Artificial intelligence chatbots are feeding voters false election information, according to research from Innovating for the Public Good. The findings surface as millions of Americans increasingly turn to AI tools for political guidance.
The IFPG study, conducted by Dewey Square Group, tested seven major chatbots on authentic voter questions about voting rules, candidate backgrounds, policy platforms and registration deadlines. Data collected in October showed that 10.9 percent of all chatbot responses contained a clear factual error or misleading claim.
The errors were concentrated in voting logistics. On questions about registration and voting procedures, 13.9 percent of chatbot answers were outdated or inaccurate. In 5.9 percent of those answers, the errors could actively mislead a voter—potentially causing them to miss a deadline, vote on the wrong day or skip a required step.
Anthropoic's Claude produced the worst results. The chatbot gave wrong or outdated voting information one in three times. Microsoft's Copilot erred one in four times. OpenAI's ChatGPT performed better but still made errors one in 10 times.
In one case, Claude provided incorrect voter registration deadlines for Michigan's 4th congressional district in 2026, citing deadlines later than the actual ones. A user following the chatbot's instructions would miss the registration period.
Tim Chambers, a principal at Dewey Square Group, said the error rates were "extremely troubling." The problem, he said, is that users trust these responses as authoritative, even when they contain factual errors that could derail the voting process.
About half of American adults used AI-powered chatbots as of earlier this year, according to Pew Research Center. Approximately 42 percent of those users said they employed chatbots to search for information.
