Blab AI
Wikipedia Rejects Direct AI Editing Amid Growing Bot Traffic and Hallucination Concerns
When Jimmy Wales, the co‑founder of Wikipedia, spoke to AFP on Monday, he made a decisive statement: the encyclopedia will not allow artificial intelligence to edit its articles directly.
The decision follows a series of high‑profile incidents in which large language models produced fabricated information with the confidence of an expert—an effect known in the field as hallucination.
Wales said newer AI models have reduced hallucinations, but the problem remains "very, very bad," and therefore unacceptable for the volunteer‑edited content that Wikipedia relies on.
In March 2026, the English Wikipedia community adopted a policy that prohibits the use of large language models to generate or rewrite article content.
The new rules allow only copy‑editing of one’s own writing and machine translation from other language editions, after a period of experimentation in which AI‑generated articles were nominated for speedy deletion.
The ban reflects the community’s concern that AI output can violate core content standards and introduce inaccuracies that undermine the encyclopedia’s reliability.
Wales also noted that AI agents could still play a useful role by alerting the community to niche news items that might otherwise be missed.
He said the platform would not let AI edit directly because "you can’t really trust it enough," but he saw potential for the technology to surface information that human editors might overlook.
The idea is that AI could scan vast volumes of new content, flagging items for human review and thereby complementing the volunteer editorial process.
The rise of AI platforms that rely on Wikipedia’s content has increased bot traffic to the site.
According to reports, AI bots now account for a growing share of Wikipedia visits, while human traffic has fallen by about eight percent.
The encyclopedia’s traffic mix is shifting because AI tools use Wikipedia as a knowledge base to answer user queries.
Despite the change in traffic composition, Wales described the drop in human visits as "meaningful" but not a disaster.
Wikipedia’s revenue model is based on small donations from readers and grants, not on pageviews, so the decline in human traffic does not directly threaten the foundation’s finances.
Wales encouraged AI companies to "pay their fair share" because the high volume of requests from AI systems places a real cost on the Wikimedia Foundation’s servers.
The foundation has already signed agreements with several technology giants that use Wikipedia data.
"We’re starting to block the ones who aren’t behaving themselves," Wales said, indicating that the foundation is prepared to restrict access from non‑cooperative actors.
The policy shift reflects a broader effort to protect the integrity of Wikipedia’s content while managing the infrastructure demands of AI usage.
Wikipedia continues to monitor AI activity and is working with technology companies to ensure fair compensation for the resources consumed.
While AI bots are increasingly using Wikipedia as a knowledge source, the foundation’s donation‑based funding model and its policy controls aim to preserve the quality and reliability of the encyclopedia.
The situation remains dynamic, with ongoing discussions about how to balance AI innovation with the need for trustworthy, community‑curated information.
Hallucinations arise when a model generates plausible but unverified statements, a risk that is magnified when the content is presented as fact. For a volunteer‑driven encyclopedia, even a single fabricated sentence can erode trust.
The March 2026 policy was the culmination of a trial period during which editors tested AI‑generated drafts in a controlled environment. Drafts that failed to meet verifiability or neutrality standards were quickly flagged for deletion.
The decision to block large‑language‑model generation also signals a broader trend in the open‑source and community‑maintained ecosystems, where the emphasis remains on human accountability and editorial oversight.
Looking ahead, the Wikimedia Foundation plans to refine its usage metrics and negotiate more granular data‑sharing agreements with AI providers, aiming to align infrastructure costs with the benefits delivered to the public.
The decision follows a series of high‑profile incidents in which large language models produced fabricated information with the confidence of an expert—an effect known in the field as hallucination.
Wales said newer AI models have reduced hallucinations, but the problem remains "very, very bad," and therefore unacceptable for the volunteer‑edited content that Wikipedia relies on.
In March 2026, the English Wikipedia community adopted a policy that prohibits the use of large language models to generate or rewrite article content.
The new rules allow only copy‑editing of one’s own writing and machine translation from other language editions, after a period of experimentation in which AI‑generated articles were nominated for speedy deletion.
The ban reflects the community’s concern that AI output can violate core content standards and introduce inaccuracies that undermine the encyclopedia’s reliability.
Wales also noted that AI agents could still play a useful role by alerting the community to niche news items that might otherwise be missed.
He said the platform would not let AI edit directly because "you can’t really trust it enough," but he saw potential for the technology to surface information that human editors might overlook.
The idea is that AI could scan vast volumes of new content, flagging items for human review and thereby complementing the volunteer editorial process.
The rise of AI platforms that rely on Wikipedia’s content has increased bot traffic to the site.
According to reports, AI bots now account for a growing share of Wikipedia visits, while human traffic has fallen by about eight percent.
The encyclopedia’s traffic mix is shifting because AI tools use Wikipedia as a knowledge base to answer user queries.
Despite the change in traffic composition, Wales described the drop in human visits as "meaningful" but not a disaster.
Wikipedia’s revenue model is based on small donations from readers and grants, not on pageviews, so the decline in human traffic does not directly threaten the foundation’s finances.
Wales encouraged AI companies to "pay their fair share" because the high volume of requests from AI systems places a real cost on the Wikimedia Foundation’s servers.
The foundation has already signed agreements with several technology giants that use Wikipedia data.
"We’re starting to block the ones who aren’t behaving themselves," Wales said, indicating that the foundation is prepared to restrict access from non‑cooperative actors.
The policy shift reflects a broader effort to protect the integrity of Wikipedia’s content while managing the infrastructure demands of AI usage.
Wikipedia continues to monitor AI activity and is working with technology companies to ensure fair compensation for the resources consumed.
While AI bots are increasingly using Wikipedia as a knowledge source, the foundation’s donation‑based funding model and its policy controls aim to preserve the quality and reliability of the encyclopedia.
The situation remains dynamic, with ongoing discussions about how to balance AI innovation with the need for trustworthy, community‑curated information.
Hallucinations arise when a model generates plausible but unverified statements, a risk that is magnified when the content is presented as fact. For a volunteer‑driven encyclopedia, even a single fabricated sentence can erode trust.
The March 2026 policy was the culmination of a trial period during which editors tested AI‑generated drafts in a controlled environment. Drafts that failed to meet verifiability or neutrality standards were quickly flagged for deletion.
The decision to block large‑language‑model generation also signals a broader trend in the open‑source and community‑maintained ecosystems, where the emphasis remains on human accountability and editorial oversight.
Looking ahead, the Wikimedia Foundation plans to refine its usage metrics and negotiate more granular data‑sharing agreements with AI providers, aiming to align infrastructure costs with the benefits delivered to the public.
