As the 2026 U.S. midterm elections approach, artificial intelligence (AI) chatbots are increasingly being used by voters seeking election information. However, recent analysis has revealed significant inaccuracies in the voter guides generated by these AI systems, raising concerns about their reliability.
A comprehensive test was conducted in late September, in which six widely used AI chatbots were prompted to produce voter guides for randomly selected addresses across all 3,100 U.S. counties. The request sought comprehensive listings of races and candidates for the November 2026 ballot. The results showed that many chatbots frequently provided incorrect data, including misidentifying congressional districts and listing candidates for races not on the ballot.
For instance, ChatGPT included a Missouri U.S. Senate race that is not scheduled for this year, and Anthropic’s Claude mistakenly named local Wyoming officeholders as federal candidates. About 48 percent of the guides failed to specify a correct congressional district, and among those that did, 22 percent contained errors, potentially misleading voters about who is running in their district. Chatbots such as Claude and Amazon Alexa showed the highest error rates, while ChatGPT, Google AI Mode, and Perplexity tended to omit district information more often rather than provide inaccurate details.
The inaccuracies were partly traced to the underlying sources from which chatbots drew their data. AI-generated websites have become a prominent information reservoir for these systems. One such site, the Ballot Brief—founded by tech veteran Ben Torres Ezrick—was cited in nearly a quarter of ChatGPT’s election-related responses and in the majority of Microsoft Copilot’s answers. The Ballot Brief compiles candidate data largely through AI tools and draws from official government sources, but several errors were detected, including conflated candidate profiles and incorrect legislative records. Ezrick acknowledged the mistakes and said improvements were made after being informed of the issues.
Experts warn that this AI feedback loop, where AI chatbots rely on content generated or summarized by other AI-driven platforms, can amplify errors and distort the election information voters receive. Isabel Linzer, an elections analyst at the Center for Democracy and Technology, noted the risk of chatbots citing information published by relatively unknown AI-generated sites and the potential impact on voter trust.
Other small websites similarly relying on AI, such as DecodeTheVote and Smarter.Vote, have also become frequent sources for chatbots. Smarter.Vote's founder reported millions of citations from AI bots, attributing the popularity to the site's practice of citing its sources clearly—a trait appreciated by AI systems.
Major AI companies have acknowledged the problems and taken steps in response. OpenAI and Anthropic have partnered with Democracy Works to integrate official voting information, and OpenAI said it updates its models when errors are flagged. Microsoft has also adjusted Copilot to prioritize authoritative state election sites. Amazon paused Alexa’s voter guide capabilities after identifying the cause of some inaccuracies. Google maintains that its AI Mode cross-checks generated claims against trusted sources.
Despite added disclaimers and links to official state election tools appearing in chatbot responses, experts suggest that many users may not notice or follow these cautions. Jevin West of the University of Washington's Center for an Informed Public emphasized the importance of consulting official sources directly, as chatbots are currently far from dependable for election information.
The increased reliance on AI for political data ahead of a crucial election underscores the challenges in ensuring accuracy and transparency in automated information dissemination. As AI-driven platforms evolve, the interplay between AI-generated content and AI-assisted retrieval remains a critical area for ongoing scrutiny.