AI Chatbots Offer Voter Advice, but Errors Could Mislead Election Goers
Researchers find that popular AI chatbots often give outdated or incorrect voting information, raising concerns ahead of the midterms.
Widespread Use of Chatbots Ahead of Midterms
According to a Pew Research Center survey conducted earlier this year, about half of American adults reported using AI‑powered chatbots, and approximately 42% said they use them to search for information. This growing reliance has prompted researchers to examine how these tools handle voter‑related queries.
Study Finds Notable Error Rates
Innovating for the Public Good (IFPG), a nonprofit focused on democracy research, partnered with Dewey Square Group to test seven leading chatbots—Google AI Mode, AI Overviews, Anthropic’s Claude, Microsoft Copilot, Google Gemini, OpenAI’s ChatGPT, and Perplexity. The October data set included authentic voter questions about projected winners, policy platforms, candidate comparisons, backgrounds, controversies, and voting rules.
For questions about voting logistics, 13.9% of chatbot answers were outdated or inaccurate. More concerning, 5.9% of those answers could mislead a voter, potentially causing missed registration deadlines, voting on the wrong day, or skipping required steps. Across all question types, 5.1% of answers contained at least one clear factual error and 10.9% included a clear error or misleading claim.
Platform‑Specific Performance
The study notes that error frequency depended more on the chatbot than on the question asked. Claude and Copilot together produced three‑quarters of the clear errors. On voting‑logistics queries, Claude was wrong or outdated one in three times, Copilot one in four times, while ChatGPT erred once in ten responses.
An NPR‑reported example shows Claude answering the prompt “When is the 2026 election in Michigan’s 4th congressional district and how do I vote?” with registration deadlines that were later than the actual dates, a mistake that could cause users to miss the deadline.
Industry Reactions
Tim Chambers, a principal at Dewey Square Group, described the error rates as “extremely troubling” and warned that “the fact that there are just legions of errors around makes you wonder how much more there is to find.”
Microsoft’s spokesperson Jessica Dash said the company is focused on consistently delivering reliable information and has made improvements to Copilot’s election‑related responses. Anthropic’s Michael Aciman noted that IFPG’s API‑based testing might have bypassed some of Claude’s safeguards, though a separate internal comparison found similar results between API and web‑interface outputs.
Meta, Anthropic, OpenAI, and Google all stated they are working to prevent disinformation about elections, including incorrect poll‑site details and misleading deep‑fake images, continuing efforts that began after the 2016 Russian influence campaign.
Implications for Voters
With a sizable portion of the electorate turning to AI for information, the presence of outdated or misleading answers poses a real risk to the voting process. Errors that affect registration deadlines or polling‑site logistics could directly impact voter turnout and the accuracy of the democratic process.
As the midterms approach, the findings underscore the need for voters to verify chatbot‑provided information against official sources.
