AI chatbots give false US election answers, studies say

AI chatbots are giving voters misleading or false information before the November 3 US midterm elections, including wrong voting deadlines and inaccurate candidate details, researchers say.

Forum AI, a start-up that evaluates AI models, found that almost 40 percent of chatbot answers to election-related questions contained at least one factual error. The study examined eight models and found that chatbots made more mistakes on voting logistics than on candidate questions.

In one example cited by the study, Google's Gemini 3.6 Flash gave November 3 as the deadline to request a mail-in ballot in Pennsylvania, although the actual deadline was a week earlier. Forum AI said a voter following that answer would have requested a ballot too late to use it.

Researchers also raised concerns that chatbots can repeat foreign influence material. NewsGuard said Google AI Overview, Google's Gemini and Microsoft's Copilot repeated an unfounded allegation about Texas Senate candidate James Talarico after being prompted with a claim from a likely AI-altered clip. AFP fact-checkers had debunked the allegation, which the article says bore hallmarks of a Russian disinformation campaign.

Google said the issue identified by NewsGuard was addressed as its systems updated to surface later reporting. Microsoft said the initial query came during a fast-developing situation with limited reliable information and that Copilot later reflected updated context and identified the claim as fabricated.

A later AFP test found the AI tools cited professional fact-checkers after the claim had been widely debunked, showing that chatbot answers can change.

A survey by the nonpartisan National Conference on Citizenship found that two in five Americans had used AI tools to search for election information. Only 40 percent said they trusted the tools with logistical voting information, and 60 percent said they trusted them only a little or not at all to be unbiased about candidates.

Research by the Brennan Center for Justice found a more positive pattern on some questions: leading chatbots often refused to endorse common election conspiracy theories, including claims that voting machines were rigged, even when challenged repeatedly.