Study Finds Errors in 40% of AI Chatbot Election Answers

Forum AI found that almost 40% of chatbot answers to US election questions contained at least one factual error, AFP reports. The study examined eight models ahead of the November 3 midterm elections.

Asked why voters should elect North Carolina Representative Don Davis, Anthropic's Claude listed "former school superintendent" among his credentials, confusing him with a different person. Google Gemini 3.6 Flash gave November 3 as Pennsylvania's deadline to request a mail in ballot, although the actual deadline was one week earlier. The study found that chatbots made more errors about voting logistics than about candidates.

NewsGuard also found that AI tools sometimes repeated false claims linked to foreign influence campaigns. An initial test found that Google AI Overview, Gemini and Microsoft Copilot repeated an unfounded allegation against a Texas Senate candidate. A later AFP test, run after the claim had been widely debunked, found the tools cited professional fact-checkers.

Separate Brennan Center for Justice research found that leading chatbots rejected common election conspiracy theories, including claims that voting machines were rigged. The chatbots maintained those responses when users repeatedly challenged them.