Can ChatGPT write a Wikipedia article?

Can ChatGPT or another AI create a Wikipedia page?

Not on English Wikipedia. The content guideline on writing articles with large language models states that using them "to generate or rewrite article content is strictly prohibited", except for the uses it lists. The rule was adopted after a request for comment in March 2026, and the guideline names ChatGPT, Gemini, Grok, Claude, DeepSeek and Grammarly as examples of the tools it covers.

The prohibition applies to drafts as well as published articles. The instructions given to Articles for Creation reviewers say so explicitly, and the main Articles for Creation page lists "Do not use large language models to write your draft" among its important rules.

The guideline gives Wikipedia's reasons in two sentences: text generated by these models often breaks several of Wikipedia's core content policies, and it can be produced faster than other editors can review it. That second point matters. Wikipedia depends on volunteers checking each other's work, and a tool that produces plausible text in seconds shifts the whole cost of checking onto them.

What does Wikipedia's AI guideline still allow?

Two things, both narrow.

Copyediting your own writing. Editors may use a language model to suggest corrections to text they wrote themselves and accept those suggestions after reviewing them. The guideline limits this to "spelling, punctuation, capitalisation, grammar, and other simple mistakes", and the editor must make sure the corrected text is still supported by the sources cited. Asking a tool to make a paragraph sound more polished, more encyclopedic or more impressive goes beyond that.

Translation from another Wikipedia. Editors may use language model tools to translate an article from another language's Wikipedia into English, but only under the separate guideline on assisted translation. Its conditions are strict:

1. The editor must be skilled enough in both languages to confirm the translation is accurate.
2. They must be reasonably sure the original article was not itself machine-written.
3. They must check for and remove invented content and other breaches of the core content policies.
4. They must check the original's sources and make sure the translation reflects them fairly, with inline citations for anything likely to be challenged.
5. They must attribute the original article as the terms of use require.

Raw machine translation may be kept in user space or draft space while it is being worked on, but it must be tagged at once as needing review by someone fluent in both languages. Our translation service works within these limits.

What happens to an AI-written draft at Articles for Creation?

It is likely to be declined with a reason written for exactly this case. The decline notice says the draft "appears to contain text generated by a large language model (such as ChatGPT)", that LLMs cannot be used to generate, draft or rewrite article content, and asks the author to delete the generated portions and summarise independent, reliable, published sources in their own words.

The notice also deals with the exceptions. An author who believes their use was covered by copyediting or translation is asked to provide a transcript of the prompts they used. Wikipedia's frequently asked questions add that some writing tools include AI features without making it obvious, so an author who believes they did not use AI should check what their tools actually did to the text before disputing the decline.

Reviewers are told to be careful in the other direction too. Their instructions say that writing style alone is not proof, that they should not rely on their impression or on an AI-detection tool, and that the speedy deletion criterion has specific requirements that should not be applied just because a draft seems machine-written.

A decline for AI use stays on the draft with every other review notice, which is part of the record described in our guide to a declined Wikipedia draft.

What is speedy deletion criterion G15?

G15, headed "Unambiguously LLM-generated pages", lets an administrator delete a page without discussion when its current revision, or a substantially similar earlier one, shows signs that it could only plausibly have been produced by a large language model. It applies to every kind of page, drafts and user pages included. The criterion lists three kinds of sign:

The criterion is deliberately narrow. It says other common signs of AI writing are not enough on their own, and that even an author admitting to using a language model does not by itself qualify a page. Tagging editors are told to rule out ordinary link rot and typos, including by checking the Wayback Machine, before treating a reference as invented. Where G15 does not fit, the criterion points editors to proposed deletion for AI content, to Articles for Deletion, or to Wikipedia's AI noticeboard.

What is presumptive removal of AI-generated content?

It is a deletion guideline that lets editors clear out a user's contributions in bulk once that user is established as having misused AI. It has two parts. Edits to existing pages may be reverted without review, and articles where that editor is the only significant contributor can be nominated through a special form of proposed deletion, known as LLMPROD.

The guideline sets requirements before it applies. The editor must be blocked for AI use, or be inactive with strong evidence of habitual use, or have disclosed AI use and consented to clean-up, or be the subject of a noticeboard consensus that presumptive removal is appropriate. The contributions removed must match the timing and pattern of the editor's AI use.

An article nominated this way is deleted by an uninvolved administrator if the tag has not been removed within five days, and unlike ordinary proposed deletion an article can be nominated this way more than once. Any editor can reverse the removal, but only by taking responsibility for the content: reviewing every claim and source, and reworking the text so that it complies with policy before restoring it.

The practical consequence is that one person's AI use can reach every page they touched, including pages about clients or employers, long after the original edits.

Why do AI-written Wikipedia drafts fail even when nobody spots the AI?

Because the output collides with Wikipedia's ordinary content rules, which apply regardless of who or what wrote the text. Wikipedia's information page on large language models describes the main problems, and reviewers see them constantly.

Invented and misattributed sources. Language models can produce citations with plausible titles, authors and addresses for articles that do not exist. Wikipedia's reliable sources guideline notes that models tend to "hallucinate" false information, including references that look as if they come from reputable publications. A reviewer who opens the links finds the problem in minutes.

Claims nobody can verify. The verifiability policy requires material that is challenged or likely to be challenged to be supported by a reliable published source. Text generated from patterns rather than from sources fails that test by construction. Wikipedia's information page describes it as, in practical terms, equivalent to original research, or worse, outright fabrication.

Promotional tone. Asked to write about a company or person, a chatbot tends towards admiring language. Wikipedia's Articles for Creation guidance says that even an accurate statement can sound promotional if it is written to impress, and promotional drafts are declined and, in clear cases, deleted.

Biographies of living people. The information page warns that generated text can libel living people. For biographies, unsourced or poorly sourced contentious material is removed immediately, so a generated claim is a liability, not a placeholder.

No help with notability. A chatbot cannot create independent coverage of a subject. If significant coverage in reliable, independent sources does not exist, no draft, however fluent, will be accepted.

Can you be blocked for using AI on Wikipedia?

Yes. The guideline states that persistent use of large language models outside the two exceptions is considered disruptive and will result in a block, following a separate request for comment that ran from March to May 2026. Wikipedia's disruptive editing guideline explains the reasoning: the effort needed to clean up generated content is disproportionate to the effort needed to create it, and users who carry on after a warning may be blocked.

The guideline also recognises that some people simply write in a style that resembles machine output. It says sanctions require evidence beyond stylistic or linguistic indications, and that the full pattern of an editor's recent edits and their compliance with content policy should be considered.

The rules extend to discussion. Under the talk page guidelines, comments, nominations and opening statements that are obviously generated by a language model can be struck or collapsed, and a deletion discussion opened that way can be closed early. An AI-written argument for keeping a page is likely to be set aside, not weighed.

What should you make of an offer of an AI-written Wikipedia page?

Treat it as a risk to the subject, not a saving. A page that is caught under G15 is deleted, with the criterion recorded in the deletion log against the title. A draft declined for AI use carries that notice into every later review. An account blocked for AI use can trigger presumptive removal across everything it wrote. Each of those outcomes is public and stays attached to the subject's name.

Wikipedia's Articles for Creation pages also warn that scammers offer publication or protection for payment. An offer that combines quick, cheap production with a promise of approval matches both warnings. Our guide to what a failed Wikipedia attempt leaves behind explains how those records affect a later, properly prepared attempt.

Should you write the draft yourself rather than use AI or a provider?

Writing it yourself is permitted and, done carefully, it can succeed. The work is mostly reading: finding the independent coverage, checking what each source actually says, and summarising it neutrally in your own words with a citation beside each claim. Nothing in that process can be handed to a chatbot under the current guideline, and a tool used only to correct your spelling must not change the meaning.

What the guideline cannot do is make the writing easier for someone who is close to the subject. The usual failures in a self-written draft are the same ones AI produces, only more slowly: praise the sources do not support, reliance on interviews and press releases, and claims that cannot be checked. A reviewer declines them for the same reasons.

Our Wikipedia article creation work is human research and drafting from sources we have read, prepared to meet the standard the first reviewer will apply. If a draft has already been declined or deleted for AI use, a free page health check will set out what the record now shows and whether a fresh attempt is realistic.