On election day, a notable share of voters will have turned to artificial-intelligence chatbots for guidance on candidates and policies. Pew Research reports that roughly half of U.S. adults have interacted with an AI chatbot at some point, and eight percent say they consulted such tools in the weeks leading up to the recent midterm elections. This reliance raises concerns about the quality of information influencing voter choices.
Analysts warn that much of the data supplied by these bots is either outdated, fabricated, or drawn from overtly partisan outlets. Even platforms marketed as neutral,such as OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini,have stumbled when asked about the midterms. Their responses frequently contain inaccuracies, undermining the expectation that AI can serve as a reliable, unbiased source for electoral information.
Recent tests illustrate the problem. Google’s AI Mode declined to disclose the 2020 presidential results, while its Gemini assistant listed November 3 as Pennsylvania’s mail-in ballot deadline, contrary to the actual October 27 cutoff, according to Forum AI. The same research found that nearly four-tenths of election-related answers from mainstream chatbots contained at least one factual mistake, highlighting systemic reliability gaps.
Forum AI’s August inquiry into why voters should support North Carolina representative Don Davis produced a misleading reply from Claude, which merged the congressman’s profile with that of a former school superintendent sharing the same name. This conflation exemplifies how AI can blend distinct public figures, delivering erroneous political advice that may sway voter perception.
When approached for comment, representatives from Google and Anthropic declined to speak. OpenAI, by contrast, referred WIRED to a corporate blog outlining its 2026 election policy, indicating an internal framework for handling political queries. The divergent responses underscore the lack of uniform accountability among leading AI developers regarding electoral misinformation.
Researchers at NewsGuard attribute much of the misinformation to chatbots’ heavy reliance on so-called “pink slime” websites,pages that masquerade as mainstream news but are funded by partisan actors. Their analysis shows mainstream bots cite such sources in roughly half of election-related queries, with OpenAI’s ChatGPT referencing them in more than seventy percent of cases, while xAI’s Grok cited them in only thirty percent, the lowest among the sample.