Wikipedia has taken a decisive step in the debate over generative AI and online knowledge. In March 2026, the English Wikipedia community formally restricted the use of AI-written article text, barring editors from using large language models to generate or rewrite encyclopedia content. The rule includes only two narrow exceptions: editors may use AI for basic copyedits to their own writing after human review, and for translation help from other-language Wikipedias, also subject to review.
This change did not come out of nowhere. It followed months of growing concern among volunteer editors, new enforcement tools introduced in 2025, and wider strategic discussions inside the Wikimedia movement about trust, verification, and the long-term health of the platform. The result is a policy that does not reject AI outright, but clearly limits where it can be used and where it cannot.
A Formal Restriction on AI-Written Content
The March 2026 policy marked a turning point for Wikipedia limits AI-written content. According to the English Wikipedia community’s new rule, editors are no longer allowed to use large language models to generate or rewrite article prose. That restriction applies to the text of encyclopedia articles themselves, the core material readers rely on for information.
The policy allows only two exceptions. First, an editor may use an AI system to suggest basic copyedits to text they personally wrote, provided that a human reviews the output and the system does not introduce new content. Second, AI may assist with translation from articles on other-language Wikipedias, again only under human review. These exceptions are narrow by design and do not permit AI to act as a co-author of Wikipedia articles.
Reporting on the policy also highlighted the reasoning behind the wording. A short quote cited by The Guardian summarized the principle clearly: “Editors are permitted to use LLMs to suggest basic copyedits to their own writing … provided the LLM does not introduce content of its own.” In practice, Wikipedia is drawing a sharp line between assistance and authorship.
Why the Community Decided to Act
The main rationale behind the policy is that AI-generated prose was judged incompatible with Wikipedia’s most important editorial standards. Wikipedia depends on verifiability, neutral point of view, and careful sourcing. Large language models, however, can produce fluent text that sounds credible while still including unsourced claims, distortions, or subtle bias.
For volunteer editors, that creates a serious problem. Wikipedia is not merely a place where readable text matters; every statement must be attributable to reliable sources and presented fairly. AI systems often generate wording that appears polished on the surface but fails under scrutiny. That mismatch makes human review more difficult, not less.
The community also acted with broad support. Reporting on the March 2026 request-for-comment said the measure passed with 44 votes in favor and only 2 opposed, with the discussion closing on 20 March 2026. Such a result suggests that among participating editors, there was strong consensus that the risks of AI-written content had become too large to ignore.
Warnings Had Been Building Since 2025
The 2026 restriction followed earlier enforcement steps. In August 2025, English Wikipedia had already escalated the issue by allowing suspected AI-generated articles to be nominated for speedy deletion. That meant the platform was no longer treating poor AI submissions as isolated editorial mistakes, but as a recurring quality problem requiring faster action.
Editors described the situation in striking terms during those discussions. One reviewer said they were being “flooded non-stop with horrendous drafts” created using AI. Others complained of “lies and fake references” that took substantial time to detect and clean up. These reports made clear that the burden of AI-generated submissions was falling directly on volunteers.
By 2025, the scale of the cleanup effort had become visible. WikiProject AI Cleanup reportedly tracked more than 500 suspected AI-written pages pending review. That kind of backlog showed that the issue was large enough to require organized triage, not just ad hoc editing by a few attentive contributors.
How Editors Learned to Spot AI Prose
As concerns grew, English Wikipedia created a dedicated “Signs of AI writing” guide in August 2025. The guide was meant to help volunteers identify likely AI-generated text more consistently. Rather than relying only on intuition, editors were given specific warning signs that frequently appeared in machine-generated drafts.
Among the reported red flags were fabricated or irrelevant citations, generic language patterns, and recognizable stock phrases such as “Here is your Wikipedia article on…” or “Up to my last training update.” Editors also flagged overuse of em dashes, words like “moreover,” promotional adjectives, and unusual formatting quirks often associated with LLM output.
This guidance reflected an important shift in moderation culture. Wikipedia was not simply reacting to a few bad edits; it was developing shared methods for identifying and handling a new type of content risk. The guide also underscored how AI-generated prose often leaves traces that experienced editors can recognize, even when the text initially looks polished.
Research Helped Quantify the Problem
The concern about AI-written pages was not based only on anecdotal evidence. A 2024 Princeton preprint offered one of the clearest attempts to measure the issue on English Wikipedia. Looking at roughly 3,000 new English articles from August 2024, the researchers calibrated detectors to a 1% false-positive rate on pre-GPT-3.5 articles.
Using that method, the study found that more than 5% of newly created English Wikipedia articles were flagged as AI-generated. Even with a cautious threshold, that is a significant share for a platform that depends on volunteer review. It suggested that AI-written content was not a fringe phenomenon, but a meaningful portion of new article creation.
For Wikipedia, those numbers mattered because the encyclopedia operates at enormous scale. When even a modest percentage of new pages may contain AI-generated issues, the downstream workload for patrollers, reviewers, and administrators grows quickly. Quantitative evidence helped justify why stronger policy responses were needed.
Not Anti-AI, but Against AI Authorship
It is important to understand that Wikipedia’s position is not a blanket rejection of AI. The new rule is specifically a limit on AI authorship in article text. Wikimedia planning documents show continued support for AI-assisted tools such as Edit Check and Structured Tasks, which are designed to guide contributors rather than replace them.
That distinction also appeared in educational guidance. In January 2026, Wiki Education warned contributors not to copy and paste generative-AI output into Wikipedia to create new articles, because the text often failed verification against sources. At the same time, it encouraged limited uses of AI, such as identifying possible article gaps or surfacing relevant sources for further human checking.
The broader Wikimedia message has therefore been consistent: AI can help with workflows, but it should not become the author of encyclopedia knowledge. Wikipedia limits AI-written content because the project values human judgment, source verification, and community consensus more than automated speed.
A Human-Centered Strategy Across Wikimedia
The editorial restriction also aligns with the Wikimedia Foundation’s wider AI strategy. In April 2025, the Foundation emphasized a “humans first” approach, describing both a promise and a commitment to the volunteers who make Wikipedia possible. The language was clearly human-centered, not based on replacing editors with machine-generated text.
That philosophy continued in the Foundation’s 2025,2026 product plan, which included a key result to “Evaluate the impact of generative AI on trust and safety, and determine product interventions to leverage opportunities and prevent threats” by the end of Q3. This showed that generative AI was being treated not only as a technical development, but as a governance and platform integrity issue.
Wikimedia’s own late-2025 phrasing captured the underlying principle neatly: “They keep knowledge human.” The “they” referred to volunteers who add citations, discuss wording, resolve disputes, and maintain standards. In that context, limiting AI authorship is less about technology fear than about protecting the social process that makes Wikipedia reliable.
AI Pressure Extends Beyond Writing
The AI debate around Wikipedia is also about infrastructure and access, not just prose quality. In the 2025,2026 annual plan, Wikimedia said that since 2024 it had seen a “dramatic rise in request volume,” with most of the increase coming from scraping bots collecting training data for AI workflows. The Foundation warned that “The load on our infrastructure is not sustainable and puts human access to knowledge at risk.”
To respond, Wikimedia set a concrete target: by the end of FY 2025,2026, 50% of requests to programmatic access channels should be attributable to a known developer or application. That goal reflects a broader attempt to place limits and accountability around automated use of Wikimedia resources in the AI era.
At the same time, Wikimedia has acknowledged why AI companies are so interested. In October 2025, the Foundation said Wikipedia is “one of the highest-quality datasets available,” which is why many generative-AI systems depend heavily on it for training. But it also called for stronger attribution standards so users can verify the sources behind AI-generated outputs.
Enforcement Is Already Happening
The new restrictions are not symbolic. Recent coverage from April 2026 described an AI-assisted editor account being banned after attempts to create and modify articles using AI-generated content. That case demonstrated that enforcement is active and that Wikipedia’s rules are being applied in practice.
The scale of the platform helps explain why this matters. Wikimedia said that in 2025, nearly 250,000 volunteers maintained English Wikipedia, while readers spent an estimated 2.8 billion hours reading English Wikipedia articles. On a project of that size, low-quality AI submissions create major costs, while strong editorial standards create enormous public value.
That is the context behind Wikipedia limits AI-written content. The encyclopedia is trying to preserve trust in a system built on citations, discussion, and careful human review. For Wikipedia, the challenge is not simply whether AI can write plausible sentences, but whether the community can maintain reliable knowledge at scale.
Wikipedia’s March 2026 decision is therefore best understood as a targeted defense of editorial integrity. The policy does not ban every use of AI, nor does it deny that AI tools may help with limited support tasks. Instead, it makes clear that writing and rewriting encyclopedia content must remain a human responsibility.
As generative AI becomes more powerful and more widespread, other knowledge platforms will likely face similar choices. Wikipedia’s answer, at least for now, is that trust depends on people: people who verify sources, challenge claims, negotiate neutrality, and accept accountability for what appears on the page. In the AI era, that human layer is exactly what Wikipedia is trying to protect.
