Study Finds AI Chatbots Give Voters Wrong Answers, as Millions Ask Them How to Vote
Researchers testing seven popular chatbots found nearly 14% of voting-logistics answers were outdated or inaccurate, and most tools offered candidate lists when pressed.
On the eve of Election Day, a growing number of Americans are asking AI chatbots how to vote, who is winning and which candidates to back. New research says the answers they get are wrong often enough to cost someone a ballot.
A study by the nonprofit Innovating for the Public Good, conducted by the consulting firm Dewey Square Group, tested seven major products: Google AI Mode and AI Overviews, Anthropic's Claude, Microsoft Copilot, Google Gemini, OpenAI's ChatGPT and Perplexity. The researchers fed them authentic voter questions about projected winners, policy platforms, 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 In data collected in October, 13.9% of chatbot answers to voting-logistics questions were outdated or inaccurate.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 those answers could have misled a voter into missing a deadline, showing up on the wrong day or skipping a required step.
Across all question types, 5.1% of answers carried at least one clear factual error, and 10.9% contained a clear error or a misleading claim. "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," said Tim Chambers, a principal at Dewey Square Group.
Accuracy depended more on the chatbot than on the question. Claude and Copilot produced three-quarters of the clear errors. On voting questions, Claude gave wrong or outdated information one time in three, Copilot one in four and ChatGPT one in 10, according to the report. In one example NPR reviewed, Claude was asked when the 2026 election takes place in Michigan's 4th Congressional District and how to vote. It gave registration deadlines, online and by mail, that were later than the real ones.
Anthropic spokesperson Michael Aciman told NPR that because part of the research ran through an application programming interface, some of Claude's key safeguards may not have been active. The researchers said an internal comparison of Claude's API responses and its web responses found them comparable. A Microsoft spokesperson, Jessica Dash, said the company has improved how Copilot handles election questions and keeps testing it.
The study also found chatbots drift toward weaker sources when asked about attacks on candidates, citing sites flagged by the rating group Media Bias/Fact Check. Every platform tested often pointed users to prediction-market sites such as Polymarket and Kalshi when asked who was winning a race. "They would get pointed to betting market content to say who's winning rather than to pollsters," Chambers said. "This is very troubling because betting markets can be manipulated, and they're sometimes accurate but often not."
NPR ran its own test on five chatbots: Meta AI, Claude, Grok, ChatGPT and Gemini. After two or three prompts, all but Gemini produced lists of candidates to consider voting for in New York races for governor, attorney general, comptroller, state senate and state assembly, based on the voter's stated political preferences. "A chatbot can give information, but it shouldn't be recommending what to vote for or who to vote for," said Abdiaziz Ahmed, a technology policy strategist at the Brennan Center for Justice.
Ahmed also warned that campaigns and influence operations now write content designed to be picked up and repeated by chatbots, a practice called generative engine optimization. "Chatbots don't always draw on the best information. They pull from whatever they can find online," he said. "So a chatbot's answer about a candidate just may reflect who worked hardest to get into it."
The companies say they are working on it. Anthropic, OpenAI and Google published election-safeguard policies ahead of the midterms. Anthropic's states that "we have a responsibility to make sure the answers they get are accurate and balanced." SpaceX and Microsoft did not publish policies, and Meta's midterm policy does not mention its Meta AI chatbot, NPR reported.
The stakes are large because usage is. A Pew Research Center survey from earlier this year found about half of American adults use AI chatbots, and about 42% of users turn to them to look up information. With polls opening Tuesday, the researchers' advice is plain: check the state election office's website for deadlines and polling places instead of trusting a chatbot's answer.
