Wikipedia restricts AI use in articles

Wikipedia has updated its editing rules to ban editors from using large language models to generate or rewrite article content, while still allowing limited AI help for copyediting.

The policy now says, “the use of LLMs to generate or rewrite article content is prohibited,” replacing earlier guidance that warned against using LLMs to create new articles from scratch.

The change followed a vote among Wikipedia editors that passed 40 to 2, 404 Media reported.

Under the updated rules, editors may use LLMs to suggest basic copyedits to their own writing and apply them after human review, but the policy warns that LLMs can alter meaning or add unsupported information.

🔗 Source: TechCrunch

🧠 Food for thought

Implications, context, and why it matters.

Volunteer burnout from cleaning AI ‘slop’ helped force Wikipedia’s hand

The policy targets an uneven load on Wikipedia’s volunteer editors 1.
AI-written text can appear in seconds. Checking, fixing, or removing it can take hours for a human editor 1.
Editors tried other steps first. They set up WikiProject AI Cleanup (a volunteer effort focused on finding and fixing AI-generated content) and used a speedy deletion process for poorly written articles 2. A steady flow of “slop” pushed the community toward a clearer restriction after earlier proposals failed to reach consensus on implementation 3, 4.

Wikipedia is trying to protect the data that trains future AI systems

Wikipedia feeds training data into many AI models, so the rule aims to keep that source cleaner 1.
The ban seeks to stop a loop where large language model (LLM) hallucinations (made-up or incorrect statements presented as fact) enter the encyclopedia, get scraped by AI companies, then return in later model training as wrong information 1.
Wikipedia still favors human-written article content. It allows limited uses like copyediting one’s own work and translation assistance 5.
The decision offers a high-profile example of leaning on human review over automated scale, which could shape policies at other user-generated content sites 6.