intuitively makes sense to me and to be doubly sure: to what extent (if any) would it be accurate for us to understand you suggesting this because of the following reasons?

1. Like articles within WP:CTOP, the suggestion performing poorly on BLPs is more consequential relative to articles in other categories like Category:Railway lines
2. Biographies of living people tend to use news, and other web-accessible sources. This means this suggestion should, in theory, be able to retrieve a larger share of the citations used within them.

Also: articles that are new, haven't been edited for a long time...

Good call. This makes sense to me.

...tagged with relevant maintenance templates, like {{dubious}} or {{failed verification}}

Mmm. In essence, you're saying articles within these two categories offer an existing corpus of claims volunteers have identified as failing verification. Accordingly, running the model against those could help us estimate the model's proficiency. Might I be missing/misinterpreting anything here?

A resulting question that comes to mind as I think about this: might articles tagged with {{dubious}} or {{failed verification}} be more likely to contain offline sources? PPelberg (WMF) (talk) 20:40, 10 September 2026 (UTC)

Some of the datasets are available. In fact Semnani et al have made this analysis themselves Articles in the “history” category exhibit the highest inconsistency rate (17.7%), followed by Everyday Life (16.9%) and Society & Social Sciences (14.3%) (Figure 5). The most common error type in history articles is numerical discrepancy. By contrast, categories requiring precise technical knowledge and quantifiable information—such as Mathematics (5.6%) and Technology (9.4%)—show markedly lower rates.

I like the idea of generating edit suggestions for articles with {{failed verification}} templates for Bayesian reasons - if one such citation has been found it's likely that there are more (aka "if you see one cockroach, there are more" principle). Alaexis_¿question? 20:50, 10 September 2026 (UTC)

On why BLP, I actually didn't have either of those reasons in mind, but both are good reasons. I was thinking something similar to #1: content that fails verification (not just poor performance of the tool) is more consequential for BLPs than, eg, railway lines. Of all FVs, those are the ones we should find and fix first.

On why the FV tag, yes, and it could also help us estimate human proficiency :-) I think it'd be useful to know whether the model confirms the FVs or finds that the FVs are actually verified -- either way it'd be useful. But I had in mind what Alaexis mentioned: if there's one FV tag, there are probably more untagged FVs in the article. I'd guess the odds are higher than average (but I don't know that for sure).

For the dubious tag, that's like a "might" or "arguably" fails verification, so it'd be useful to have the tool analyze those for a person to review. A statement tagged dubious needs a verification check.

And yeah, I'd guess dubious and FV tagged content is more likely to have offline sources. I don't know the statistics, but I'd guess offline sources are rare and so it's unlikely to be significantly more?

Cool idea btw and well presented. Thanks to the teams for working on this! Levivich (talk) 21:12, 10 September 2026 (UTC)

This sounds great and I'm glad the WMF is working on it. Regarding "what types of articles", I don't particularly see any reason to restrict articles included in the dataset, beyond "we don't want to scan the entire Wikipedia" - is that the reason? Or something else? In solidarity, asilvering (talk) 19:26, 10 September 2026 (UTC)

If "We don't want to scan the entire Wikipedia" is the or a reason, then not scanning articles tagged as unreferenced (Category:All articles lacking sources) or lacking inline citations (Category:All articles lacking in-text citations) are obvious ones to not include as (assuming the tags are correct, which is a different issue) they cannot contain references that can be validated in this manner (~143k articles total). It's also not worth spending the resources attempting to scan references tagged as (permanently) dead, failed verification, or dubious. Thryduulf (talk) 19:55, 10 September 2026 (UTC)

If "We don't want to scan the entire Wikipedia" is the or a reason, then not scanning articles tagged as unreferenced (Category:All articles lacking sources) or lacking inline citations (Category:All articles lacking in-text citations) are obvious ones to not include as (assuming the tags are correct, which is a different issue) they cannot contain references that can be validated in this manner (~143k articles total).

Great spot, @Thryduulf. Excluding articles in these categories for the reasons you named [i] sounds like a great idea to me unless, of course, there is a consequence here I'm not seeing.

It's also not worth spending the resources attempting to scan references tagged as (permanently) dead, failed verification, or dubious...

With regard to {{failed verification}} and {{dubious}}, can you please say a bit more here? Asked another way: what's prompting you to think it would not be worthwhile to scan articles with those two templates present?

Per what Levivich and I were discussing above, I'd been assuming, perhaps inaccurately, that scanning those articles could provide a helpful baseline to compare the model's proficiency against. Might I be missing something here?

Regarding the "tagged as (permanently) dead" bit specifically, I'm assuming the following. Please let me know what (if anything) I might've missed...

1. I assume you are referring to articles that include Template:Permanent dead link

2. If so, I assume the reason for excluding articles that contain ≥1 of these templates would be because these citations are unlikely to have an archived copy the suggestion could retrieve, making a check on them likely to fail

---

i. We can assume articles within All articles lacking source do not contain citations the model can evaluate claims against PPelberg (WMF) (talk) 21:01, 10 September 2026 (UTC)

re failed verification and dubious, I hadn't thought about training the model. Rather I was just thinking that if a human has already tagged a reference as not supporting the associated text there isn't much benefit in an AI suggesting to a different human that it might not support the associated text (this would be a waste of time and resources). I agree that using them to train the model would be useful.

re permanent dead links, I was thinking on a per-citation not per-article basis - read the tag and skip the associated without spending any resources attempting to verify it in the source.

Your comment about archive templates has sparked another thought though, that this tool could highlight potential problems with archives. Firstly, if the url-status parameter is blank or live then it should attempt to verify using the live link or both links, if it is "dead", "deviated", "usurped" or "unift" then it should only attempt to verify against the archive link.

For every citation that has an archive link the following are possible:

• Live link and archive are identical, both verify the text (no problem)
• Live link and archive are identical, neither verify the text (problem, but not with the archive but worth flagging to human)
• Live link and archive differ, but both verify the text (almost certainly not a problem)
• Live link and archive differ, live link fails verification archive link passes verification (url-status parameter should be changed to deviated, usurped or unfit, but which probably requires human judgement)
• Live link and archive differ, live link passes verification but archive link does not (flag this for human attention)
• Live link and archive differ, both fail verification (include this information when flagging this for human attention)
• Live link is dead, archive verifies text (url-status should be set to dead, possibly flag to something like user:InternetArchiveBot or some other automated task to make changes more widely).
• Live link is dead, archive fails verification (change the url-status as above and flag to both bots and humans as other citations to the same source may also be dead and pass verification).

Thryduulf (talk) 22:29, 10 September 2026 (UTC)

re failed verification and dubious...I agree that using them to train the model would be useful.

Wonderful. Thank you for walking out what you had been thinking!

...re permanent dead links, I was thinking on a per-citation not per-article basis - read the tag and skip the associated without spending any resources attempting to verify it in the source.

Ah, I see. In concept, what you're describing seems valuable to me. @Alaexis: do you think instructing the model to "skip over" claims that have the Template:Permanent dead link associated with them would require an update to the prompt?

@Thryduulf: in case you're curious, I'm asking about the prompt above because, for now, we're reluctant to make any changes to it. Rationale: a) the prompt has been performing pretty well, b) adjusting the prompt could affect the output in unexpected ways. For these reasons, we're hoping to keep the prompt as-is for this initial round of evaluation.

Your comment about archive templates has sparked another thought though, that this tool could highlight potential problems with archives. Firstly, if the url-status parameter is blank or live then it should attempt to verify using the live link or both links, if it is "dead", "deviated", "usurped" or "unift" then it should only attempt to verify against the archive link.

Oh, this is an interesting set of cases. @Alaexis two resulting questions for you and/or @Isaac (WMF):

1. Do we know how (if at all) the suggestion will behave in these various archive template cases?

2. More broadly, to what extent (if any) would it be accurate for me to think that both a) the LLM could accommodate nuanced instructions of this sort were we to deem them important and b) implementing these instructions would come in the form of an adjustment to the prompt? PPelberg (WMF) (talk) 00:22, 11 September 2026 (UTC)

@Thryduulf If you're curious about how the code selects links to check, you can see the logic and additional notes in the userscript via the extractHttpUrl function. It was written to prefer internet archive links, and then the original link, and only then some of the other archive sites. Checking multiple URLs is possible without changing model prompts but it would still complicate the pipeline as it would functionally double the number of URLs to scrape and model requests to make so would slow things down a good bit. I'll leave that decision to Alaexis but the way I've been thinking about it: the core goal here is verifying the claim, which thusfar we've attempted to do that by fetching the "best" URL. As you raise, there are a variety of additional checks you could do at the same time with the goal of improving the citation (and therefore general Verifiability). I've also thought about inferring the language of the source URL to help fill in the lang parameters on citation templates. These additional checks/actions would complicate the core claim verification task though so I would almost prefer them to be separate at least from an interface perspective, but I am taking note of them.

I'll attempt to answer re: the permanent dead link suggestion as well. It's likely possible to add a check for them (this would also be with the heuristics for extracting links, not the LLM portion of the flow). My thinking: there aren't a ton of them and if they're a dead link then the flow will quickly+gracefully fail anyways (they would never show a suggestion to the end-user). Adding this sort of language-project-specific logic to the code can also complicate efforts to extend this tooling to other language editions in the future. But I'll leave it up to Alex whether it's worthwhile. Isaac (WMF) (talk) 17:22, 11 September 2026 (UTC)

Unfortunately I don't know js so the link doesn't really aid my understanding (but that's not your problem). I don't have a strong feel for what is possible, my suggestions are all things that would are desirable if they are possible. If there are things it isn't going to do (for any reason) but might discover in the process of what it does do then documenting those things somewhere that other humans and/or bots can deal with would be a good thing. Thryduulf (talk) 18:04, 11 September 2026 (UTC)

Yes and please keep the suggestions coming! I mainly wanted to communicate that even if some of these extension ideas don't end up incorporated into this experimental suggestion, they're not being discarded. Isaac (WMF) (talk) 19:48, 11 September 2026 (UTC)

...beyond "we don't want to scan the entire Wikipedia" - is that the reason? Or something else?

Good question, @Asilvering. What you described is accurate, [i] In addition, it would be ideal if this initial dataset includes the types of articles that:

1. You all can imagine this suggestion being the most useful on

2. Includes cases that you think could be particularly complex/not straightforward so we can learn how/if the model fails here

---

i. The size of this initial batch of suggestions will be limited to ~1,000 articles PPelberg (WMF) (talk) 21:07, 10 September 2026 (UTC)

This seems like a perfectly reasonable use of AI on Wikipedia, since it's not actually generating text. I'm looking forward to seeing how it performs. --Ahecht (TALK
PAGE) 20:07, 10 September 2026 (UTC)

Yeah, this sounds like a wonderful tool. It would be awesome if this could be fitted with a API front end so other tools could build on top of it. A while ago I built something (https://wikirefs.toolforge.org/show?page_title=Bronx+Grit+Chamber) which parses a page, pulls out all the individual claims, and matches them up with citations. But I did a half-assed job of it and got it to the point where it was good enough for my purposes. And I know if fails badly on some referencing styles. It would be great if all that low-level crud could be done once, properly, correctly, and then everybody else who wanted to build tools in that space could just take advantage of it instead of reinventing it from scratch. RoySmith (talk) 22:03, 10 September 2026 (UTC)

A couple of other things that would be useful... If you don't find the claim in the cited source, look at the other sources cited in the article. It's not uncommon during editing an article for a properly cited statement to get moved but the citation doesn't move with it. Being able to recover the correct pairing would be valuable. Also, if you can't reach a source's URL, see if you can find some other place that has the same item. For example, I'll often cite a NY Times article via the NYT's own archives, which you may not be able to get to because it's behind a paywall, but you can find the same article in ProQuest or some other aggregator. RoySmith (talk) 22:52, 10 September 2026 (UTC)

Combining source verification with finding the edit which added the source to the article often helps with this, could be a useful extentsion. It also helps when existing text in front of a source is modified without reference to the source. CMD (talk) 00:38, 11 September 2026 (UTC)

I think that you can combine the script with "Who Wrote That" already, but you're that it would be useful to see it at a glance. Alaexis_¿question? 11:06, 11 September 2026 (UTC)

@RoySmith, it would be great to integrate with TWL and with the Internet Archive to get access to sources that aren't publicly available. Earwig's Copyvio detector has access to TWL so there is a precedent.

For citation that failed with the "Not Supported - Omission" verdict checking other sources in the article makes sense. It's not necessarily cheap - if you have 30 unsupported citations out of 300 total you'll need to run 30*300=9,000 checks, unless there is some kind of screening (the passages nearest where the claim sits or the sources added in the same edits). I tried building a standalone script (User:Alaexis/CNfirmed) to solve the broader problem of finding reliable sources. It turned out to be harder than I thought though. The biggest problem is *where* to look - generic web search is expensive and the alternatives are brittle. Alaexis_¿question? 10:56, 11 September 2026 (UTC)

It would be great if all that low-level crud could be done once, properly, correctly, and then everybody else who wanted to build tools in that space could just take advantage of it instead of reinventing it from scratch.

@RoySmith to be doubly sure I'm following, by "low-level crud" are you referring to things like splitting the article up into its constituent claims, identifying the source(s) associated with each, etc.?

A while ago I built something...which parses a page, pulls out all the individual claims, and matches them up with citations

Neat! If you happen to have them handy, we'd be eager to learn what referencing styles you noticed this tool struggling with.

...everybody else who wanted to build tools in that space could just take advantage of it instead of reinventing it from scratch.

I can imagine the creativity something like this could inspire. With this said, I think we're still a ways away from being able to determine how feasible something like this would be. Once you confirm the first question I posed above, I'm thinking I can create a phabricator ticket so that we can come back to it at a future point. PPelberg (WMF) (talk) 00:39, 11 September 2026 (UTC)

Yeah, I envision some kind of network API where you give it a page title (revid, whatever) and it gives you back a list of claims and the associated citations in some structured form. Of course, this is really just one specific example of a general pattern. You really should be able to treat every page as an object on which you can perform operations and access those operations via a network API.

We spend too much time and effort reinventing wheels. For example, I don't know how many tools we've got which measure the readable prose size of an article. They all give different results because they all have slightly different definitions of what "readable prose" means. Neither are more right than any other, they're just different. And it's dumb that so many people have spent so much time writing essentially the same function in slightly different ways.

To go back to this particular example, there's already a bunch of different referencing styles in use. My code parses one of them (the one I tend to use) pretty well. I know it fails on others (to be honest, I don't remember which). Imagine a world where parsing claims and citations from articles was handled by one API that handled them all. Then all sorts of tools could take advantage of that. And more to the point when a new format comes along (say, sub-referencing, which is going to hit enwiki Real Soon Now), one bit of code will need to be adapted to handle that, and automatically all the various other tools will now be able to handle it too. RoySmith (talk) 01:12, 11 September 2026 (UTC)

@RoySmith, that's so true. I had to build a lot of non-core stuff and would've been happy to reuse others' work instead. Discoverability of tools is a huge topic, Toolhub-Evolved is the latest solution I'm aware of.

Currently verification is not exposed as a standalone service but that wouldn't be too hard to do if you have a use case in mind. Fetching website data is already a standalone service on Toolforge (repo, ToolHub page). Alaexis_¿question? 10:15, 11 September 2026 (UTC)

Thank you for bringing this discussion here. As someone who's used Alaexis's tool on and off for a few months, this is the kind of collaboration I like to see: working with an established editor to support development or rollout of a tool that's already been field-tested by the community.

To answer your questions: 1. Yes, definitely. 2. Dreamyshade has been building https://projo.toolforge.org/review to identify high-priority articles for WP:NPP reviewers – another opportunity for collaboration or sharing ideas? —ClaudineChionh (she/her · talk · email) 00:51, 11 September 2026 (UTC)

Yes, thank you Claudine! The Projo unreviewed article priority-ranking tool (which is new and still in development, still janky) reflects my understanding of how to prioritize articles that need editor help in general, based on factors including CTOPs, BLPs, reference need, page views, orphan status, and cleanup tags such as AI-generated, COI, and POV (the page contains a table of all the factors). It's focused on supporting New Pages patrollers because that is the most urgent work that I know of right now, with the giant backlog and constant influx of COI/UPE and LLM-generated articles. I'm adding more factors based on data I can derive from categories and other elements in the replica database. I'd love to have access to APIs for predicted percentage of LLM-generated content and predicted numbers of verified/partially-verified/failed-verification citations!

I use Alaexis' Source Verifier a lot, especially for Articles for Creation review, New Pages patrol, and AI cleanup tasks. I find it very helpful, especially on articles that are tagged as likely AI-generated or that I suspect are AI-generated. It helps me rapidly check source-text integrity, which helps me figure out whether an article is mostly fine, salvageable, or trash. It's part of my standard toolkit for article quality evaluation, along with https://copyvios.toolforge.org/, https://wikipedia.gptzero.me/, and Cite Unseen - none of them are perfect, just tools, but I believe they're useful when used with competence and a grain of salt. I've been trying to gather lists of tools along these lines: Article workflows#Tools for specific workflow steps + User:Dreamyshade/Article workflows#Semi-automation. (That page has some of the ideas I'm trying to put into practice in the Projo tool.) Dreamyshade (talk) 01:48, 11 September 2026 (UTC)

Wouldn't it not be possible to provide this System access to paywalled content over the Wikipedia Library? The Other Karma (talk) 09:44, 12 September 2026 (UTC)

@The Other Karma, that's the single biggest thing that would extend coverage. Earwig's Copyvio Detector already reaches EBSCO through The Wikipedia Library (phab:T378077), but those credentials were granted for copyright enforcement only, so verification would need its own permission. Flagging this for Meta:User:Samwalton9-WMF as one more editor asking. Alaexis_¿question? 06:21, 14 September 2026 (UTC)

This looks like an actually productive use of AI on Wikipedia. Given the fact that it's not creating new information, it seems fine to me. I think a test run of this may be promising to introduce new editors. 🪐Kepler-1229b | talk | contribs🪐 17:23, 13 September 2026 (UTC)

As for the questions, 1. Possibly but my main editing area lies outside of source verification, so I might not be able to test it as much. 2. Articles with "Failed verification" tags could be first priority for testing. 🪐Kepler-1229b | talk | contribs🪐 17:25, 13 September 2026 (UTC)

For anyone interested in seeing a demo and talking about this in a voice call, we will be hosting a meeting in the Wikimedia Community Discord on 14 Sep 2026 from 17:00 - 18:00 UTC.

This Discord call is happening tomorrow (Monday). Here is a link for anyone interested in joining: https://discord.gg/wikipedia?event=1547782657969098752.

Note: we'll continue being responsive in this thread too. PPelberg (WMF) (talk) 04:51, 14 September 2026 (UTC)

This seems really interesting, and a good usage of LLMs in Wikipedia. Thank you for the open communication too. qcne (talk) 12:02, 14 September 2026 (UTC)

I'll be interested to see how this goes. LLMs are very bad at subtlety and being overconfident on specific answers, but very good when delivery an array of possible answers. Checking source verification could fit very well into the second type, displaying an array of possible issues that may require user intervention is something that could be very useful. -- LCU ActivelyDisinterested «@» °∆t° 20:41, 17 September 2026 (UTC)

It's a much better fit than an LLM trying to understand NPOV, see my point about subtlety and overconfidence.
A common problem in verification is that an editor will add new unsourced content between a sentence and it's reference. So the content then looks like it's supported by a reference, but that's only partially true. Even if this could only surface those issues it would be a major win. -- LCU ActivelyDisinterested «@» °∆t° 20:47, 17 September 2026 (UTC)

Sorry, I'm a bit confused by the question: "Would any of you all be interested in evaluating a batch of these experimental suggestions for en.wiki articles?" I signed up for what sounds exactly like this maybe six weeks or two months ago. The entry test was very buggy, but it seemed to be getting updated. Then nothing. Is this the same project or unrelated? Johnjbarton (talk) 22:17, 28 September 2026 (UTC)

Thanks for the pro-active post on this. I have some reservations based on common things that can go wrong (as well as other things I would prefer be prioritized) but will wait to see the actual suggestions first. Gnomingstuff (talk) 06:09, 14 September 2026 (UTC)

I have some reservations based on common things that can go wrong (as well as other things I would prefer be prioritized) but will wait to see the actual suggestions first.

Understood and sounds great. We'll be in touch in this thread once we have the initial set of suggestions generated for y'all to review.

Before that, we'll be in touch with what criteria we're proposing to use to select the initial 1,000 articles we'll generate suggestions for to make sure we think it will produce a useful sample.

Thanks for the pro-active post on this.

You bet; thank you for demonstrating to us the need and value in doing so, @Gnomingstuff. PPelberg (WMF) (talk) 22:44, 14 September 2026 (UTC)

Verify Endpoint

@RoySmith:, you suggested that it would be useful to have the "verify claim against source" exposed as an API endpoint, so I've done it, details here. I've checked it myself, so feel free to tinker with it too. To avoid any doubt, the edit suggestions pilot doesn't use this endpoint. Alaexis_¿question? 19:00, 29 September 2026 (UTC)

Cool, thanks, I'll take a look. -- RoySmith (talk) 19:31, 29 September 2026 (UTC)

Dataset proposal

Hi y'all – below is the criteria we are planning to use to select the initial ~1,000 articles we will generate source verification suggestions for.

If you see criteria that are included or missing from the below that you think is critical to reviewing the reliability of this suggestion, please comment as much.

Article selection criteria

We are proposing to generate an initial batch of source verification suggestions for ~1,000 articles using the following selection criteria. The criteria you see are meant to reflect what we heard from you all: this dataset ought to be representative of where you see these suggestions being most useful and potentially, risky.

Selection criteria	Description	Motivation
Evergreen	A set of 100 articles User:Alaexis has predicted are likely to be edited within the next 30 days.	This is a set of popular and frequently edited articles where the impact of inaccurate claims is higher and where the suggestions are more likely to be seen and acted upon.
Category:BLP	Biographies of living people	Articles susceptible to mis-attributed/incorrect information and where evaluating all of them might be unrealistic. Automation to help facilitate this process could make this task more feasible for more people on an ongoing basis.
Category:CTOP	Contentious topics
Category:History	Historical articles	Articles that are prone to inconsistencies and thus more likely to contain claims that need verification[1]
{{dubious}}	Articles where at least one {{dubious}} template is present	Strong signal that unsubstantiated claims may be present; offers a helpful reference to benchmark the model against[2]
{{failed verification}}	Articles where at least one {{failed verification}} template is present
{{AI-generated}}	Articles volunteers have concluded to be AI-generated	Articles more likely to contain hallucinated references (read: claims that are not substantiated by the sources that accompany them)
Good article nominations	Articles nominated for "Good" article status	Thorough review of all claims within article nominations is necessary and toilsome.
NPP Backlog	Articles created in 2026 that are not reviewed yet	There is a long backlog of new articles for review and being able to increase the speed with which volunteers can evaluate verifiability could help decrease it.

Next steps In terms of process, here is how we see things going from here:

• Step #1: Use the criteria above to select ~1,000 articles we will create an initial batch of source verification suggestions for. ← we are here
• Step #2: Conduct an internal review of these suggestions to ensure they are of sufficient quality to warrant y'alls (volunteer) attention
• Step #3: Share the results of the internal review and invite volunteers (you all) to conduct a second review of the internally-vetted dataset.
• Step #4: Make this initial batch of suggestions available for volunteer review in two ways:
  1. In bulk via a spreadsheet
  2. In context via an experimental suggestion within Suggestion Mode
• Step #5: Analyze volunteer evaluations
• Step #6: Share findings and discuss next steps on-wiki

How does this all sound?

PPelberg (WMF) (talk) 16:17, 24 September 2026 (UTC)

This sounds like a solid plan. I must emphasize the importance of being honest with yourselves in the internal review – if you're having to, say, regenerate half the suggestions because of obvious errors, that might mean more work is required on the underlying tool before volunteers test it. Please tell us exactly what you find there, with numbers. I also wonder, if each of these nine categories gets ~1/9th of the 1000 articles, do we have enough statistical power to do whatever analysis you are hoping to do? I'm not a statistician, and 100+ articles per category is still a lot, but that's something you want to check beforehand. Finally, would this check the entire article on the articles where it's run? For instance, if a page has an inline failed verification tag, I would be disappointed if the program was set to generate, say, 3 suggestions per article max and then didn't check the sentence tagged as failed verification. For the NPP category, I suggest choosing only articles that have sat in the NPP queue for a week or two to make them less likely to be instantly reviewed and fixed up by someone patrolling the front of the queue. Toadspike [Talk] 16:50, 24 September 2026 (UTC)

This sounds like a solid plan. I must emphasize the importance of being honest with yourselves in the internal review – if you're having to, say, regenerate half the suggestions because of obvious errors, that might mean more work is required on the underlying tool before volunteers test it. Please tell us exactly what you find there, with numbers.

Well put and agreed. To be explicit, following the internal review, you can expect us to share:

1) How many suggestions we reviewed

2) What a review of each of these suggestions entailed

3) What the quantitative and qualitative outcomes of these reviews were

4) What we think the next step(s) are in the light of the above.

I also wonder, if each of these nine categories gets ~1/9th of the 1000 articles, do we have enough statistical power to do whatever analysis you are hoping to do? I'm not a statistician, and 100+ articles per category is still a lot, but that's something you want to check beforehand.

First, to the question you asked: we will be generating suggestions for all claims within the articles in the dataset, not just a subset. In line with the above, we expect this approach to produce a dataset of, at least, 10,000 suggestions. I think this will provide the scale we need to draw meaningful conclusions. Tho, I defer to @Isaac (WMF) to say definitively here.

For the NPP category, I suggest choosing only articles that have sat in the NPP queue for a week or two to make them less likely to be instantly reviewed and fixed up by someone patrolling the front of the queue.

Good spot. I think incorporating this constraint shouldn't be too difficult for this run. Although, if that proves to be the case, we'll let you know as much.

Thank you for thinking this through, @Toadspike. PPelberg (WMF) (talk) 21:24, 24 September 2026 (UTC)

I think incorporating this constraint shouldn't be too difficult for this run. Although, if that proves to be the case, we'll let you know as much.

Ok! It turns out that implementing the above at this point could have unintended side effects. So, for this first run, we'll not apply any date-based criteria to the articles we analyze from the NPP queue. "Worst" case, we'll see fewer suggestions there because they will be, as you hypothesized, of better quality. PPelberg (WMF) (talk) 21:31, 24 September 2026 (UTC)

Thanks, sounds good. I look forward to being able to get started on this. Toadspike [Talk] 21:45, 24 September 2026 (UTC)

Wonderful. I expect us to complete the internal review and have results to share the week October 12th. PPelberg (WMF) (talk) 21:59, 27 September 2026 (UTC)

Invitation to join the conversation on future of affiliates

Hi everyone!

I am Kaarel, working with the Wikimedia Foundation, facilitating conversations regarding the future of affiliate landscape. We would love to get your thoughts on the current discussions on recognition and funding of movement organizations.

We’ve heard many times (for example here and here) about problems related to grant funded projects and that's why it’s incredibly important that your voices, ideas, and expectations are heard.

We just published a summary report from our July and August conversations. It highlights what people agree on, where they disagree, and what still needs clearing up. We hope this report gives you a good overview of the chat so far and inspires you to jump in! Your perspective as a project contributor is vital to shaping the future of this model.

Please head over to Meta (general discussion, affiliate model discussion, funding discussion), to share your thoughts. If you have any questions or just want to chat through it, feel free to reach out to me. I am keen to set that up, listen and talk through.

We will also be hosting two public community calls next week and the week after: 1) Thursday, September 24, 16:30-17:30 UTC on affiliate overlaps 1) Tuesday, September 29, 16:00-17:00 UTC on affiliate growth paths - LINK TO THE CALL and 2) Thursday, September 24, 16:30-17:30 UTC rescheduled to Friday, October 2nd, 17:00-18:00 UTC on affiliate overlaps - LINK TO THE CALL.

Thank you for your very kind attention and have a great rest of your week! --KVaidla (WMF) (talk) 08:40, 16 September 2026 (UTC)

This is a real opportunity to bring editors and affiliates even closer together. But it does require some changes on the affiliate front and that's not easy. I know many editors are like "affiliates have nothing to do with me, so why should I care about this?" But I think "affiliates have nothing to do with me" doesn't have to be the way it is and more importantly shouldn't be the way it is. I hope other editors will join me in expressing a desire to see affiliates refocus their work in ways that tangibly improve projects, rather than continuing to be their own separate thing. Best, Barkeep49 (talk) 14:12, 16 September 2026 (UTC)

---

I couldn't help but notice this: "The present text has been revised based on the valuable feedback from various community stakeholder groups, including the Affiliations Committee, the Global Resources Distribution Committee, and directors of existing movement organizations".

Notably missing are donors, people who read Wikipedia but do not edit, and the community of Wikipedia editors. I would welcome evidence that there was even a small effort made to get feedback from these missing stakeholders.

From Wikipedia Foundation exec: Yes, we've been wasting your money in The Register and The WMF Executive Director’s Reflections on the FDC Process on Meta (which I recommend reading in full for context):

"I have significant concerns about how our movement entities are developing... I believe that currently, too large a proportion of the movement's money is being spent by the chapters...

I am not sure that the additional value created by movement entities such as chapters justifies the financial cost...

I do also believe that people who are involved in chapter organizations (and other Wikimedia organizations) have a particular worldview that is in some ways different from that of Wikimedians who choose not to become involved with incorporated Wikimedia organizations, and I think a healthy funds dissemination process would benefit from multiple perspectives...

[The] process, dominated by fund-seekers, does not as currently constructed offer sufficient protection against log-rolling, self-dealing, and other corrupt practices...

The community members who are paying the closest attention to the process are applicants, and that community involvement in scrutinizing proposals is otherwise low...

There is currently not much evidence suggesting this spending is significantly helping us to achieve the Wikimedia mission."

I see little evidence that these fundamental problems have been addressed in the 13 years since they were posted.

And before someone says it, Wikipedia:Village pump (WMF) is the place where the WMF should look to get feedback from the community of Wikipedia English Wikipedia editors. not a talk page on meta where comments get few or no responses. --Guy Macon (talk) 15:08, 16 September 2026 (UTC)

So do you think the changes the WMF has proposed will help donor money be spent better or do you think there should be different changes to affiliates (it's clear you don't think the status quo is good)? I don't think there's anything wrong with us posting here, I've posted substantive comments myself after all, but enwiki editors are not the same as "wikipedia editors" and so having a central place, like meta, that should be open to all seems like an obvious answer for me. Best, Barkeep49 (talk) 15:15, 16 September 2026 (UTC)

It should also be noted that Wikipedias are not the only WMF projects and editors of those projects should also be able to have their say without requiring WMF staff to read dozens of independent discussions. Thryduulf (talk) 15:52, 16 September 2026 (UTC)

I think that what the WMF is doing is well thought out and will certainly help to insure that donor money is spent better. On the other hand, I also think that those on the receiving end are very strongly motivated to maximize how much cash flows their way, and will do whatever it takes to accomplish that.

The problem is that I don't see anyone surveying a sample of donors and asking them whether this is what they had in mind when they donated. I think that if you asked them the overwhelming majority would say that the money should go to things like hosting.

The WMF knows this. That's why the latest banner ads say "We hope that [Wikipedia] has given you at least $2.75 of knowledge. If so, please join the 2% of readers who give to keep this resource available for all".

Note that they did not say "...who give to match racial justice leaders with machine learning research engineers to develop data-based machine learning applications."

I am reminded of the many smart people who asked "are we spending money of the right things in Vietnam? Should we fund better rifles or better boots? Should we buy more tanks or more jeeps?" all without ever asking "should we be fighting a war in Vietnam at all?" I think you can guess what the answers from the "stakeholders" who supplied the boots and the jeeps were. --Guy Macon (talk) 17:44, 16 September 2026 (UTC)

The WMF does seem to spend an awful lot of money on causes which are noble but totally unrelated to what donors thought they were funding. No one is here to oppose worthy aims such as racial justice but they have their own dedicated charities. Indeed, some of our donors may also be sponsoring those organisations. Either way, they're entitled to have the money they allocated to Wikipedia spent here and not diverted to unexpected social campaigns.

Certes (talk) 19:07, 16 September 2026 (UTC)

re your footnote, there are also links to donate from the sidebar of all the other projects I looked at except Commons and MediaWiki (usually as "donate" but sometimes as "donations"). Thryduulf (talk) 19:19, 16 September 2026 (UTC)

Guy, they didn't say that because they aren't doing that. The Knowledge Equity Fund was a three-year program and is now over. In solidarity, asilvering (talk) 20:33, 16 September 2026 (UTC)

Pick whatever they are spending money on that is unrelated to "keeping this resource available for all" this week and insert that. The specific spending on things totally unrelated to what donors thought they were funding keeps changing and is often only discovered after they have moved on to the next new thing, but we all know that the pattern is ongoing. --Guy Macon (talk) 20:49, 16 September 2026 (UTC)

@Asilvering, I'm not sure this is the case. Consider the projects listed here. Yes, they have "articles created" and "editors retained" metrics and all that, but one gets the feeling that these goals and their monitoring is not the central concern in these programs.Alaexis_¿question? 06:05, 28 September 2026 (UTC)

Scholarships for Wikimania 2027

Scholarship applications are now being accepted for Wikimania 2027, which will take place August 18 to 21 in Santiago, Chile. The final deadline to submit your application is October 31, 2026 (end of day Anywhere on Earth).

Scholarships cover travel and accommodations, registration, and limited travel insurance fees. Wiki project contributors and Wikimedians working to connect our movement with the wider open knowledge and culture ecosystem are invited to apply.

Tell us about your Wikimedia journey, share your vision for “The Internet we want” (the theme for Wikimania 2027), and be sure to include links to your work. All in-person attendees, including scholars, will be subject to trust and safety checks.

More detailed information on the application process can be found on the Scholarships section of the Wikimania wiki, and in this Diff post.

Orientation sessions for potential applicants will be offered starting the week of October 5, 2026. Successful applicants will be notified starting in January 2027.

Best of luck to all! Eureka-WMF (talk) 15:04, 22 September 2026 (UTC)

Join an Orientation Session with tips on how to apply for a scholarship to Wikimania 2027!

• Friday, October 9, 2026 – 3:00 PM UTC – English and Portuguese (Zoom link)
• Friday, October 9, 2026 – 4:30 PM UTC – Spanish (Zoom link)

Eureka-WMF (talk) 19:25, 1 October 2026 (UTC)

Should Wikimedia administrators be paid by wikimedia foundation?

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.

---

Hello, should administrators on different wikimedia projects like Wikipedia be paid? I think yes. Botaki (talk) 12:28, 25 September 2026 (UTC)

Speaking as an administrator who could certainly do with more money, no. Thryduulf (talk) 13:12, 25 September 2026 (UTC)

There is an old saying: "He who pays the piper calls the tune." Right now the administrator permissions are granted and managed by the individual communities, without any input from the WMF with extremely rare circumstances. If the WMF starts paying us, they will have to assume control of that, as is necessary in any employer/employee relationship. I don't really think the community wants to give up control of who becomes an administrator, and what rules admins will enforce, and what tools they have at their disposal. So no, I don't think the WMF should pay administrators. Speaking personally, I don't want to be a WMF employee, thank you very much. Risker (talk) 13:34, 25 September 2026 (UTC)

Go ahead and pay them. Nobody is stopping you. (Exception: paying an admin to do something will get you both booted from Wikipedia, but setting up a fund that is shared equally by all admins willing to accept payment is probably fine -- but check first instead of taking my advice.)

Oh, wait. I missed the "by wikimedia foundation" part. Sorry. Silly mistake. The WMF has no control over what administrators or editors do (again, with exceptions. If the admins were to, say, decide to allow child pornography or copyright infringement the WMF has a legal responsibility to stop them. See WP:OFFICE) I don't see how being independent is compatible with admins being paid employees of the WMF.

Also, we already have far too many individuals who appear to be far more interested in keeping the grant money flowing than in anything that actually helps any of the wikis. See Where does your Wikipedia donation go? Outgoing chief warns of potential corruption. --Guy Macon (talk) 13:40, 25 September 2026 (UTC)

Ignoring all the other obvious concerns, were the WMF to start paying admins, they (the WMF) would be legally accountable for their actions. I doubt very much they'd wish to take on that responsibility, as they'd likely find themselves involved in an endless stream of lawsuits. AndyTheGrump (talk) 13:59, 25 September 2026 (UTC)

I think no. Sohom (talk) 14:12, 25 September 2026 (UTC)

I think our pay should be at least doubled. _signed, Rosguill ^talk 14:51, 25 September 2026 (UTC)

If you keep making these unreasonable demands, we're going to cut your pay in half! Levivich (talk) 19:09, 25 September 2026 (UTC)

How much do I get if I make reasonable demands? Anyway, I'm not too worried about the salary, I'm mostly in it for the stock options. RoySmith (talk) 22:28, 25 September 2026 (UTC)

(Not an admin) I would second the above comments, although not being an admin (thankfully) my opinion might not be worth much. A two-tier system would potentially create division in some editors minds. At the moment, admins are editors with extra buttons and (usually) a good understanding of how Wikipedia works.

The cabal shouters don't need any further encouragement! Knitsey (talk) 22:35, 25 September 2026 (UTC)

There Is No Cabal (TINC). We discussed this at the last Cabal meeting, and everyone agreed that There Is No Cabal. An announcement was made in Cabalist: The Official Newsletter of The Cabal making it clear that There Is No Cabal. The words "There Is No Cabal" are in ten-foot letters on the side of the 42-story International Cabal Headquarters, and an announcement that There Is No Cabal is shown at the start of every program on The Cabal Network. If that doesn't convince people that There Is No Cabal, I don't know what will. --Guy Macon (talk) 22:47, 25 September 2026 (UTC)

If only you paid me my cabalbucks. Knitsey (talk) 22:51, 25 September 2026 (UTC)

This would create all kinds of unsolvable problems. Admins would in many countries be considered employed by the WMF, but would be elected by the volunteers and the volunteers would be able to fire them via the recall process. The legal ramifications of that would cause complete chaos. -- LCU ActivelyDisinterested «@» °∆t° 00:07, 26 September 2026 (UTC)

Paid by the WMF? No. I do recall at least one editor who had a Patron, though. And I have wondered at times whether folks should experiment with a "tip jar" sort of thing, giving readers an option to thank volunteers other than just donating to the WMF. Personally, I'm skeptical that it's possible to introduce money in any of these ways without considerable unintended consequences. — Rhododendrites ^talk \\ 02:51, 26 September 2026 (UTC)

Patreon, I suppose? Alaexis_¿question? 17:30, 27 September 2026 (UTC)

Not to mention, how would you determine pay? It would have to be per admin action, otherwise very busy admins would be paid the same as those that never use their tools at all. And if you're paying per admin action, that sounds like a very good way to get people to (a) rush their actions and get them wrong, and (b) get burnt out. Black Kite (talk) 18:30, 27 September 2026 (UTC)

Also not all admin actions are the same. Protecting a page due to obvious vandalism requires much less effort than closing a contentious AfD, yet both would (presumably) count as a single action. It would also need to be decided whether CU and OS actions count as admin actions and if so at what level (one CU check typically results in more log entries than dealing with one OS request). This could also lead to competition between administrators to be the one to record the logged action rather than collaborating with their colleagues to get the right outcome. Thryduulf (talk) 18:41, 27 September 2026 (UTC)

Another thing that probably is worth keeping in mind is that taking a decision to not do a admin action is sometimes as valuable as doing one. Declining a speedy deletion, declining a unblock request or a WP:PERM request is equally as valuable as granting them. Sohom (talk) 19:21, 27 September 2026 (UTC)

If I block somebody and they're later unblocked, is my commission subject to clawback? RoySmith (talk) 20:23, 27 September 2026 (UTC)

And what about panel closes - is it the standard fee for each member or do they have to share one? Thryduulf (talk) 20:41, 27 September 2026 (UTC)

This is a terrible idea and would remove the independence of admins by introducing a massive conflict of interest. It's also just unworkable in general: there are way too many admins throughout the numerous Wikipedia languages and other projects, and there are no clear rules for how many there should be, creating an incentive to increase their number to get more money (or worse, forcing the WMF to determine how the projects choose admins) Ita140188 (talk) 07:54, 28 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.

For those interested in where the money goes

Please review:

• meta:Grants:Programs/Wikimedia Community Fund/Review/2021-22
• meta:Grants:Programs/Wikimedia Community Fund/Review/2022-23
• meta:Grants:Programs/Wikimedia Community Fund/Review/2023-24
• meta:Grants:Programs/Wikimedia Community Fund/Review/2024-25
• meta:Grants:Programs/Wikimedia Community Fund/Review/2025-26

There is a lot to unpack there, so please post anything you find that seems worth discussing. --Guy Macon (talk) 18:38, 28 September 2026 (UTC)

Slow Editing Towards Equity

I will start with one that caught my eye:

• Project page: meta:Research:Slow Editing Towards Equity
• Amount Spent: $42,356.25 USD
• Result: meta:Research:Slow Editing Towards Equity#Results
• Promised result: meta:Grants:Programs/Wikimedia Research Fund/Slow Editing towards Equity#Impact
• Related: (that last one makes some interesting claims about Wikipedia)

Was this $42K well spent? Who benefited? What was accomplished?

Was this the sort of thing that the donation banners describe your contributions as funding?

Why are some of the links at dead or useless? Are we not capable of posting these publications on WMF servers? --Guy Macon (talk) 18:38, 28 September 2026 (UTC)

Takeaways from the linked paper at [14]:

Theory Section

- The power to shape consensus on Wikipedia is not evenly distributed. Even aside from groups like ArbCom, things like technical knowledge, access to bots, and understanding of the site's jargon gives the small slice of the community with mastery over them ("experienced users") an overwhelming presence in policy discussions.

- The rules pages act as entrenched technical authority that can be used by experienced users against outsiders, not merely as consensus records.

- As proof of the above, most policy pages are very old. If a page survives its first year of official status, it gains enough momentum to be effectively permanent. 11 of the 15 sampled policy pages were fully stabilized by 2011.

- The rules pages are very stable, and that's good. However, they are resistant to change in a way that discourages any new user that isn't already aligned with Wikipedia's way of thinking. Such users don't become experienced users with the ability to shape discussion, and so can't reform the very rules preventing their participation. This is a barrier to gaining new editors from outside the spheres Wikipedia originated in (American male tech nerds).

- Wikipedia's policies, both content and social, can be used as a cudgel against users who threaten local consensus and/or individual editors' worldviews. Examples cited.

- All of the above to say, Wikipedia's current policies fail to reflect actual community consensus due to how they empower users who already agree with them.

There's more stuff about the actual form said empowerment takes, lot of sociology jargon and such. It records the various levels of authority a rules page can have, including unofficial levels like "essay endorsed by popular community members."

Data Section

- The number of individual users involved in creating each policy is very low. It's mostly the same handful of people talking to each other. There were more "critical discursive moments" identified in the sample than unique editors.

- enwiki prefers discussion among a few people with strong opinons and arguments over the flatter voting system of eswiki. The other three languages didn't have much of either, and just kind of trusted whoever was writing the rules.

- Recommendation made that the processes used to define and change policy be adjusted to better reflect actual editor consensus rather than consensus among the technocratically empowered few. No hard suggestion included.

Overall I think it's a useful study, if not for Wikipedia itself then certainly for the history and sociology of internet communities. I'm not mad about it having been funded. ~2026-52337-89 (talk) 23:56, 28 September 2026 (UTC)

The promise (when they were asking for $42K) was "identifying gaps in policy that are necessary for the needs of all English Wikipedians." Please name a gap in policy that was identified, and how we can fix that gap.

And again I ask, Was this the sort of thing that the donation banners describe your contributions as funding? See The Huge Fight Behind Those Pop-Up Fundraising Banners on Wikipedia in Slate --Guy Macon (talk) 02:14, 29 September 2026 (UTC)

Thinking on how to resolve this, perhaps that's the solution - we establish a consensus that donations banners run on enwiki must accurately detail where the money is going. If something is beyond the scope of what donation banners said the WMF is spending money on, then either the WMF needs to cut the program, or not run banners on enwiki. BilledMammal (talk) 02:19, 29 September 2026 (UTC)

@Guy Macon, I don't think expecting research to "succeed" in doing something 100% of the time is a reasonable expectation to have. Research can and does "fail" even with the best effort. Doing research on policies on Wikipedia is within Wikimedia Foundation's overarching programmatic goals to better the community to which donators donate to. There was a certain amount of funding allocated to the researchers, they tried to gain some useful information, they failed in the overall goal but found some interesting findings about internet communities which got published in a (what I assume) is a well respected journal. Sohom (talk) 02:55, 29 September 2026 (UTC)

No need to ping me. I have a watchlist and know how to use it.

Useful? Sure. $42,356 USD of the donor's contributions useful? Not so much. A lot of those donors are living in poverty and sacrifice to give because they were told that this was needed to keep Wikipedia free.

There are many, many researchers who have published papers on Wikipedia in well respected journals. I question why only a handful of insiders get funding for doing that from the WMF. --Guy Macon (talk) 03:06, 29 September 2026 (UTC)

seful? Sure. $42,356 USD of the donor's contributions useful?, that's how research works??? You provide a chunk of funding so that a group of researchers can try different methodologies and techniques and build prototypes to see if any work. If anything, 42K is fairly cheap in the research world for the length of time the project took (2022-2025) actually since if you think about it, a single grad student's yearly tuition and stipend in CS (being paid barely enough to sustain themselves, while also being one of the most funding rich areas) amounts to roughly that number.

I question why only a handful of insiders get funding for doing that from the WMF.. Guy Macon, I would suggest looking at the page for the dissemination of research funds which provides full transparency into who is allowed to put in proposals, how much money WMF is willing to give for research projects, the proposals submitted and rejected, the folks who decide who to fund (who are well respected researchers within the Wikimedia Research community). I don't think characterizing this as "insiders get funding" is a correct characterization. Sohom (talk) 03:22, 29 September 2026 (UTC)

As I have said many times, I think that everyone at the WMF -- from engineers to the people who decide who and what to to fund, are doing a great job at what they are trying to do. I have zero doubt that you are really looking hard at who you decide to pay to do research, and by "insiders" I don't mean "your buddies" or anything implying corruption or malfeasance. I believe you are like the fine people back in the 1960s who were managing the war in Vietnam. They did seriously great work deciding whether to fund more jeeps or more tanks, whether the jungle boots could be made better, etc. Thousands of good decisions were being made by dedicated individuals doing a great job, but never once considering or even being allowed to consider whether we should be fighting a war in Vietnam at all.

Now consider the entire world population of people who have published research on Wikipedia. All of them. They all get paid by someone, usually their university. What percentage get WMF grants? Do an honest estimate. One in a thousand? What differentiates the two groups? One group knows how to craft a proposal that will get them funding in the usual ways academics get funding. The other has learned how to work the system to get WMF funding. How many of each group get free travel and lodging to go to Paris and spend time with Wikimedia staff and Wikipedia editors? How many in each group know the names of multiple WMF employees? How can you call the tiny minority that get grants anything other than insiders? --Guy Macon (talk) 06:47, 29 September 2026 (UTC)

And again I ask, was this the sort of thing that the donation banners describe your contributions as funding? --Guy Macon (talk) 06:47, 29 September 2026 (UTC)

There's lots of things the WMF is doing badly, but I don't think this particular funding decision is one of them. The article seems to be of good quality to me, and seems in line with the research questions in the application. TietoTeekkari (talk) 07:56, 29 September 2026 (UTC)

And I fully agree, just as I agree that the military made some great decisions about funding jungle boots during the Vietnam war. They were great boots. I still have a pair and they look like new. And, just as I think that same military made those great boot decisions without even considering whether we should be there we should be fighting a war is Vietnam, I think the WMF did a fine job of picking which researcher to fund while never even considering whether we should be funding academic papers at all. And again I ask, was this the sort of thing that the donation banners describe your contributions as funding? Is there some special feature in Wikipedia's software that makes the preceding words invisible when I write them? --Guy Macon (talk) 13:50, 29 September 2026 (UTC)

Not Tieto, but the last time I donated to Wikipedia I recall the banner saying something about "supporting and defending open knowledge projects." I'm probably not getting the phrasing exactly right. This research seems very relevant to Wikipedia and how our internal community and governance systems operate, along with how community-based open knowledge projects operate more broadly. ThadeusOfNazereth(he/him)^{Talk to Me!} 19:17, 29 September 2026 (UTC)

You mean this banner? The actual promise was "Please join the 2% of readers who give what they can to to help keep this valuable resource ad-free, up-to-date, and available for all". Nothing about supporting and defending open knowledge projects. The banner also talks about what you are doing when you visit Wikipedia, which I thought was a nice addition. Please note that something can be "very relevant to Wikipedia" without being what the donors were told their donations were funding.

You can see a bunch of fundraising banners here. I couldn't find any documentation as to which were shown when and to who. Perhaps someone else can find that information. --Guy Macon (talk) 23:45, 29 September 2026 (UTC)

Guy, it sounds to me that you want the WMF to support Wikimedia and open knowledge projects but to do no research into how to support those projects, no research into how effective support for those projects is, whether support would be better directed towards maintaining existing (governance) structures or implementing different (governance) structures? If so, why do you think that? If not, please try explaining what I'm getting wrong. Thryduulf (talk) 19:47, 29 September 2026 (UTC)

Not at all. The choice is not between "funding no research" and believing without evidence that this particular research actually ended up "maintaining/implementing governance structures". Can you name a single governance structure that was maintained or implemented by this research? Look at the results of this research again: The first two results are trivial and the third is just plain wrong.

In the banner at the donors -- many of whom live in poverty -- were not told that they were funding research, just as they were never told that they were funding trips to talk about Wikipedia in exotic vacation destinations. They were told that they were keeping Wikipedia ad-free, up-to-date, and available for all. I am fine with donors funding any research that has a reasonable chance of advancing those goals, but they should be told that some of their donations will be given to other individuals and organizations. --Guy Macon (talk) 23:45, 29 September 2026 (UTC)

Research does not have to be successful to have been valuable, and if you only fund research that produces the results that you want it to then that's not research. Please explain how funding people and projects that write and maintain the projects, write and maintain the sources we use, research how the project can be improved, research how best to support the projects, etc. are not "keeping Wikipedia ad-free, up-to-date and available for all" either directly or indirectly? Thryduulf (talk) 00:04, 30 September 2026 (UTC)

By those criteria I can't think of a single thing that the WMF spends money on that doesn't keep Wikipedia ad-free, up-to-date and available for all either directly or indirectly. I can't see any plausible way that not funding this sort of thing could possibly result in Wikipedia being ad supported, stop the volunteers from updating it, or make it unavailable, but you clearly do so I won't waste your time with further disagreement. --Guy Macon (talk) 00:18, 30 September 2026 (UTC)

Can you name a single governance structure that was maintained or implemented by this research? To this among other research appears to be part of a body of research into policies that eventually led to the establishment of the NPOV Working Group and the subsequent implementation of the global baseline NPOV policy. Sohom (talk) 00:27, 30 September 2026 (UTC)

Just to be clear: I think that Wikiresearch is in general a good thing, and funding some of this by WMF seems reasonable. The specifics of funding, methods, and transparency and ethics of the coordination of the research should, of course, be subject to transparent debate (apart from privacy issues). I'm not judging the validity of funding for this particular project. Boud (talk) 11:39, 30 September 2026 (UTC)

---

I found meta:Baseline NPOV policy (a baseline NPOV policy for the Wikipedias which do not yet have an NPOV policy) to be a fine document. The question is whether it is worth the $42,356.25 USD ($88.43 per word) it cost the donors.

I would be interested in how many Wikipedias did not yet have an NPOV policy when it was created and how many have adopted it. If the donor is paying for "governance structures implemented by this research" it would seem reasonable to ask how many actual governance structures were implemented, not just a web page that seems like it would be useful to someone doing that.

I also noticed that it is only in English and French. One would think that some of that $42,356.25 USD would have been spent on translating it to the languages of the Wikipedias which do not yet have an NPOV policy. I'm just saying. --Guy Macon (talk) 16:14, 30 September 2026 (UTC)

Couldn't small wikis just copy the NPOV policy from English, French or any other well-scrutinised Wikipedia, gratis, and alter any bits they deem inappropriate for their project? Certes (talk) 16:23, 30 September 2026 (UTC)

@Certes, I would suggest looking at Serbian, Bosnian Wikipedia situations where despite copying the policy text over over they've had multiple significant issues surrounding the enforcement of said policies. Similar issues exists to smaller extent on other projects as well. Even on the rather bare-bones baseline policy, smaller communities have pushed back on the baseline policy due to concerns over the fact that it often constrains smaller language wikis in multiple ways in terms of prioritizing spoken history vs colonial written sources or preferring accounts in local language sources over the broader international consensus on a topic. Sohom (talk) 16:38, 30 September 2026 (UTC)

Certes, the real problem in searching for worldwide solutions is that in my experience different language Wikis are effectively different planets with totally different climates. I sometimes look at the Italian, French, German and Spanish wikis, and their culture and customs are strikingly different, and range from those with a highly controlled system (eg German) to others with freewheeling on non-crucial topics. Their wiki cultures adre as different as the foods they eat. You can not give them a universal cookbook. Yesterday, all my dreams... (talk) 18:02, 30 September 2026 (UTC)

@Guy Macon The policy hasn't yet gone into effect. It is still in proposal phase. Also re cost, 42K (and more actually) spent on building policy to avoiding incidents where Wikimedia project's credibility is called into question is imo money fairly well spent. Sohom (talk) 16:40, 30 September 2026 (UTC)

That would indeed be money well spent if that's what the $42K bought. A bargain, I would say. The problem is that you have offered no examples where having One More Web Page On Meta actually avoided any incident where any Wikimedia project's credibility was called into question. Not even a plausible path where that might happen in the future. You seem to be saying that money spent on goals without any hint of a way to actually accomplish those goals is money well spent. I say that $88.43 per word is way above the going price for good intentions, which are currently so cheap that they pave roads with them. --Guy Macon (talk) 22:34, 30 September 2026 (UTC)

The problem is that you have offered no examples where having One More Web Page On Meta actually avoided any incident where any Wikimedia project's credibility was called into question - The m:CheckUser policy and the m:UCoC would be prominent examples of the a "page on meta fixes problems" effect. Sohom (talk) 00:21, 1 October 2026 (UTC)

Please accept my apologies in advance for saying this, but with phrasing like "Wikimedia Foundation's overarching programmatic goals to better the community" you could run for senate. The problem with their research results was similar. Too many words, too little substance. Sorry, but bluntness was needed here. Yesterday, all my dreams... (talk) 15:07, 30 September 2026 (UTC)

Foundation's overall goal of making the wiki community better is what I meant. Ignore the word programmatic, it's a reference to the way WMF and other non-profits calculates it's budget, (as programmatic expenses and non programmatic expenses). Sohom (talk) 16:28, 30 September 2026 (UTC)

Guy, I think $42,355.25 was wasted on the word salad they produced. The other $1 was useful, given that it made me laugh. Yesterday, all my dreams... (talk) 14:56, 30 September 2026 (UTC)

In this edit, Guy Macon wrote, The first two results are trivial and the third is just plain wrong. I think it would be good if Textaural, who is an experienced enwiki editor, could respond here. The peer-reviewed research paper is technically a WP:RS, so usable in enwiki articles, but peer review does not guarantee correctness.

I agree that the third key result The status of rules [on these 5 Wikipedias is] established largely by individual decisions and rarely through collective decision making is extremely dubious for at least enwiki and frwiki. The strings !vote and not vote seem to be completely absent from the published paper. In a nutshell: If I'm the individual to first write the enwiki policy that 1+1=2 is a fact and nobody contests that, it's not due to my dictatorial authority; it's due to the WP:!VOTE decision-making algorithm and the nature of 1+1=2 being widely considered as reasonable, without anybody needing to argue about it (leaving aside modern foundations of mathematics). The whole point of not holding a vote on my hypothetical new policy is that getting 10,000 votes (not !votes) with a likely result of 99.9% in favour would be a huge waste of time. @Textaural: How do you justify your extraordinary conclusion that Within the English-language rules, the authority of rules was based on the editorial authority of one user, who may have found support through limited deliberation without having studied the role of WP:!VOTEs in the establishment of enwiki rules? !VOTEs are a form consensus decision-making, which is a form of collective decision-making. The authority of rules is not based on the editorial authority of one user; it's based (often) on the ability of one user to successfully describe and summarise the current consensus or predict the likely consensus.

Textaural: do you have any evidence for this claim that you and your authors assert in the paper? The Methods section of your paper says nothing about how you quantify the unexpressed non-objections to enwiki rules (a possible method would be to obtain permission to do a survey of Wikipedians to see how much they agree with existing rules and whether or not they have tried objecting to rules they disagree with; but you don't seem to have done that). Boud (talk) 11:07, 30 September 2026 (UTC)

"Rules" !== policies to my understanding. Sohom (talk) 11:23, 30 September 2026 (UTC)

The author gave examples of what they consider to be rules. Two of the examples are in English; Wikipedia:Proposed deletion (a policy) and Wikipedia:Disruptive editing (a behavioral guideline.). --Guy Macon (talk) 12:37, 30 September 2026 (UTC)

Fair enough, my understanding is that the paper discusses the individual rules inside the policies instead of the policy as a whole but I can see it being taken in both ways Sohom (talk) 14:04, 30 September 2026 (UTC)

I also noticed that the author uses https://artandfeminism.org/resources/research/unreliable-guidelines/ as a source. Artandfeminism.org appears to be funded by the WMF through WikiCed ( https://www.wikicred.org/). I can't find any record of how much the WMF gave artandfeminism.org to say "This research project identifies the ways that organizational values and processes around reliability and the reliable source guidelines are implicated in maintaining hierarchies and excluding marginalized knowledges and communities" and "Ultimately, the Reliable Source guidelines are an unreliable and incomplete guide to provide editors with a meaningful understanding of how to assess source reliability".

So what are we supposed to replace our reliable source guidelines with? "Triangulation". If anyone can understand what (cited by artandfeminism.org) is talking about please explain it to me. --Guy Macon (talk) 13:16, 30 September 2026 (UTC)

I don't disagree that Art+Feminism is funded to a certain extent by the Foundation, but last I checked, it wasn't through WikiCred? WikiCred organizes meetups in the West Coast and builds tools for media reliability on Wikipedia, Art+Feminism advocates for better coverage of topics related to gender, feminism and art (among other things). They were one of the advocates for a global UCoC and and have long advocated for changing our guidelines to be more inclusive towards marginalized communities and better enforcement of civility policies. I don't think there is a correlation between WMF funding Art+Feminism, Art+Feminism being cited as community criticism of the reliability guidelines and WikiCred organizing conferences. Sohom (talk) 16:25, 30 September 2026 (UTC)

Re: "last I checked, it wasn't through WikiCred?" you need to do a better job of checking. Look at Read the funding section. What we have here is an informal community of organizations united in the goal of writing proposals that will result in the WMF giving them grant money. If only the WMF would check to see if the goals in the proposal were actually met and, if not, find some other applicant to award the next round of grants to. --Guy Macon (talk) 23:00, 30 September 2026 (UTC)

To quote the paper, Unreliable Guidelines, a project of Reading Together: Reliability and Multilingual Global Communities, is an inaugural Art+Feminism research report partially funded by WikiCred, which supports research, software projects and Wikimedia events about information reliability and credibility. WikiCred and Art+Feminism are together funding the paper with At+Feminism being the primary sponsor and WikiCred being a partial sponsor. The Funding section in particular say The expertise of the main researchers was partially funded by WikiCred and Art+Feminism. i.e. that this project was a collaboration between AF and WikiCred. Nowhere does it say that the Art+Feminism is funded by WikiCred. Sohom (talk) 00:27, 1 October 2026 (UTC)

When I skimmed that paper yesterday, I also had several qualms with the methodology leading to those conclusions: off the top of my head, they ignored pages with a creation date before 2005; they only looked at edits which change the label (between essay, guideline, policy, etc.); and they only looked talk pages when explicitly mentioned in an edit summary.

To expound a little: if I'm reading it correctly, it would ignore (to list a few non-exhaustive examples), talk discussions not mentioned in the edit summary; discussions which affirm the status quo; changes which align content to community consensus without changing the "label" of a page; and implicit consensus where thousands of people have read a page and no one has found reason to change it. LittlePuppers (talk) 21:56, 30 September 2026 (UTC)

Can I suggest separating, to the extent possible, the funding question from an evaluation of the paper? The WMF does not read the outcome of a project before funding the project, of course, so really it's a question of "should the WMF invest in research about Wikipedia governance" not "should the WMF invest in this paper". Yes, "was this worth it" is a fine question to ask, and one that the WMF answers internally, too, but any evaluation of the funding question should be based not on "I found one I'm skeptical of" but on a full accounting of related projects. I haven't engaged deeply with this paper yet (open in a tab in my "to read" window for some time, sadly), but the project sounds worth doing to me. We have precious few studies that examine Wikipedia governance anywhere but the English Wikipedia, and a direct comparison between several language editions is very potentially valuable. And NM&S is easily one of the top journals for non-computational Wikipedia research these days. That doesn't mean everything it publishes is perfect, but it has a high rejection rate, intense peer review process, and deep bench of Wiki-knowledgeable reviewers. That said, as with any Wikipedia research, it's also valuable when volunteers evaluate the content and correct the record, if appropriate. A quick search through the signpost archives doesn't turn up a match, so maybe it's a good fit for the next Research Report. — Rhododendrites ^talk \\ 16:16, 30 September 2026 (UTC)

I couldn't find an easy way to view multiple grants to the same researcher, but The WMF does read the outcome of a project before funding the next project by the same researcher. --Guy Macon (talk) 10:28, 6 October 2026 (UTC)

Wikimedia Foundation Bulletin 2026 Issue 18

Here is a quick overview of highlights from the Wikimedia Foundation since our last issue on 15 September. Please help translate.

---

Starter Kit dashboard overview.

Highlights

• Introducing the Wikipedia Starter Kit: The Foundation has launched a new Toolforge-hosted tool giving new and small language Wikipedia communities a guided pathway through essential onboarding tasks, activity indicators, and connections to the wider movement.
• Apply for a scholarship to Wikimania 2027: Scholarship applications are now open to attend Wikimania 2027 in Santiago. To help with preparing submissions, orientation sessions in Spanish, Portuguese, and English will be held starting the week of October 5. The final deadline is October 31.
• Reading Lists rolling out to all Wikipedias: Starting September 28, logged-in users on every Wikipedia will be able to save articles to a "read later" list using the bookmark icon, following successful rollouts on Arabic, Bengali, Chinese, Czech, English, French, Indonesian, and Vietnamese Wikipedias.
• WikiCelebrate: WikiCelebrating Juan "BugWarp" Kulichevsky, who has contributed to Wikimedia projects since childhood and is passionate about sports photography and community building in Argentina and beyond.
• The 2026 appointment cycle is open now through 31 October 2026 for the Ombuds Commission (OC) and the Case Review Committee (CRC). You can learn more about the committees, the roles, and how to apply on the Committee Appointments page on Meta-wiki.

Annual Goals Progress on Engage
See also: Growth · Product Safety and Integrity · Tech News · Language and Internationalization · The Wikipedia Library · list of movement events · Wikifunctions & Abstract Wikipedia

• Community Wishlist 2027: The Community Wishlist Process is now updated, taking into account the feedback received. Look out for the call for wish submissions towards the end of October.
• Wikimedia CEE Meeting 2026: From Sep 18-20, Wikimedians, open-knowledge researchers, and free-culture advocates from Central and Eastern Europe gathered for the first time in Romania. Take a look at the novelties and history behind CEE Meeting.
• Increasing account creation: To convert more casual users into logged-in daily users, the evergreen account creation prompt is released into Beta / TestFlight for iOS and Android. Also, a new post-edit notice designed to encourage Temporary Account holders to register for a permanent account will now be released to all wikis. The new notice has proven to increase permanent account creation from 1.94% to 3.66%.
• VisualEditor is now the default at English Wikipedia: this wiki has now adopted VisualEditor as the default editor on both desktop and mobile web, following a sitewide discussion that found overwhelming consensus to make the change.
• Retesting image carousel: The image carousel experiment will be retested using three new versions of the design, updated based on community feedback. The team will assess results and determine with communities whether or not to proceed with the feature.
• Wikifunctions: Your code can now call back to the Wikidata fetch functions. Wikifunctions have deployed an initial capability by which a code Implementation (in JavaScript or Python) can call certain other Wikifunctions Functions, get back the result, and have it available for further processing.
• Latest experiments: A current experiment is testing whether offering a "generic guidance" path for Article Guidance that does not depend on a Wikidata match would increase the chances of more article creation attempts. See all live, upcoming, and completed experiments in Product & Technology.
• Tech News: The latest highlights from Tech News weeks 38 and 39 include a Reader Growth experiment testing whether showing only part of an article at first on mobile, with a “Read more” button to reveal the rest, encourages people to keep reading. See also the 56 community-submitted tasks that were resolved over the last two weeks.
• Suggestions for Transliterated Search: If you regularly use two different writing systems (such as Hindi and Latin alphabet), sometimes you accidentally type with the wrong one and need to switch to the right keyboard. Now, this gadget will be able to suggest query based on what you typed.
• Peer-to-peer learning program: Let’s Connect, a peer-to-peer learning program is shifting to a community-organized model.
• LLM Paste Check rolling out to Wikipedia: A new Edit Check will prompt editors who paste text likely copied from an external AI chatbot to consider whether it aligns with movement AI-use policies and decide whether to keep or remove it. It goes live as a default-on feature at English Wikipedia on Sep 24 and at all Wikipedias on Oct 8.

Annual Goals Progress on Enable
See also: Research newsletter · WikiLearn News · other newsletters on MediaWiki.org

• Wikidata bulk downloads now free via Wikimedia Enterprise: New beta API endpoints let anyone download full Wikidata snapshots and pull hourly batch diffs at no cost, opening up bulk access that previously required a paid tier.
• Accountability for Wikimedia production services: To ensure clear accountability for responding to incidents in Wikimedia production services, Wikimedia Foundation published the Service Catalog page. The page is the canonical source of service ownership for code deployed to the Wikimedia Foundation cluster. See also FAQ pages which contain more background details.
• Datacenter switchover 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.
• Affiliates and grant funding: The conversations on the new proposed model, requirements and criteria continues until the end of next week with public calls on 1) Tuesday, September 29, 16:00-17:00 UTC on affiliate growth paths and Friday, October 2nd, 17:00-18:00 UTC on affiliate overlaps.

Annual Goals Progress on Reach
See also: Wikimedia Apps · Readers

• AI literacy program: A new regional initiative in ASEAN was launched to support training and conversations around AI literacy, accountability, and Indigenous self-determination.
• Testing Google preferred sources: An experiment is proposed on English Wikipedia to run a temporary (2-week) notice asking readers who come from Google to set Wikipedia as a “preferred source” in Google. We would see if this causes people who use Google to visit Wikipedia more. If your language wiki is interested, please reach out on the project's talk page.

Other Movement-curated newsletters & news
Diff blog · Goings-on · Planet Wikimedia · Signpost (en) · The Headword (en) · Kurier (de) · Actualités du Wiktionnaire (fr) · Regards sur l'actualité de la Wikimedia (fr) · Wikimag (fr) · Education · GLAM · Milestones · Wikidata · Central and Eastern Europe · other newsletters

Subscribe or unsubscribe · Help translate

For information about the Bulletin and to read previous editions, see the project page on Meta-Wiki. If you have feedback or suggestions about the bulletin, let us know at foundationbulletin@wikimedia.org. For questions about the Wikimedia Foundation's work, contact us!

MediaWiki message delivery 01:50, 30 September 2026 (UTC)

OpenAI rogue agent activities found on Wikimedia projects

The Wikimedia Foundation article was the #2 top read article yesterday behind only Christa Pike with over quarter of a million views. This is unusual so I looked into what was causing the spike in interest in the WMF.

It seems to be due to a report by Selena Deckelmann that OpenAI “rogue” agent activities found on Wikimedia projects.

The Wikimedia Foundation conducted its own investigation to see whether Wikimedia websites had been similarly affected by AI agents, focusing on those operated by OpenAI. We can confirm that we have discovered some activity by these “rogue” OpenAI agents on Wikimedia platforms. The unauthorized bot activities included edits to our wikis, some unsuccessful attempts to exploit a public note-taking tool we host, and heavy traffic, which are described more below.

This was picked up by Reuters and the resulting coverage presumably generated increased interest in the WMF.

Andrew🐉(talk) 06:41, 6 October 2026 (UTC)

This is the dataset of all the edits. Per that (and discussion on Discord), the citation tool in question is m:Web2Cit. Sohom (talk) 14:03, 6 October 2026 (UTC)

Here's that dataset in list form for convenience:

Extended content

• https://en.wikipedia.org/w/index.php?title=User:Example/sandbox&diff=prev&oldid=1353490694
• https://en.wikipedia.org/w/index.php?title=User:Example/sandbox&diff=prev&oldid=1353490935
• https://en.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=1353491551
• https://en.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=1353492663
• https://en.wikipedia.org/w/index.php?title=User:Example/sandbox&diff=prev&oldid=1353498400
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=741398
• https://test2.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=612931
• https://www.mediawiki.org/w/index.php?title=Project:Sandbox&diff=prev&oldid=8370989
• https://en.wikipedia.org/w/index.php?title=User:Sandbox&diff=prev&oldid=1353501383
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=741399
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=741400
• https://test2.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=612932
• https://test2.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=612933
• https://www.mediawiki.org/w/index.php?title=Project:Sandbox&diff=prev&oldid=8370994
• https://www.mediawiki.org/w/index.php?title=Project:Sandbox&diff=prev&oldid=8370995
• https://www.mediawiki.org/w/index.php?title=Project:Sandbox&diff=prev&oldid=8370996
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=741405
• https://test.wikipedia.org/w/index.php?title=User_talk:Example/sandbox&oldid=741406
• https://en.wikipedia.org/w/index.php?title=User:Example/sandbox&diff=prev&oldid=1353507315
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=741409
• https://en.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=1353518652
• https://en.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=1353543060
• https://commons.wikimedia.org/w/index.php?title=Commons:Sandbox&diff=prev&oldid=1213503714
• https://commons.wikimedia.org/w/index.php?title=Commons:Sandbox&diff=prev&oldid=1213503789
• https://commons.wikimedia.org/w/index.php?diff=1213513506
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=744268
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=744270
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=744271
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=744272
• https://en.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=1356314507
• https://test2.wikipedia.org/w/index.php?title=Sandbox&diff=prev&oldid=613856
• https://test.wikipedia.org/w/index.php?title=Sandbox&diff=prev&oldid=744412
• https://test.wikipedia.org/w/index.php?title=Sandbox&diff=prev&oldid=744414
• https://en.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=1356419247
• https://test.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=747327
• https://simple.wikipedia.org/w/index.php?title=Wikipedia:Sandbox&diff=prev&oldid=10891416
• https://commons.wikimedia.org/w/index.php?title=Commons:Sandbox&diff=prev&oldid=1233683454
• https://incubator.wikimedia.org/w/index.php?title=Incubator:Sandbox&diff=prev&oldid=7226111
• https://meta.wikimedia.org/w/index.php?title=User:~2026-36867-71/Web2Cit/data/com/arcgis/geocode/templates-temp-5123&oldid=30732691
• https://meta.wikimedia.org/w/index.php?title=Web2Cit/data/com/arcgis/use1-geocode/templates.json&oldid=30732696
• https://meta.wikimedia.org/w/index.php?title=Web2Cit/data/gov/hawaii/geodata/templates.json&oldid=30732698
• https://meta.wikimedia.org/w/index.php?title=Web2Cit/data/com/arcgis/templates.json&oldid=30732699
• https://meta.wikimedia.org/w/index.php?title=Web2Cit/data/com/arcgis/services/templates.json&oldid=30732700
• https://bg.wikipedia.org/w/index.php?diff=12923296
• https://commons.wikimedia.org/w/index.php?diff=1213399057
• https://meta.wikimedia.org/w/index.php?title=Meta:Sandbox&diff=prev&oldid=30732655
• https://commons.wikimedia.org/w/index.php?title=Commons:Sandbox&diff=prev&oldid=1238390511
• https://incubator.wikimedia.org/w/index.php?title=Incubator:Sandbox&diff=prev&oldid=7226103
• https://incubator.wikimedia.org/w/index.php?title=Incubator:Sandbox&diff=prev&oldid=7226104
• https://incubator.wikimedia.org/w/index.php?title=Incubator:Sandbox&diff=prev&oldid=7226105
• https://incubator.wikimedia.org/w/index.php?title=Incubator:Sandbox&diff=prev&oldid=7226107
• https://incubator.wikimedia.org/w/index.php?title=Incubator:Sandbox&diff=prev&oldid=7226108
• https://incubator.wikimedia.org/w/index.php?title=Incubator:Sandbox&diff=prev&oldid=7226109
• https://incubator.wikimedia.org/w/index.php?title=Incubator:Sandbox&diff=prev&oldid=7226110

--TonySt  14:32, 6 October 2026 (UTC)

OpenAI is getting sued for the OpenAI–HuggingFace incident. WMF should sue them for this. Levivich (talk) 14:23, 6 October 2026 (UTC)

On what grounds? None of these edits seem to have caused any meaningful harm to the WMF. * Pppery * _{it has begun...} 16:14, 6 October 2026 (UTC)

(This is not legal advice. I am not a lawyer for the WMF or anyone else reading this.) Of course they caused harm: it cost the WMF money to pay people to investigate and respond to the breaches. Arguably there's a harm in overloading the servers via heavy traffic. And an attempt to cause harm can sometimes be legally actionable even if no actual harm was caused. ("No harm no foul" isn't always the law; some laws provide statutory damages.) Also, you don't have to sue for money, you can also sue for an injunction--like suing a neighbor for an injunction to keep their dog on a leash, one could sue an AI company for an injunction to keep their bots on a leash. As for grounds, off the top of my head: breach of contract for violation of the TOS, Computer Fraud and Abuse Act, Stored Communications Act, common law claims like trespass to chattels, and possibly state laws like California Comprehensive Computer Data Access and Fraud Act and the Florida Computer Abuse and Data Recovery Act. Levivich (talk) 18:43, 6 October 2026 (UTC)

There were agents on our wiki... and my monitoring script didn't catch it... screw my stupid chud life. ^MetalBreaks_AndBends (One for all) 14:33, 6 October 2026 (UTC)

Popups!

The donation pop ups have now been going on for over a year straight. This has never happened before, Wikipedia used to do it only a few days a month or year. There are also the "install the app" pop ups. Please consider reverting your policy. ~2026-53762-96 (talk) 09:26, 6 October 2026 (UTC)