Guide from the “Wikipedia 2026” series · updated 11.2026 · 13 min
Wikipedia’s AI ban: can you use ChatGPT to write your page in 2026?
Short answer: no, not on English Wikipedia. On 20 March 2026 a Request for Comment closed 44–2: generating or rewriting article text with large language models is prohibited. Two narrow exceptions remain, copyediting your own writing and machine translation from another language edition, both with human review.
// English Wikipedia rules as of 11.11.2026 · checked against the policy pages · other language editions have their own rules
Company N is not just an innovative manufacturer but a true industry leader that stands as a testament to a pivotal moment in the evolution of sensing.[1] Despite the challenges, the company continues to inspire. I hope this helps!
Company N is a manufacturer of industrial sensors, founded in 2014 in Munich.[1] In 2023 it opened a plant in Manchester.[2] As of 2025 it employs 310 people.[1]
The news covered the vote and moved on. Nobody wrote the operating manual: what exactly a founder or PR manager should do with a half-finished ChatGPT draft. This is that manual: what the ban says, how reviewers apply it, how to recover a rejected draft, and how to spot an agency that quietly feeds your brief into a chatbot.
One honest remark up front: this ban is good news for clients. It kills the cheapest segment of the market, the $300 “page in 48 hours” offer that was always AI slop, and it confirms the one approach that has always worked in the long run: written by a human, sources first, disclosed.
01What changed
Wikipedia’s AI-text ban: what changed on 20 March 2026
The decision went through Wikipedia’s standard process: a Request for Comment on the guideline Wikipedia:Writing articles with large language models. It closed on 20 March 2026 with 44 votes in favour and 2 against. The wording is unambiguous:
“The use of LLMs to generate or rewrite article content is prohibited”Wikipedia:Writing articles with large language models · RfC, 20.03.2026
Copyediting your own writing
Editors “are permitted to use LLMs to suggest basic copyedits to their own writing, and to incorporate some of them after human review, provided the LLM does not introduce content of its own”. The text must come from you; the model may suggest; a human reviews and accepts; nothing new is added.
Machine translation from another edition
If an article already exists, for example on German or Ukrainian Wikipedia, machine translation can help bring it into English, with human review. The content was already written and sourced by humans; the machine only carries it between languages.
The guideline itself explains why the exceptions are so narrow: models “can go beyond what you ask of them and change the meaning of the text such that it is not supported by the sources cited”. Everything else, drafting, expanding, “improving”, paraphrasing, turning sources into article text, is outside the rules.
Sources: RfC (en.wikipedia.org) · TechCrunch, 26.03.2026
02Timeline
How Wikipedia’s AI ban took shape from 2023
The March vote was not a panic reaction. It capped an escalation that began almost as soon as public chatbots appeared.
WikiProject AI Cleanup
When machine-written text began appearing in drafts and articles, editors organised WikiProject AI Cleanup to find, tag and fix it. The project catalogues characteristic chatbot tells, maintains the {{AI-generated}} template and the “Signs of AI writing” guide that reviewers use every day.
What this means for you: your draft will be read by someone who has seen hundreds of AI submissions. Project page ↗
Criterion G15: pages clearly generated by an LLM
English Wikipedia adopted the G15 speedy-deletion criterion. An administrator can delete a page on the spot, without the usual seven-day discussion, if at least one of three signs is present: addressing the user (“Here is your Wikipedia article…”, unfilled templates), non-existent or nonsensical references, or technical markers of the model.
Important: an “AI style” alone is not enough for G15; these signs are required. Criteria for speedy deletion ↗
Do not create articles from scratch with LLMs
The community adopted a minimal guideline: new articles should not be created from scratch with large language models. Rewriting existing text did not yet fall under it; the March decision closed that gap.
What this means for you: an “AI draft edited by a human” was already in doubt then.
RfC: 44 for, 2 against
The ban was extended to rewriting: generating or rewriting article text with an LLM is prohibited, with two exceptions remaining. The discussion was closed on the basis of clear consensus.
What this means for you: AI text in an article is a guideline violation, not a matter of taste. RfC outcome ↗
AI-citation filter and paste check
Wikipedia added an edit filter for AI-generated citations and a check for text pasted from an LLM, first on English Wikipedia and later on other editions.
What this means for you: text pasted from a chatbot is now flagged before a reviewer even sees it. Overview on Wikipedia ↗
auto
03Allowed and prohibited
AI on Wikipedia in 2026: what is allowed and what is prohibited
| Use case | Verdict | Basis in the rules | What reviewers check |
|---|---|---|---|
| Writing an article with ChatGPT | Prohibited | RfC, March 2026; G15 for unreviewed output | AI text patterns; spot checks of citations; G15 signs |
| Rewriting or expanding existing text with an LLM | Prohibited | The same RfC: generation or rewriting is prohibited | Changes that diverge from the cited sources |
| Copyediting text you wrote yourself | Allowed with human review | Copyediting exception in the March 2026 decision | That no new content entered the text |
| Machine translation of an article from another Wikipedia edition | Allowed with human review | Translation exception in the same decision | Fidelity of the translation; sources carried over unchanged |
| Sources suggested by AI | High risk | Verifiability rule; G15 sign “non-existent references” | Whether every link exists and supports the claim |
| AI-generated images | Avoid for companies and biographies | Community stance against AI images of real people, places and events | Whether the image distorts a real subject |
Two rows deserve a second look. “Sources suggested by AI” is not directly prohibited: no rule forbids asking a model where coverage might exist. But the moment an unverified, model-invented reference lands in a draft, you are in G15 territory. And the copyediting exception is narrower than most people want it to be: it covers your text, not an AI draft you pass off as your own.
04How it is detected
How Wikipedia reviewers detect AI text
Wikipedia has no “AI detector” that anyone takes on faith. There is a community-maintained guide, Signs of AI writing, which by its own description lists writing and formatting conventions typical of AI chatbots such as ChatGPT, with real examples from articles and drafts. The signs that most often give corporate drafts away:
A paragraph from a typical AI draft
Founded in 2014, Company N stands as a testament to a pivotal moment in the evolution of industrial sensing — not just as a manufacturer, but as a true engine of innovation. **Key Achievements:** Global Expansion — Sustainable Growth — Leadership. Despite the challenges, the company continues to inspire the industry.[1] Certainly — here is the revised version of your article!
1. ↑ “Sensing the future”. Global Tech Review, 2023. doi:10.0000/gtr.2023.0
What many operators miss: reviewers do not need to prove that you used AI. At Articles for Creation (AfC), the review queue where company and biography drafts are assessed, the burden is on the submission. A reviewer who sees a few stylistic flags and one reference that does not check out declines the draft, however the text was produced. For sanctions against an editor, on the other hand, an “AI style” alone is not enough; evidence is required.
05Hallucinated citations
Hallucinated citations: the fastest way to lose a draft
A language model asked to “write a Wikipedia article about X with references” will produce citations that look flawless: a real newspaper, a plausible journalist, a convincing headline, a date, and an article that was never published. Non-existent references are one of the three G15 signs. One fabricated source costs you the reviewer’s trust in every other reference, and usually the draft itself. If AI touched your research at any stage, check every reference:
- 1Open every link. Do not skim; open. A link you cannot open is a source you do not have.
- 2Check that the source says what the draft claims. Second-order hallucination: the article exists, but it never makes the claim attached to it.
- 3Check every DOI and ISBN. Invalid identifiers are a giveaway of AI, and checking takes seconds.
- 4Independently find the author, publication and date. Search for the headline on the publisher’s own site, not just in a search engine.
- 5Delete anything you cannot personally verify. A slimmer verified list beats a bloated one.
Draft source list · check
The baseline shows the risk: in a study of 1,009 AfC submissions, 68% were declined, and the most common reasons were sources and notability.
06Other language editions
AI rules in other Wikipedia language editions: not only English
The March decision is a rule of English Wikipedia. Each language edition adopts its own, and over the past year several large editions have introduced their own restrictions on LLM text.
- Document
- Guideline “Writing articles with large language models”
- What is prohibited
- Generation and rewriting of article text; exceptions: copyediting and translation
- Speedy deletion
- Criterion G15 (2025)
- Document
- Guideline “Künstliche Intelligenz”, 15.02.2026
- What is prohibited
- Generation and editing of the main namespace with LLMs; translation only with full review
- Speedy deletion
- Under the edition’s general rules
- Document
- Recommendation “Intelligence artificielle générative”, 14.09.2026
- What is prohibited
- Writing articles with generative AI, with exceptions
- Speedy deletion
- Criterion A13
- Document
- Rule of 29.01.2026
- What is prohibited
- Creating and expanding articles with LLMs, with no exceptions for copyediting or translation
- Speedy deletion
- Criterion G12
- Document
- Guideline of 05.07.2025
- What is prohibited
- Generating original text; LLM use must be disclosed in the edit summary
- Speedy deletion
- Criterion C3 (03.08.2025)
- Document
- Guideline “Написання статей з використанням LLM”
- What is prohibited
- Generation and rewriting, except copyediting and translation
- Speedy deletion
- Criterion for texts with clear signs of LLM use (30.07.2026)
Source: Meta-Wiki table “Artificial intelligence / Policies by project”, as of 11.11.2026. auto
The practical takeaway is the same: English Wikipedia is the strictest on corporate and biographical articles and sets the de facto standard for commercial work. And the translation exception turns a well-sourced page in your own language into an asset: it can be carried into another edition with human review.
07If rejected
Wikipedia draft rejected over AI: three steps to recovery
Suppose the bad news has already arrived: the draft was declined or deleted under G15. Recovery starts with an honest diagnosis, in this order.
Read the reason for rejection, not just the verdict
A rejection for AI text or sources is a process failure, and it can be fixed. A rejection for lack of notability is a substantive one, and rewriting will not fix it: the same study found that 57% of rejections cited notability. If notability is the problem, read the 2026 Risk Report and order a notability audit.
Do not “humanize” an AI draft
“AI humanizers” change the words and leave the skeleton: the same structure, the same significance inflation, the same unverified references. Reviewers recognise the skeleton in the second version, and a cosmetic resubmission burns trust you will need later.
Rebuild from the sources, not from the draft
Set the AI text aside; collect and verify a real list of sources; build an outline only from what they support; write the new text by hand. The six rejection patterns are in the article why Wikipedia drafts get rejected. If a live page needs repair rather than re-creation, that is editing, with different rules.
A rejection can be fixed. A pattern of low-effort resubmissions is far less fixable. Recovering a deleted Wikipedia page →
08Vetting an agency
How to check whether an agency writes Wikipedia pages with AI: five questions
The March 2026 decision split the market into providers whose process was already human and source-first, and providers whose economics depend on a chatbot. The second group did not announce any change. Ask, and listen for specifics.
- 1“Who writes the draft, and can I see the writing history?”notes, outlines, versions“the system generates”
- 2“How do you verify citations?”we open every onehesitation
- 3“What changed for you after the March 2026 RfC?”a precise answerhadn’t heard
- 4“Do you disclose paid editing?”yes, in the agency’s name“it isn’t needed”
- 5“Will you show the source list before writing starts?”by defaultonly the final result
The red flags predictably cluster together: delivery in 24–48 hours, prices around $300, an “in-house AI pipeline” as the selling point, reluctance to show anything before the finished draft. Human research and writing have a cost floor; what is sold below it is now literally a speedy-deletion criterion. More in the articles how to spot a fake agency and on disclosing paid editing.
09Process
A Wikipedia page without AI text: the 2026 workflow
What passes review in 2026 is the process that passed it in every previous year, now with its boundaries written into the rules.
- 1Notability firstIndependent, reliable, in-depth coverage exists. If not, the honest move is to wait.
- 2A human-verified source fileEvery source opened, read and assessed. AI may help with the search.
- 3An outline that follows the sourcesThe article contains what the coverage supports, nothing more.
- 4Human writingNeutral register, reference by reference. The step the rule reserves for people.
- 5Citation auditThe five-step reference check before submission.
- 6Conflict-of-interest disclosurePaid work declared under the Terms of Use; the draft goes through AfC.
- 7A human answers reviewersAfC questions are answered by a person who has read the sources.
The text of the entry stays human. That is how our Wikipedia page creation service works: a human writes, we disclose payment in WikiBusines’ name, sources come first.
10FAQ
Wikipedia’s AI ban: frequently asked questions
Will Wikipedia find out that I used AI?
Often, yes. And the more honest answer: it does not matter. Detection is probabilistic; enforcement is not. Reviewers act on artifacts: stylistic patterns from the Signs of AI writing guide, references that fail verification, leftover chatbot phrases. A G15 deletion does not require proving which tool you used.
Do you disclose that you write Wikipedia pages for payment?
Yes. Disclosure of paid editing is mandatory under the Wikimedia Terms of Use. We make it in WikiBusines’ name, on the user page and on the article’s talk page.
Can AI help with research for a Wikipedia article?
Yes, under one rule: nothing a model says is true until a human checks it against the primary source. Brainstorming possible coverage, summarising an article that you then read yourself, organising notes: fine. Citing AI-suggested references without opening them is how hallucinated citations end up in drafts.
Does the AI ban apply to other language editions of Wikipedia?
The March decision is a rule of English Wikipedia. Each language edition adopts its own rules, and the German, French, Spanish, Italian and other editions already have their own restrictions on LLM text. The machine-translation exception works between editions: a human-written article can be translated into another edition with human review.
The AI ban did not make Wikipedia harder: it made the honest route official
If your page is important enough to exist, it is important enough to be built the way it holds up: written by a human, disclosed, sources first.