Chatbots gave inaccurate answers to US election questions

France 24 described the Nov. 3 vote as the first federal election in which millions of voters actively relied on AI tools for political information.[1] The Institute for Strategic Dialogue characterized it as the first US election with AI tools built into every major tech platform.[1] A National Conference on Citizenship survey found that two in five Americans had used AI tools to search for election information.[1] France 24 reported that chatbots can help debunk conspiracy theories and educate voters, but studies found they often draw on unreliable open-web sources.[1] Those findings put the quality of routine answers, as well as political claims, in focus.

Almost 40 percent of chatbot answers to election-related queries contained at least one factual error.[1] In a study examining eight models, chatbots answered voting-logistics questions incorrectly more often than questions about candidates.[1] One chatbot gave Nov. 3 as the deadline to request a Pennsylvania mail-in ballot, although the actual deadline was a week earlier.[1] The report also described a chatbot that listed “former school superintendent” among a person’s credentials after confusing him with someone else.[1] For voters and teams helping them, a confident-sounding answer about a deadline or eligibility should not be treated as confirmation.

A likely AI-altered clip purported to show a former student of James Talarico accusing him of sexual misconduct.[1] When NewsGuard prompted them, Google’s Gemini and Microsoft’s Copilot repeated the false claim.[1] AFP later found that AI tools cited professional fact-checkers after the allegation had been widely debunked, illustrating that chatbot responses can be inconsistent.[1] Google said the issue had been addressed after its systems updated to surface the latest reporting, while Microsoft said Copilot later reflected updated context and identified the claim as fabricated.[1] These examples show why a chatbot’s answer may depend on when it is asked.

NewsGuard said users risk being fooled by foreign influence operations and other bad actors, based on a separate study of 10 AI tools.[1] It also tested five false narratives pushed by the Pravda network and found chatbots repeated them as fact in half of cases.[5] In a separate test, The Washington Post asked six popular chatbots to create voter guides for addresses in every US county; the guides invented a Missouri Senate race and misidentified candidates, including a House candidate who was not on the ballot.[2] The picture was not uniformly negative: Brennan Center research found leading chatbots resisted common election conspiracy theories, and the center said they could play a constructive role in fighting election disinformation.[1]