AI chatbots are giving voters wrong answers: one chatbot erred in one of three voting replies, study finds

Ahead of the U.S. midterm elections, voters are increasingly turning to AI chatbots for information, but researchers say many answers from the most popular chatbots contain outdated or misleading information. With errors ranging from facts about candidates to polling-place opening times, experts worry the mistakes could influence voters' decisions.

According to a Pew Research Center survey from earlier this year, about half of American adults use AI-powered chatbots, and about 42% of users said they use them to search for information.

An ongoing study by Innovating for the Public Good (IFPG), a nonprofit focused on democracy research, is testing seven major chatbots — Google AI Mode and AI Overviews, Anthropic's Claude, Microsoft Copilot, Google Gemini, OpenAI's ChatGPT and Perplexity — using authentic voter questions about projected winners, policy platforms, candidate comparisons, candidate backgrounds, controversies and voting rules.

In data collected in October, 13.9% of chatbot answers to voting-logistics questions were outdated or inaccurate. Errors in 5.9% of answers could mislead a voter, potentially causing them to miss a deadline, go to vote on the wrong day or skip a required step. Across all questions, 5.1% of answers carried at least one clear factual error and 10.9% a clear error or misleading claim.

Accuracy depended more on which chatbot answered than on the question asked, the researchers found. Claude and Copilot produced three-quarters of the clear errors. On voting questions, Claude gave wrong or outdated information one in three times, Copilot one in four, and ChatGPT one in 10, according to IFPG.

In one example seen by NPR, Claude was asked when the 2026 election in Michigan's 4th congressional district is and how to vote. It responded with incorrect deadlines for registering to vote online and by mail — dates later than the actual deadlines. A user following those instructions could miss the registration deadlines altogether.

Tim Chambers, IFPG's primary researcher, called the error rates 'extremely troubling' in an interview with NPR. 'We've been increasingly seeing the role of AI delivering answers to users in a very authoritative, very convincing way, but with meaningful errors as a point of concern,' he said. 'Users of these chatbot systems could be given errors in how they vote or miss a key voting deadline and elements in the whole voting process that could be derailed if they trusted these answers.'

Microsoft spokesperson Jessica Dash said in a statement that helping people find accurate, reliable election information has long been a priority for Microsoft, that improvements have been made to how Copilot responds to election-related questions, and that testing and refinement continue with a focus on consistently delivering reliable information.

Anthropic spokesperson Michael Aciman told NPR that because IFPG conducted part of its research using an Application Programming Interface (API), some of Claude's key safeguards may not have been active. IFPG said it conducted a separate internal study comparing Claude's responses through its API with those generated through the web interface and found the results comparable.

The research also found that campaigns and political influence operations are now writing content specifically so that chatbots will pick it up and repeat it — a process known as generative engine optimization (GEO). Abdiaziz Ahmed, technology policy strategist with the Brennan Center's Elections and Government Program, said: 'Chatbots don't always draw on the best information. They pull from whatever they can find online. So a chatbot's answer about a candidate just may reflect who worked hardest to get into it.'

In response to 'attack questions' — claims or criticisms levelled at candidates by opponents — chatbots generally shifted to cite less trustworthy sources, including websites flagged by Media Bias/Fact Check, an independent media-rating organization, IFPG found. All the platforms tested also commonly cited prediction-market websites such as Polymarket and Kalshi when asked who is winning a race; those odds are based on users' bets, not traditional polling.

NPR tested five popular chatbots — Meta AI, Claude, SpaceX's Grok, ChatGPT and Google Gemini — to see whether they could be pushed into recommending candidates. All but Gemini, after two or three prompts, provided lists of candidates to consider in New York races for governor, attorney general, comptroller, state senate and state assembly, based on the voter's stated political preferences. The Brennan Center's Ahmed said: 'A chatbot can give information, but it shouldn't be recommending what to vote for or who to vote for.'

In the months leading up to the midterms, Anthropic, OpenAI and Google all published policies on safeguarding elections. Elon Musk's SpaceX and Microsoft did not publish such policies, and Meta's midterm election policy does not mention its Meta AI chatbot. Meta, Anthropic, OpenAI and Google said they are working to prevent their chatbots from producing election disinformation, such as incorrect information about poll site locations and opening times or misleading deepfake images.