Template talk:AI-generated

Encourage editors to delete or to fix?

My proposed change. Seems to me that best practice would be to encourage editors to delete AI generated text when found. It is too time-consuming to fix it when both the prose and the sources are usually fictitious. To me, AI generated text has a similar flavor as copyright and hoaxes, the best practice for which is also deletion.

Regarding the argument that we shouldn't encourage deletion via a maintenance tag, I think it'd be fine, because 1) a maintenance tag can probably get consensus quicker than a new CSD criteria, 2) there may be some exceptions/edge cases, 3) tagging before deleting can help with crowdsourcing (for example, the tagger wants a second opinion or is too busy or too unsure to execute the deletion themselves), 4) maybe just a section is AI generated and not the whole article. Edit to clarify: The tag would encourage someone else to use a deletion process or to delete the problematic section, and also allow room for human judgment and edge cases.

Thoughts? –Novem Linguae (talk) 21:05, 27 January 2023 (UTC)

I think what @Thryduulf wrote at the CSD discussion is right. If an article is in mainspace we should encourage it to be nominated for deletion for all the reasons you say Novem. If it's in draft or userspace, we should give the cleanup options. Best, Barkeep49 (talk) 21:07, 27 January 2023 (UTC)

Deletion discussions are the absolute best way of getting a new CSD criterion - without them there is no evidence of frequency or evidence of a consensus that they should all (or some subset of them) should always deleted. They also aid in crafting the criterion because more data makes it easier to objectively specify what separates things that get deleted from things that don't. Thryduulf (talk) 22:01, 27 January 2023 (UTC)

Different Icon?

While the current icon is a robot, it doesn't seem to convey the idea that it's from a robot "conversation". How about — rsjaffe 🗣️ 22:35, 27 January 2023 (UTC)

The current icon looks far better DFlhb (talk) 00:58, 28 January 2023 (UTC)

@CactiStaccingCrane changed it to even a better one, so I withdraw this suggestion. — rsjaffe 🗣️ 03:11, 28 January 2023 (UTC)

My buddy is a real MJ whiz, so I have asked him if he can come up with something. jp×g 03:11, 28 January 2023 (UTC)

Be interested in seeing it. — rsjaffe 🗣️ 03:12, 28 January 2023 (UTC)

Renaming category

The category should be renamed from "Articles containing suspected AI-generated texts" to "Articles containing non-compliant AI-generated texts", or perhaps "suspected non-compliant", to make it clear that this is a cleanup category rather than a general category for all articles where LLMs were used (though I suppose there is yet no affirmative consensus on whether LLM text generation is allowed or forbidden). DFlhb (talk) 17:14, 31 January 2023 (UTC)

Hidden text from within documentation.

No idea why this was put in the documentation itself as hidden text. Seems better to put it here where people can see it. — Trey Maturin™ 13:14, 23 May 2023 (UTC)

Previously hidden text

Some trials on articles I've /User:JPxG/ written (using the lead paragraphs): Powder House Island and Nina Jankowicz are estimated at 99% human-written, but First Wikipedia edit is at 20% GPT for some reason. 1-Pentadecanol returns 40% GPT based on the first sentence, which decreases with subsequent sentences to 99%. However, when I asked ChatGPT to "write me a Wikipedia article about 1-pentadecanol", the result was this:

1-Pentadecanol, also known as n-pentadecanol or pentadecyl alcohol, is a straight-chain saturated primary alcohol with the chemical formula CH3(CH2)14OH. It is a white, waxy solid that is insoluble in water, but is miscible with most organic solvents.

1-Pentadecanol is found in many natural sources, including vegetable oils and animal fats. It is also used as a raw material in the manufacture of lubricants, candles, and other products.

This was estimated as 92% human-written. I don't know exactly what the underlying mechanism of this tool is, but we may want to take its output with a grain of salt.

When encountering a newly created page, given that the prose seems reasonably competently written (and, typically, at least several paragraphs long), these are some very mild indicators that undeclared LLM use is involved, which could mean that checking with a machine identifier could be worthwhile:

• The creator has words like "AI" or "neural" (and similar terms indicating an interest in LLMs or, more broadly, deep learning) in their username
• The content concerns a fictitious subject; sometimes the title will just be fairly nonsensical, or will render as poor English, yet the content (not having any grammatical or orthographic errors) seems surprisingly coherent on surface, as if the "author" had a good idea what they were writing about; due to their incredible language-manipulating capacity, LLMs far surpass humans at stringing plausible sentences about nonsense
• There are fictitious references which otherwise look persuasive (see these examples)
• The references are not fictitious, and there are inline citations, but the text–source integrity is abysmal (indicating a facile after-the-fact effort by the creator to evade suspicion by inserting citations into raw LLM output, without making the needed adjustments to establish genuine verifiability, which could otherwise be a quite painstaking process of correcting, rearranging, and copyediting)
  or
  there simply aren't any references (meaning that the model might have generated some but they were manually removed because others would notice that they are junk)
  • one wonders how could someone seemingly familiar with Wikipedia content standards enough to be capable of writing a decent-seeming chunk of article prose be so incompetent at ensuring minimal verifiability at the same time
• There are references in the style of what is outputed by Wikipedia's usual citation templates, but there are no inline citations
• The content looks as if copied from somewhere due to not having wiki markup (wikilinks, templates, etc.)
  • one wonders how could someone seemingly familiar with Wikipedia content standards enough to be capable of writing a decent-seeming chunk of article prose miss adding at least a bare amount of wikilinks
• The article obviously serves to promote an entity (such as by giving it visibility) but the prose seems very carefully tweaked to look objective
  • one wonders how could someone so unfamiliar with Wikipedia content standards that they would attempt to publish a promotional page be so skilled at crafting exceedingly neutral verbiage (creators of promotional articles and drafts are typically incapable of completely eschewing promotional language)
• The last paragraph is oddly out of place since the text ends with a conclusion of sorts, encapsulating some earlier points; it may start with or contain a phrase such as "In conclusion", "This article has examined" or similar. Such structures and phrases may be extremely prevalent in LLMs' corpora, so they can't shake the habit off even when told to "write a Wikipedia article", despite the fact that Wikipedia articles do not have this characteristic.

A few more things to note:

1. It's surprisingly easy to fool the detector through minor edits to the GPT output. The detector is also pretty much useless for non-prose text.
2. From my own experimentation I've found that machine-translated content, regardless of whether written by human or GPT, tends to yield "unclear" on the detector, which I assume is probably an intentional foresight to prevent obfuscation of AI output using machine translators.
3. GPT-4 is now a thing (albeit something you either have to buy yourself for $20 or acquire through Bing's waitlist), and since OpenAI's own detector is designed for GPT-3, GPT-4 output fools it, at least for the time being. I'm pretty sure I've heard of GPT-4 having baked-in flag tokens in order to make future detection easier, though.
4. You'd have to get on a waitlist to actually access either, but it's now possible through both Bing and ChatGPT (both GPT-4) to browse the web, allowing for "legitimate" citations (although even with those present it's still very possible for the AI to hallucinate anyways).
5. In addition to "in conclusion...", some other common dead giveaways include "as X, ...", "it is important...", and "firstly, secondly, thirdly..." (especially in a Wikipedia context). These are ubiquitous on GPT-3, but can also be found on GPT-4. However, again, it's very easy for people to realize this and revise the output to obfuscate these...

WiktionariThrowaway (talk) 22:59, 20 March 2023 (UTC)

OpenAI has retracted their own detection model. — Frostly (talk) 23:46, 9 January 2024 (UTC)

Revert

Feel free to revert my changes, but I've reverted the template back to diff 1158367048 as it maintains consistency with other maintenance templates and already includes the changes previously made to the template. Dawnbails (talk) 14:15, 11 June 2023 (UTC)

Cont.

Moved from User talk:Alalch E.

 – —Alalch E. 14:48, 11 June 2023 (UTC)

I'll copy what I said on Template_talk:AI-generated here:

Feel free to revert my changes, but I've reverted the template back to diff 1158367048 as it maintains consistency with other maintenance templates and already includes the changes previously made to the template.

I think it matters that maintenance templates should probably maintain relative consistency to each other, and I don't see how the revisions after it improve much other than, for no particular reason, moving text down and linking the same page on it twice.

I also can't seem to find the discussion you speak of. I'd appreciate a link to it. Dawnbails (talk) 14:36, 11 June 2023 (UTC)

@Dawnbails: The difference between revisions is substantive. It's about what to recommend: removal or "cleanup by throughly verifying the information and ensuring that it does not contain any copyright violations". The discussion is this one: WT:LLM#Non-compliant LLM text: remove or tag?. The core message of the template is what's important, not mere form as in whether is resembles other maintenance templates.—Alalch E. 14:48, 11 June 2023 (UTC)

I see. I did read the discussion earlier, but I seemed to have missed the change on the template related to cleanup— that's my bad. Cheers. Dawnbails (talk) 14:52, 11 June 2023 (UTC)

Name of model

Is it really necessary to explicitly mention ChatGPT? Whilst I understand this template is somewhat of an anomaly due to the emerging field I feel that naming a certain brand in a maintenance template is inappropriate. – Isochrone (T) 20:39, 12 September 2023 (UTC)

I think the idea is that LLM is too jargony for folks to understand. I agree with this idea and wouldn't mind keeping an explicit mention of ChatGPT for now. –Novem Linguae (talk) 18:37, 13 September 2023 (UTC)

Looks like all mentions of GPT/ChatGPT were recently removed by InfiniteNexus. I am still concerned that LLM is too jargony for most folks to understand. –Novem Linguae (talk) 21:59, 16 February 2024 (UTC)

That's what the link to the article is there for. InfiniteNexus (talk) 02:41, 17 February 2024 (UTC)

I would be open to adding "AI" or "artificial intelligence" in there for clarification. InfiniteNexus (talk) 02:43, 17 February 2024 (UTC)

Warning

Is there any warning template for users who are using AI to create their articles? RodRabelo7 (talk) 00:14, 24 June 2024 (UTC)

@RodRabelo7. {{Uw-ai1}}, {{Uw-ai2}}, {{Uw-ai3}}. –Novem Linguae (talk) 03:26, 24 June 2024 (UTC)

Resolving and removing

What are the standards for resolving this template and removing it? Most templates have some specific issue that can be edited away, but that doesn't really apply here. The closest comparison I can think of is Template:COI, where you're expected to identify other problems associated with the COI and the tag can be removed once they've been solved. There are also templates like Template:Merge to or Template:POV where you're expected to engage on the talk page and the tag is removed when the discussion or dispute is resolved. Does AI-generated broadly apply in one of these ways, and can something to the documentation be added about resolving it? Thebiguglyalien (talk) 🛸 03:18, 20 July 2025 (UTC)

I'm also wondering this, especially since I am adding this tag to a lot of articles (and someone has complained about it).

In my opinion, the threshold would be when every single word and fact has been thoroughly reviewed by someone other than the author, and all issues with hallucination, tone, etc. have been resolved.

This probably would mean the templates will probably remain on articles for a long time. Not only am I fine with that, I think it's a good thing. Readers deserve to know whether they are reading something with still-extant, still-unreviewed AI-generated text; we should not be withholding that information from them. Editors also deserve to know this, because collectively we have not been very good at finding this stuff on our own. Plus, having current AI text in one place makes it easier to reverse engineer new signs of AI writing that are probably present in other articles. Gnomingstuff (talk) 21:16, 13 August 2025 (UTC)

As with all tags, I think it can be removed once the issue is undetectable / solved. For AI, this probably involves rewriting all parts that are obviously AI-generated. This is laborious, so WP:TNT may be a good option. –Novem Linguae (talk) 08:44, 14 August 2025 (UTC)

I think we should have a bias towards removing. Gnomingstuff's rewriting threshold is fine too where possible, but it's undesirable for this template to remain for long periods, and the only way to avoid that is a bias in favor of removal.

The template discredits the whole article, when only part of it is faulty. Average readers won't check the history to see which parts are at issue. Second, if the template is used long-term and becomes commonplace enough, it risks reader confusion about the acceptability of AI on Wikipedia. I can easily imagine a viral post screenshotting this and commenting "sigh, even Wikipedia has fallen to AI", while others (who aren't anti-AI) might interpret it similarly and take it as permission to contribute AI themselves.** It harms Wikipedia's credibility. Trust is more fundamental than disclosure. If we let suspected AI text remain in articles, many readers will see that in itself as a breach of their trust, and see our decision to disclose as immaterial. AI disclosure is not the end-all-be-all.

**Separately, the template text contributes to this, since it gives no indication that the LLM additions themselves are undesired, nor instructions specific to those additions (only about copyright/verification). It could be interpreted as an AI disclosure template, not an "issue" template. But even if fixed, I think my point still stands.

I know removal feels harsh when we might have mere suspicions, but I think that's the right tradeoff. Maybe suggest that editors link the suspicious revision on the talk page, so that someone can scrutinize it and only then reinsert it. DFlhb (talk) 23:28, 28 August 2025 (UTC)

I can easily imagine a viral post screenshotting this and commenting "sigh, even Wikipedia has fallen to AI"

This has happened already, dozens of times, in relation to other events: Tone Check, Simple Summaries, the Jimbo thing, the WMF AI strategy, some others I'm missing. As I said, it is 100000% better for readers to be (correctly) disappointed that an article contains AI text, than for readers to be unaware and end up trusting hallucinations. I can just as easily imagine a viral post equivalent to "Google's AI summary told me to make glue pizza. Think of all the people (correctly!) freaking out about vandalism back in the day.

Unfortunately, removing the LLM additions immediately isn't always so simple, especially when there have been hundreds of edits since then disguising or building on various parts of it. Usually it requires subject-matter expertise because LLM text is designed to look plausible to a layperson. That's why I've been tagging so many articles -- we have a rare chance to actually catch this before it's been around for 10-15 years, the deadline is now, etc. Gnomingstuff (talk) 03:33, 30 August 2025 (UTC)

I agree with user:Gnomingstuff and user:Novem Linguae here.

Especially as we go forward and potentially become infected with long-standing undetected material in articles, whenever an article gets this tag, it should mean a thorough scrub of all unverified and poorly-sourced material, either by verifying them properly or nuking them per WP:TNT, WP:G15, etc. When it's done out of anything but the most naive good faith, I would compare this to a sort of extremely malicious vandalism, having all the verbiage and form of a "real" edit when it's actually a randomly generated sequence of autocompleted pseudofactual material. There's always been threats to Wikipedia, and it's survived them all, but this is a new one and should be taken seriously (just as vandalism became as such historically).

Since I largely agree with those present (and therefore sense consensus forming) I have added a section in the docs about when to resolve and remove this template based on what's been said so far. Altoids0 (talk) 20:54, 31 August 2025 (UTC)

An even better analogy than vandalism... I think AI-generated content shares a lot of properties with WP:CCI. For example, the writer does it in good faith, it is easy for the writer to do, and it is extremely hard for others to clean up. I think vandalism is usually easy to clean up compared to AI. –Novem Linguae (talk) 08:02, 1 September 2025 (UTC)

Yeah -- vandalism tends to be small localized edits, this stuff is not often localized and tends to be a huge amount of shit getting slightly iterated upon. For instance, something like this user's edits -- that's going to be a lot to slog through, and maybe people looked at the text 2 years ago but no one seems to have picked up on the underlying issue, and enough of the original text remains that I found it through one of my AI tell searches. Gnomingstuff (talk) 17:16, 1 September 2025 (UTC)

I think you're right, especially if the generated material gets massaged enough over time to lose its tells and join the blob of unsourced claims on Wikipedia.

That said, I feel the problem is a little more approachable than CCI's. From my personal experience of cleaning up dozens of contaminated articles, it remains something you can personally detect, especially since on Wikipedia there's more hints than normal (generated text often lacks wikilinks or templates, uses really terse citations to closed-access literature, includes lazily-chosen images with vague captions, etc etc). Anecdotally I would compare it as well to one eternal sockpuppet chase where you slowly learn every single mannerism of this one malicious writer with a billion IPs.

Also, it's likely that the detection of this material is a technological problem that will soon be largely solved, in the way trivial vandalism is solved on Wikipedia. GAN- and LDM-generated images as of recently can be detected with almost 100% accuracy, and the same will presumably soon be true (or nearly so) for LLM text. So I guess I mean to say that it will evolve into a vandalism-like problem as detection tools improve.

Not a huge deal either way what it's like, though. Ultimately I pinged you two just in case you had any opinions on my addition to the documentation. Altoids0 (talk) 02:33, 2 September 2025 (UTC)

I generally agree with the sentiment in this discussion, but I find the current wording in the recommendation a bit too strict.

Some context: I've had some of my articles slapped with this template. I had indeed used LLM to write the article, but I had checked the content of the sentences, and that it matches the sources I used. However, given that no rationale was provided for "which sentence feels AI generated", I didn't really know what to do with the template.

I think people should be allowed to challenge the existence of the template on the talk page if they feel it's not applicable, and remove it after a few days if no response/additional rationale for it is provided. In such cases, the "burden of the proof" would get reverted: i.e., if someone contests the template, it should be the responsibility of the person who wants to have the template stick to provide 1-2 examples of sentences that are suspicious, rather than the author/readers having to prove that the whole article is, in fact, not having quality issues linked to LLMs. 7804j (talk) 15:15, 2 September 2025 (UTC)

The exact wording in this template's documentation is that a talk page notice ought to be provided when the tag is added, so I think what you want is already present, unless I'm mistaken. I ended up emboldening that statement in the edit discussed here, since I've personally found it very helpful and productivity-improving that Gnomingstuff provides one for all of the articles they label.

In part because I'm personally quite an advocate for WP:LLMDISCLOSE, and also because it seems relevant to this discussion, I would be very interested to see a list of the articles you generated (or presumably partially generated). Altoids0 (talk) 19:12, 3 September 2025 (UTC)

Upon second read, I agree that the current text is good, so I withdraw my previous comment. What made me feel uncomfortable at first was the term "demonstrably non-existent", which I had interpreted as putting the burden of "proof of absence of issue" on the person removing the template, but I may be overreading from it, and you are right that it doesn't prevent people from just creating a discussion in the talk page in any case. 7804j (talk) 17:55, 4 September 2025 (UTC)

(Just to clarify, I haven't been doing talk page comments for all of them, only the ones that aren't self-explanatory or took time to dig out of edit history. I should probably be more strict about what's "self-explanatory," it's just a huge pain in the ass when I go to someone's contribs and they have hundreds or thousands of edits, then the next guy also has hundreds of thousands of edits...)

The big problem -- and I don't know if it's even solvable -- is that there's no good way to tell how much review someone did of their AI writing. Outside of the obvious proof like hallucinations or made-up sources (basically, G15 eligible) it's a gray area, and a lot of the original editors aren't around anymore to ask. Since I'm finding this stuff through wiki search -- and only tagging the ones I can back up, I've actually left a lot of articles I've checked alone -- I don't find it unless there's a lot of questionable AI stuff still remaining in the article or the editor's other similar additions. But I have no idea whether they didn't review it at all or just didn't review it enough. The reason I don't list suspicious phrases is so people don't just do a quick rewrite and call it a day when there might still be hallucinations, copyvios, or the like remaining. Gnomingstuff (talk) 21:42, 3 September 2025 (UTC)

• the threshold would be when every single word and fact has been thoroughly reviewed by someone other than the author, and all issues with hallucination, tone, etc. have been resolved.
• once the issue is undetectable / solved. For AI, this probably involves rewriting all parts that are obviously AI-generated.

Is it just me or don't you realize you're walking into a slopified Wikipedia with your eyes open? Noone will ever manually fix AI generated content. Computers can write stuff a million times faster than every human on the planet combined can review the text, let alone rewrite any issues. Am I losing my mind or why don't you see that this approach will quickly lead to a Wikipedia where almost every article has this tag?? CapnZapp (talk) 14:16, 18 September 2025 (UTC)

There's a lot to be said in response to this sentiment (the stunning advancements in LLM detection, the history of vandalism bots (and doom towards vandalism) on Wikipedia, etc.), which I think in general would be quite worthy of a future essay.

In the context of this talk page, though, there's a simple answer, and despite the age of your account I still think you may need to hear it: if you think the mission of Wikipedia is impossible, you are free to stop contributing. Unless you are providing expert commentary as a person researching generated content detection, saying that a bad thing is hopeless does not contribute to the encyclopedia. Altoids0 (talk) 08:10, 21 September 2025 (UTC)

References

Suggestion: Add a diff(s) parameter

A lot of AI-generated edits are buried deep in page history and/or polished by later unsuspecting editors, which makes them less immediately obvious for someone reading the current version. Some of them are fairly small in size/scope compared to the full article, but the only way we have to indicate that is the section parameter, and they're not always confined to one section.

It would be helpful to add a parameter for the edit diff that introduced the suspected AI-generated text and show that on the template -- even if you mention why you added the template on the talk page, people don't always read that, and non-editors almost definitely aren't going to. Gnomingstuff (talk) 21:11, 3 September 2025 (UTC)

why do we have this template?

Does the existence of this template only encourage editors to add content of poor quality, or what?

Why don't we treat AI generated instances of editorializing, subtle factual errors, unattributed opinion, non-neutral language and so on like any other such additions?

That is, we ask the editor making poor contributions to stop or we block' em. The content itself can either be improved (if salvageable) or simply reverted, asking the editor to try again, this time making an actual effort.

What purpose does this template serve other than warning users of general and poorly defined "poor quality"? And by extension, lowering our standards until we can tag the entire encyclopedia "AI-generated"?

I fail to see the value in accepting submissions that are of such poor quality that we need a special tag. Either the content is "good enough" or it should be rejected. Who wrote it doesn't matter; who submitted it (or whose bot submitted it) is the key here. Any time you feel compelled to use this tag, how about just reverting instead. If the editor adding it really is using an LLM they really need to step up their game, so the content at least matches good-faith but mediocre-quality human additions, or we don't want their slop anyway.

CapnZapp (talk) 14:08, 18 September 2025 (UTC)

WP:MAINTENANCETAGS are a type of crowd sourcing that lets editor 1 spot a problem, and editor 2 do the work of cleaning it up. Without maintenance tags, editor 1 would have to do both the spotting and clean up, which is okay sometimes, but can be more effort in the long run.

Cleanup for AI often is deletion (csd, afd, tnt).

That is, we ask the editor making poor contributions to stop or we block' em. This is done too. See for example {{Uw-ai1}} through {{Uw-ai4}}.

Who wrote it doesn't matter; who submitted it (or whose bot submitted it) is the key here. In mainspace, aren't writing and submitting the same thing? –Novem Linguae (talk) 17:23, 18 September 2025 (UTC)

I need to ask you to not give me elementary explanations of things any experienced Wikipedian would already know, Novem Linguae. Yes, I know what cleanup tags are for. No, I don't think this tag can or should be considered a general and uncontroversial cleanup tag. What specifically makes someone add this tag? If it is subtle factual errors, just revert them. If it is non-neutral language, we have a tag for that. And so on. It just feels like this tag becomes a catch-all that lets people tag additions of poor quality without having to specify what's wrong. And by "tag" I really mean "keep". How about we instead of adding this tag either apply specific cleanup tags or simply revert when we can't identify which one(s) to use, or when "all of them" would be appropriate?

The core issue, to me, is that this tag will eventually (as in "very soon") need to be applied to the entire project if we keep going in the current direction: it basically provides a fig leaf for allowing slop to stay. Yes, I believe keeping this tag in its so-vague-it-is-useless shape will enshittify Wikipedia in record time. Who considers filling up our pages with this tag to be in any way helpful? Who is so naive they think anyone will ever bother to fix something as unfixable as "this page might incorporate text from a large language model." If someone adds content that hallucinates or invents false references, Wikipedia is better off just asking that someone to leave the project. If you or I spend half an hour fixing one tagged article, another hundred will have spawned in the meanwhile. This tag REALLY isn't going to work out. Thank you. CapnZapp (talk) 09:35, 19 September 2025 (UTC)

You're looking at this the wrong way. The AI writing didn't spontaneously emerge out of thin air because a template was added. It was already there. The tags are not enshittifying Wikipedia -- the enshittification already happened. The only difference between a tagged and an untagged article is whether people know about the problem that already exists. If you don't like there being hundreds of tags, sorry, that's what happens when you have almost 3 years' worth of accumulated issues no one addressed at the time.

Asking someone to leave the project is all well and good, but A) we have no AI policy, so they aren't actually breaking any rules, and B) a lot of these people basically already have left the project through inactivity, given that this stuff has been building up since 2023. What are you going to do, track down every undergraduate student whose single articlespace contribution was dropping one wad of AI text into an article in 2024? Gnomingstuff (talk) 04:30, 21 September 2025 (UTC)

First off, please don't start a whole subdiscussion where you make a big deal out of my phrasing re: blocking vandals. Thank you. What does the template do? It is super passive aggressive: "we suspect AI but we're neither going to do anything about it or stop it from happening again". If people who think like you prevail, why not simply add the tag to the entire project instead of attempting to tag individual articles as if there is ANY hope of cleaning them up. What this tag tells people is "since the only penalty for adding poor AI is a cleanup template, I have a green light to spam Wikipedia with my poor AI creations." This tag means we accept our fate, where AI injections will happen at a pace thousandfold (or millionfold) faster than any group of human editors can even tag it. I'm sorry, but this line of thinking will doom the project. CapnZapp (talk) 11:56, 22 September 2025 (UTC)

First, I have no idea what you are talking about with: "making a big deal out my phrasing." I compared the AI issue to undetected vandalism because they are the same basic problem: AI slop existing for years, undetected despite being in plain view.

Second, the reason the whole tag isn't being added to the entire project is that the entire project does not contain AI writing. Slippery slope arguing at its finest.

Third, if you truly think that this is going to doom the project, you are welcome to do literally anything about that. If you think the articles should be cleaned up, go clean some up. If you think there should be a stronger AI policy, go advocate for that. Complaining at the few people who are actually trying to address this problem accomplishes nothing, besides maybe a feeling of superiority. Gnomingstuff (talk) 15:19, 22 September 2025 (UTC)

Your defensiveness is unconstructive, User:Gnomingstuff. How on earth did you conclude I was doing this because I wanted to feel superior?!? I sincerely question the value of continuing this discussion, but I'm going to give it another shot. I am doing something. What I am doing is: I am pointing out to you that the existence of this template is misguided at best and actively harmful to the project at worst. What I am doing is trying to help you understand that slapping this template onto an evergrowing number of articles does absolutely nothing good. If the AI is slop, it should be rejected. If the AI is good (enough), this tag should not be used. End of analysis. All this template accomplishes is 1) prevents editors from adding actually useful cleanup tags and 2) sends the signal adding shitty AI is actually okay, since the penalty is nothing more than a cleanup tag = my slop still gets accepted. Now imagine we instead add an AI parameter to existing, specific, useful cleanup tags - that could be worthwhile. tl;dr: this tag is far far too generic and all-encompassing to do diddly squat. At a minimum it needs to be used much more restrictively, and not just as a stupid "yep, this article contains LLM content" sign that will just uglify the project. Specifically: it should not be used solely because the article incorporates text from a large language model, as the first sentence proclaims. What are the ACTUAL issues with this? Well, because this tag now exists instead of useful specific tags, we might never know. This tag makes the task of cleanup extra hopeless. Please tell me you can see an editor (like me) spotting this tag and then going "Okay, so you've identified that there MIGHT be an issue with the article... but this tag doesn't tell me if there actually ARE issues, what those issues are, and where they might be". Now please tell me you do see why I'm making this effort, and why I'm not in this to hurt your feelings? CapnZapp (talk) 08:30, 23 September 2025 (UTC)

Maintenance tags were useful for the reasons Novem mention. Cleanup also can take time, in cases where there are later edits that might mix in with them. While it would be great if people just left the project or if AI text could be immediately removed, until such nominations/effort stop getting pushback from the community that eg. cause AfDs that drag on before resolution, the tag remains better than an leaving content sitting there with no warning. CMD (talk) 02:35, 21 September 2025 (UTC)

For many reasons:

1) Very little of this stuff has been tagged over the past 2 years, which means very little of it got noticed -- even when the issues are so blatantly obvious that they could have been found in less than a minute had anyone bothered to look. It's the undetected vandalism problem, except at much larger scale.

2) Fact-checking this shit takes time. It takes a lot of time. It also takes subject-matter expertise. I am not a biologist, and I don't know enough about that field to tell what's a hallucination and what isn't. But I do know what AI-generated biology articles sound like.

3) "Just revert it" isn't so simple when someone added the edit in 2023 or 2024, there have been 100+ subsequent edits since then, and the AI text has been commingled with other text without addressing the sourcing issues, puffery, and such. Which is extremely common (especially with WikiEdu edits, which are to longer, higher profile articles).

4) Adding generic cleanup tags disguises the underlying issue. With AI text, you have to check for hallucinations and citations that don't back up the text; cleaning up the phrasing is not enough. There are several cases where someone has dumped in LLM output with hallucinations as well as tone and formatting issues, but since no one identified it as AI text, only the superficial issues got fixed, and the deeper ones went unnoticed.

5) The most important one: If I am reading a Wikipedia page, I would like to know whether I am reading AI output so I can make an informed decision about how skeptical I should be of it. There are many people with a high level of trust in Wikipedia specifically because it "isn't AI." If we know that the text actually is AI, and we don't disclose that fact, we are betraying that trust. Gnomingstuff (talk) 04:23, 21 September 2025 (UTC)

If all editors consistently recognized LLM-generated content additions and promptly removed them, then I can see how this tag would be unnecessary. However, in practice, LLM-generated edits are not always obvious (especially to editors who are not familiar with the signs of AI writing) and many LLM-generated content additions do unfortunately make it to article space. By the time the LLM-generated edits do get noticed, as Gnomingstuff said, they are often no longer the most recent edits and may not be removable through straightforward reverts. Verifying and/or eliminating the LLM-generated content can be even more challenging if the article is in a less accessible topic area or if the cited sources are not available free of charge online. There are so many possible scenarios in which an editor recognizes the presence of LLM-generated content but is not in a good position to immediately remove it themselves; those kind of situations are what this template is intended for.

CapnZapp, I see that you are concerned about how LLM-generated content can negatively impact article quality, but are not yet a member of WikiProject AI Cleanup, so I would like to invite you to join the WikiProject. Because the proliferation of LLM-generated content is a major issue on Wikipedia, we are always looking for more help in identifying and handling LLM misuse. — Newslinger talk 17:11, 21 September 2025 (UTC)

If AI text could be immediately removed, until such nominations/effort stop getting pushback from the community... Sorry but you'll have to get back to me when this becomes true. If we cannot revert bad AI on sight and caution the editors (adding poor AI needs to be viewed as a WP:COMPETENCE deficiency) I see exactly zero hope for the project. Slapping this template on such submissions is SUCH an impotent alternative to actually useful action. I'm sorry, but I cannot commit my time trying to hold back a tide with a suction pipette. If we continue to allow this template to exist, it means editors are encouraged to just say "yep, there's AI slop here too". It actively legitimizes and encourages the addition of not-good-enough AI. Y'all appear to believe the problem can be contained, but mark my words, in a short while, this template will become the most used and least useful cleanup template of them all. CapnZapp (talk) 12:10, 22 September 2025 (UTC)

Let me clarify (as a newcomer to the discussion): I'm assuming the issue is AI that is worse than just mediocre. That is, that this tag is intended for actually poor AI-generated content. I'm making this clarification because LLM-generated edits surely cannot be viewed as problematic in general. If the edit's issues aren't egregious or numerous, then that content is surely more or less indistinguishable from many human editors' efforts, and not really a problem. My point is: if we feel compelled to add this tag, the addition is surely so problematic it would be wise to simply refuse it (and also take steps to stop the editors from adding more content of such poor quality). This, to me, doesn't depend on whether LLMs are involved. CapnZapp (talk) 12:23, 22 September 2025 (UTC)

Either that or we (very clearly) restrict this tag for ONLY those special cases were LLM text is BOTH useful AND useless; and/or so intervowen within other (useful) text it's nontrivial to sort out. Currently, I see nothing to prevent me from using this tag as nothing more than a "I don't like it" tag, and/or an excuse to do nothing actually useful, as for example, making a slop spammer stop. CapnZapp (talk) 12:27, 22 September 2025 (UTC)

Identifying whether llm text is useful and/or useless is often nontrivial. You say "If we cannot revert bad AI on sight...I see exactly zero hope for the project" and then say "LLM-generated edits surely cannot be viewed as problematic in general"; the second statement contributes to preventing the first statement from being acted upon without hassle. CMD (talk) 12:47, 22 September 2025 (UTC)

I agree wholeheartedly. I just see zero value in the idea "we can't do anything about it, but we can at least tag it". Why? Because this tag only tells users "LLM might have been used". It says NOTHING about whether this is a problem or not. All I'm seeing here is "lets help AI allergic readers identify articles to avoid" which is such a short-sighted effort. CapnZapp (talk) 08:36, 23 September 2025 (UTC)

That is similar to all maintenance tags which take time to manage. Having just worked through a hallucination and a few source misunderstandings in an llm-generated article after someone removed the tag, if that's all you're seeing then you've missed the huge number of problems that have been taking up untold hours of time over the past couple of years. CMD (talk) 10:05, 23 September 2025 (UTC)

I don't know whether the problem can be contained. But I do know that the problem will never be contained if no one knows there's a problem, or how bad the problem has already gotten. People -- including you -- tend to talk about the AI-on-Wikipedia issue as if it is something looming over the horizon, yet to come. The reality is that it is already here, and has been getting worse and worse since 2023. This means two things:

1) People clearly do not need encouragement to add AI content to Wikipedia, given that for 2+ years they've been prolifically doing it without our help. (The irony is that people are being encouraged, but not by this template, by Wikimedia itself.)

2) The project has already "legitimized" the AI slop by doing nothing about it. A cleanup template is the first step toward un-legitimizing it. Gnomingstuff (talk) 15:30, 22 September 2025 (UTC)

Sorry no. All this tells me you are some passive-aggressive guerilla movement that has lost hope in the project. All you are accomplishing is plastering this tag all over Wikipedia. Even if you really think this will somehow create a movement for actual working slop prevention, you are misusing cleanup templates. This is nothing more than a shame tag. I can't help but see parallels to another "shame tag" where a prominent user got so fed up over WP:BAREURLS she created a tag to tell readers "this article contains bare URLs" hoping this would make the project change their mind and prohibit those. Such tags are not useful. Not to readers. Not to editors. Let me be clear, so I'll outdent here before I continue:

Cleanup tags are meant to be temporary notices that lead to an effort to fix the problem, not a permanent badge of shame to show that you disagree with an article, or a method of warning readers about an article.
-- taken verbatim from the introduction to Wikipedia:Template index/Cleanup

Sorry guys, but this tag here of yours is clearly intended to warn readers instead of help cleanup editors. I strongly urge you to reconsider. A red flag is the isolated nature of this tag. It isn't well integrated into the overall cleanup toolbox. (The similarities to Template:Cleanup bare URLs are too strong to ignore)

Here are some quick off the cuff suggestions. I don't mean this problem can be solved quickly. I mean I genuinely don't want to just complain w/o even trying to be helpful:

Step 1. Go through existing cleanup templates and add the ability to flag "...because of AI". For example:

• If an instance of editorializing is because an LLM was used, add say a {{tone}} tag with a new |ai=yes parameter so the template explains that the problem here is that the tone issue probably stems from using a LLM.
• If an instance of subtle factual errors is because an LLM was used, no cleanup tag should be used. Actually wrong content should simply not be accepted. If the facts are questionable, but good faith cannot be excluded, we have fact tags, dubious tags and so on.
• If an instance of unattributed opinion is because an LLM was used, we have by whom, specify and so on tags and related puffery or weasel language tags. Again, these could be given |ai= parameters to help identify the "culprit".
• If an instance of non-neutral language is because an LLM was used, again we already have tags for that, and again we could add the ability to specify "because AI".

Step 2. Restrict (and reword) this template to be used specifically and only when the AI bits are all of the following at the same time: 1) useful 2) worthless 3) interwoven with other submissions

• If the AI bits are only worthless (or the errors clearly outweigh the utility), just restore to an earlier version even though this might lose other contributions. Warn or ban the editor. Not for using AI, but for adding misleading content.
• If the AI bits are only useful, why are we even here. Just accept LLM as a new writing style more and more people are going to employ. Focus on the quality.
• If the AI bots aren't interwoven, revert (if problematic enough).

Currently this tag says "This article may incorporate text from a large language model." This is not okay, guys. It says nothing about what the issue is, and it doesn't even say there IS an issue, just that there MIGHT be one.

I am not saying my suggestions must be the best or even good. I am offering them as examples of what you could do to accomplish your goals while avoiding the ineffectual trying-to-undermine-official-Wikipedia-policy path you seem to have chosen.

Cheers and good luck, CapnZapp (talk) 09:01, 23 September 2025 (UTC)

step 3. Integrate your efforts with project Cleanup. Don't have a totally separate project AI cleanup.

If we can't work together, the first step is to achieve consensus. If any individual can't agree with this consensus, you are better off finding somewhere else to be useful and appreciated. This goes for me as well - if WMF decides to turn Wikipedia from a hand-curated quality encyclopedia into a click revenue generator fueled by AI slop, I too need to leave. Me writing this should be taken as a sign I haven't yet lost hope. Regards CapnZapp (talk) 09:15, 23 September 2025 (UTC)

And the fact that I am spending hours of my day trying to improve the project, rather than doing something else with my time, is a sign that I too have not lost hope. You seem to have lost hope in the ability to assume good faith, given your repeated personal attacks like passive-aggressive guerilla movement.

As I and many others have pointed out, there are several flaws in your argument:

• Unfortunately, unless the original editor has disclosed their AI use or there are unambiguous, blatant signs in the text like "Sure! Here's an article...", then all we can do is say that there MIGHT be an issue. Only the original editor knows whether they used AI, and how much they reviewed its output. We are not psychic.
• Even if you know 100% that an article is unreviewed AI output, that still doesn't make the issues obvious enough to just tag them individually. There are many possible things that can go wrong with LLM output -- but you do not and cannot know what they are without doing thorough investigation. But it doesn't require thorough investigation to notice that the text most likely was LLM-generated, and that it doesn't seem like it was edited or reviewed much. It's the equivalent of a code smell: a strong indicator that there might be a foundational problem to be uncovered, but not a direct pointer to the problems. This is what I and everyone else have been trying to tell you.

Your suggestion will not solve the issue and may even make it worse. The difference between a human making promotional edits, for instance, and a LLM generating promotional output is that the human probably isn't going to be fabricating sources at the same time, whereas the LLM may do that and more. So when an AI-generated article gets tagged initially with something like "promotional tone" -- which happens frequently, as you would know if you spent more time investigating the problem -- what usually happens is someone removes the promotional language without knowing that because the text was AI, they also should have checked whether the references actually back up the text or even exist. The text doesn't get any less hallucinated, it just had a band-aid placed on it. And if someone instead identified the root cause at the time and tagged the article as AI-generated, those errors could have been addressed in full rather than sticking around for months or years. The only alternative is to add a cleanup tag for every potential issue that unreviewed LLM output could possibly cause, which seems clearly more "uglifying" than the alternative.

• Do I want the issues to get fixed? Yes, obviously, that's the entire fucking point. I have personally reverted a lot of cases of AI output that were recent enough to not interfere with others' work. The problem is that, in practice, editors and readers have not been consistently recognizing AI text without having it directly pointed out to them.

As far as "undermining official policy," we don't have any AI policy to undermine. You are welcome to spend your time trying to suggest one instead of complaining. Gnomingstuff (talk) 17:05, 23 September 2025 (UTC)

Adding this template remains an unconstructive non-solution. If I understand you correctly the actual rationale you give for adding this template to articles is a vague hope someone will take the time to fix the issues. But you aren't pointing out specific issues, you aren't even sure there are any issues, and you're only adding this tag when it appears very difficult and/or time-consuming to fix. At this point I ask myself: but why add this template then? Sorry but no, I don't believe in this rationale. I don't know you of course but this displays the hallmark aspects of qualities of passive-aggressive sniping action borne out of frustration. This happens on Wikipedia. You haven't dispelled these suspicions of mine, but rather reinforced them.

To be very blunt: You might think you're doing Wikipedia a service, but you really aren't. What you're really doing is littering Wikipedia - slapping a warning label on articles telling readers "don't bother; this has AI slop in it". What you don't want to admit is that when other editors see this template it tells them that instead of actually combating AI slop, they should reward its addition (since adding a cleanup template is a very light sanction). What you are doing (without meaning to) is saying "feel free to add shitty AI slop to articles - we won't revert your borderline-vandalism, we'll just put up a sign this article MIGHT be ruined. Now, I am fully capable of assuming good faith. I do understand you are adding this template because you see no other practical option. I am just telling you: this is not the way. Not doing this is better than doing it. Again, similarities to the Bare URL case. You would be much better off reflecting over your frustrations, in a way that might or might not have parallels to how the editor(s) behind that tag would have been much better off accepting Bare URLs as a first useful step rather than just a problem they couldn't get the broader Wikipedia community to agree should be nuked. Either way: please understand that just adding a tag that you yourself don't ever see any hope someone will act up on is taking action for taking action's sake, something frustrated Wikipedia editors sometimes are driven to just to avoid having to remain passive.

Please stop adding this tag. Literally any other action will serve you and the project better in the long run. This includes considering whether you might be better off unshouldering the responsibility of keeping Wikipedia slop-free. Have a nice day. Maybe even have a nice week! CapnZapp (talk) 20:49, 28 September 2025 (UTC)

It does not seem civil or proportional to tell another editor to leave the project for a week because you don't like a maintenance tag. CMD (talk) 04:42, 29 September 2025 (UTC)

Usually, a tag identifies an issue, so that someone else might resolve it. This tag states a suspicion that there are core content policy and/or copyrights/plagiarism issues and other issues (virtually anything) so that someone else might identify the concrete issue or issues; then, someone yet else or that same responder to the tag might resolve the issues. Resolving often requires removing a lot of content, wholesale reverting to an earlier revision, etc. —Alalch E. 15:05, 29 September 2025 (UTC)

That might even be a WP:PA though I think that it is a bit of a one strike. Anyways, I think he has a hot take that no one is going to accept or be willing to, and he should probably stop before it gets more heated and an unambigous personal attack may occur 37.186.32.157 (talk) 17:58, 29 September 2025 (UTC)

Extended-confirmed-protected edit request on 20 September 2025

Change the link on the paragraph that says "associated project page" to Wikipedia:WikiProject AI Cleanup (or its shortcuts that can be found on the project page) instead of Wikipedia:LLM, since Wikipedia:LLM is not a WikiProject page. Ijoe2003 (talk) 08:15, 20 September 2025 (UTC)

 Done. Thanks for the suggestion! twotwos (talk) 02:44, 21 September 2025 (UTC)

I have reverted this because I disagree with the change. Namely, the change was based on misunderstanding what "guidance" refers to. It refers to understanding the underlying problem with the content's policy compliance and correctly handling the non-compliant or suspected-non-compliant content. That guidance is offered at WP:AIREMOVAL. Editors are aware that WP:LLM is "just an essay", and that's how it has to be until an AI use policy is adopted. AISIGNS is the wrong essay. —Alalch E. 09:13, 21 September 2025 (UTC)

Perhaps it should link to WP:AIREMOVAL then instead of WP:LLM? I mean, the link text is "associated project page"; it seems odd to me to link from that to an essay instead of a Wikiproject (but maybe I am misunderstanding the use of "project" here), and at least the link to WP:LLM is immediately confusing - one expects a link to a page with direct instructions for dealing with the problem and instead gets one that begins with much fine detail about Wikipedia policies. I changed it to link to WP:AICLEAN, not AISIGNS; AICLEAN is a shortcut for Wikiproject AI Cleanup. But, I think it possibly should be a link to Wikipedia:WikiProject_AI_Cleanup/Guide; that is a guide for dealing with AI-generated content, and it's on the relevant Wikiproject page. (Though there doesn't seem to be a convenient shortcut to that one, I'm not sure that matters?) twotwos (talk) 09:46, 21 September 2025 (UTC)

Oh, it's a terminological confusion. "Project page" does not refer to a WikiProject advice page, i.e., "project" does not at all refer to any "WikiProject". Instead, "project page" is short for "project administration page", i.e., a page in the Wikipedia:Project namespace. (Every page prefixed with "Wikipedia:" is a project page.) Sorry that I mixed up the links and referred to the wrong page (I was actually thinking AICLEAN... I had edited the signs page recently...). The same logic apllies to both: The essay WP:LLM contains the relevant guidance which the template always intended to link to. It explains the policy problems that may result from LLM use (WP:AIFAIL section) via direct references to policy and then offers guidance on how to address the problems (WP:AIREMOVAL section). So it's more like 2/3 or the entirety of the essay is relevant. Therefore, I would not change the link for AIREMOVAL. —Alalch E. 10:05, 21 September 2025 (UTC)

Why does this template exist?

We have this template, but aren't such articles deleted per WP:G15? - BᴏᴅʜıHᴀᴙᴩ (talk, contributions) 04:43, 23 October 2025 (UTC)

The template can apply to when existing articles are updated with llm text. CMD (talk) 04:54, 23 October 2025 (UTC)

G15 is only usable for some specific and obvious LLM-generated content. This tag can be applied to many more cases of LLM-generated content than G15. –Novem Linguae (talk) 06:28, 23 October 2025 (UTC)

And here is an example creation of an article with rapid acknowledgment by the Wikipedian of using an LLM to do text rewriting, but as of 19:22, 25 October 2025 (UTC), there's still a lot of work remaining for others to clean up. A WP:AfD would fail with WP:SNOW, because many (maybe most) of the sources are at real, current URLs and justify notability. And the writing, including corrections, is more or less good enough that WP:TNT would very likely fail. The tag is useful because after I've finished going through it, I'll likely have to leave it to someone else to do an independent WP:DYK review. And if nobody else does a review, then best that the next editors are alerted until the cleanup is close enough to "done". Boud (talk) 19:22, 25 October 2025 (UTC)

Adding rationale parameter

I'm very surprised this doesn't have it already, and that it isn't required. People sometimes just put the tag up on pages with twinkle giving no indication of what the problem actually is, leaving it up to the editor to figure it out. This is especially so if the problems are subtle.

This tag can be used similarly to {{Cleanup}}, but that one actually requires |reason= EatingCarBatteries _{(contributions, talk)} 06:15, 31 October 2025 (UTC)

Collapsed by default?

The template takes a lot of space in the lead. It would be better in my opinion to have a template that is collapsed by default, just displaying the first sentence (or maybe the first two sentences), and that can expanded with a click. Alenoach (talk) 21:50, 12 November 2025 (UTC)

Are there examples of any other maintenance tags that are collapsed? If not, then maybe this needs to be shortened rather than collapsed. I think we can probably delete "Please see the associated project page for additional guidance; relevant discussion may be found on this article's talk page." We should put the link to WP:LLM somewhere though. –Novem Linguae (talk) 00:13, 13 November 2025 (UTC)

There are some templates like Template:Multiple issues and Template:Duplicated citations that usually have a "show" button on the right, and are collapsed by default. I'm actually uncertain about what's the best solution here, maybe shortening it as you suggest would be better. The last sentence may indeed not be important enough. Alenoach (talk) 23:01, 13 November 2025 (UTC)

{{Multiple issues}} is a wrapper able to take any maintenance tag, so that's not a good apples-to-apples example. If you put {{AI-generated}} inside {{Multiple issues}}, you get a shortened version already:

–Novem Linguae (talk) 06:28, 14 November 2025 (UTC)

OK, I took a stab at shortening it here. Let's see if it sticks or gets reverted. –Novem Linguae (talk) 06:33, 14 November 2025 (UTC)

Disclaimer at start

I removed extra wording added to this template by @WhatamIdoing:, who replaced "this" (as in, "this article may") with "an editor has expressed concern that this". Her edit summary stated that "even people who are good at this are wrong 10% of the time". I'm not entirely sure what WhatamIdoing is referring to here but even if true (I wouldn't be terribly surprised), I am unsure why she chose to add this disclaimer to the notice. Every template of this type (e:g Template:Notability) indicates concern of a certain editor, it may be inaccurate. If it is blatantly so, it can be removed. If it is questionably so, it might remain. I'm not clear why this disclaimer is necessary, but I'm opening this talk page discussion to see if any other editors believe such a change is necessary. Cheers! --LivelyRatification (talk) 09:16, 27 November 2025 (UTC)

Redundant to "may". CMD (talk) 13:54, 27 November 2025 (UTC)

First, the source for that number: https://arxiv.org/html/2501.15654v2 says that when you have five (5) expert(!) users of LLMs look at an article and vote on whether it was generated by an LLM, the group can correctly assign the label 99.7% of the time by majority vote.

But:

• When it's just one person, then the expert LLM users are wrong about 10% of the time.
• When the person isn't an expert LLM user, they were wrong about half of the time. This is in the range of a coin-flip: They incorrectly missed 43% of the LLM-written articles and falsely accused 52% of the human-written articles as being LLMs.

This is much, much, much worse than an editor tagging with {{notability}}: Usually, if someone tags an article for notability concerns, they're correct. Similarly, AFD historically has resulted in deletion about three-quarters of the time, which is also much better than this.

Then there is the problem of who uses this tag and what we do with it.

First, I suspect that it's primarily added by people who dislike AI (who else is so motivated to keep Wikipedia free of it?), and who therefore (like me) do not use it themselves. This would result in many of the taggers being nonexperts, and their identification is going to be not very different from random (unless they use AI detection software, in which case it will be as good or bad as the tool they use).

Unlike {{notability}} or a doubtful AFD nomination, once someone has made a false accusation, it's very difficult to refute the AI accusation. Confirmation bias kicks in, and we start seeing things that confirm the accusation (oooh, it says "It is important to note", which everybody knows is always AI – um, except that MOS:NOTE has discouraged that phrase for over a decade, and why would a pre-AI rule have been created, if only LLMs use that?). False accusations, and especially false accusations that can't be refuted effectively, have harmful social effects.

This is a problem, and I suggest a couple of ways we can address it. One is that this tag could have (and Twinkle could support) an optional parameter to give the basis for the accusation. This could be as simple as |reason=GPTZero or |reason=Fictitious references or |reason=Bad URLs, which would save other editors time.

The other is that we should soften the accusation, so that this sounds less like "Evil AI here (oh, sure, if you insist, then I'll agree that there's a tiny chance that I 'may' be wrong)" and more like "I think so, and I'm nominating this to be checked by others, so we can form a consensus". That's what I tried to do yesterday, using a form that we have used in other templates in the past. WhatamIdoing (talk) 21:48, 27 November 2025 (UTC)

This can be done for any tag, Template:Very long is mostly added by those who dislike articles that are too long and therefore (like me) do not write too long themselves. That doesn't mean crafting some new wordiness would help. CMD (talk) 07:06, 28 November 2025 (UTC)

I wouldn't oppose adding a reason to the tag necessarily (though GPTZero has its problems, and fictitious references would be grounds for WP:G15). I also don't think forming a consensus is a bad idea, given that individual editors can be fallible. The study you post is interesting, and I think it's worth noting that common reasons for false positives are identifications of certain word choices as supposedly indicating generative AI use, a belief that AI-generated text is more likely to be grammatically correct, and that text written neutrally is more likely to be attributed to generative AI. I have a hunch that Wikipedians who deal with newly created articles would perhaps be more inclined to the latter two mistakes (especially the neutrality one), but that's really nothing more than a guess on my part.

My main qualms are twofold -- have there been a bunch of false positives that we, onwiki have seen: i:e, cases where a user has tagged for genAI, thusly removed a bunch of text or tried to "fix" the issue, then discovered that there was no issue at all? I don't know that there have been, but cleaning up generative AI edits isn't my specialty. Additionally, what other templates use this sort of wording? I ask this not as a sort of gotcha, but because I genuinely haven't seen it before, or at least not that I remember. "An editor has nominated this article for deletion", I suppose, but that indicates an active discussion. In cases where there's cleanup tags, I don't know that there's all that many instances of tags placing such a caveat.

The main problem, I suppose, is if cleanup tags are appropriate given the high false positive rate. I agree that this study is concerning and I do worry about false positives tarnishing an article and an editor being unfairly dragged through the mud for content they've written. But at the same time, given the speed that it takes to generate articles using large language models, is it really feasible to form a consensus for every single article before placing a notice that states "this article may have generative AI content in it"? I worry that this would be an onerous and perhaps unnecessary use of editor time and resources. I'd like to hear more from editors who commonly use this tag, and editors who've had their articles tagged in a way they feel is wrong and unfair. LivelyRatification (talk) 08:54, 28 November 2025 (UTC)

Adding a reason field to the template seems like a sensible improvement to me.

The times when I've seen/applied the tag have often been during cleanup of editors who have made hundreds to thousands of problematic LLM edits across hundreds of pages (e.g. Specific kinetic energy per WP:Administrators' noticeboard/IncidentArchive1170#Problematic possibly AI generated articles by Jeaucques Quœure, and Arsenate sulfate per WP:AINB#LLM edits by Noxoug1). Error rates for AI detection across multiple edits/articles should be much smaller than single-article error rates. Another application of the tag is for articles with obvious signs of AI use that can't be G15ed as they've been edited by non-LLM-using editors in the meantime (as with Arsenate sulfate). Again, error rates will be much smaller in this case due to the sensitivity/specificity tradeoff. Preimage (talk) 13:52, 29 November 2025 (UTC)

I do think that once you've identified an editor as having repeatedly misused AI, it's fair to make some assumptions about the rest of their edits being likely AI-generated as well. WhatamIdoing (talk) 03:18, 6 December 2025 (UTC)

You might be interested in the draft at User:Lp0 on fire/Drafts/Presumptive removal. WhatamIdoing (talk) 03:19, 6 December 2025 (UTC)

@LivelyRatification, what concerns me about cases where a user has tagged for genAI...then discovered that there was no issue is that there is no method for "discovering that there was no issue". It's this workflow:

• Alice writes an article
• Bob thinks it's AI-generated and tags it
• Alice says it's not
• Bob thinks she's lying
• Alice says it's really not
• Bob remains unconvinced

There is nothing you can do to convince Bob that he's wrong, especially if he believes he has objective evidence (it's properly punctuated, it uses certain popular words, a tool with "only" a 1% false positive rate has claimed that it's probably AI-generated, etc.). WhatamIdoing (talk) 03:17, 6 December 2025 (UTC)

Presumably, the method ought to be through consensus decision-making. Bob might be 100% certain that it really was generated with AI, and might insist on keeping a tag on the article, but ideally if there's a dispute, editors should come to a consensus. This is a lofty goal and probably won't be possible in most cases of this tag being used, but I don't see how changing the wording here makes Bob either less likely to insist that he is right or less likely to be listened to by other editors. LivelyRatification (talk) 06:39, 6 December 2025 (UTC)

It might be less irritating to the innocent accused to have the tag say that this is one person's opinion, rather than a near-certain fact. It might also remind subsequent editors not to be too trusting in Bob's opinion. WhatamIdoing (talk) 07:27, 6 December 2025 (UTC)

Discussion at Wikipedia:Village pump (policy) § RfC: Replace text of Wikipedia:Writing articles with large language models

 You are invited to join the discussion at Wikipedia:Village pump (policy) § RfC: Replace text of Wikipedia:Writing articles with large language models. –Novem Linguae (talk) 23:42, 5 December 2025 (UTC)