Study Finds 29% of Chatbot Voting Replies Inaccurate

AI chatbots are providing unreliable information about elections and voting procedures, raising concerns about voter access and trust. A study by the Institute for Strategic Dialogue found that 29% of responses to questions about registration, mail-in voting and election dates were incomplete, inaccurate or outdated, while another 16% were technically correct but omitted important guidance. Errors included wrong deadlines and requirements, and performance was substantially worse for Spanish-language questions. Among the systems tested, GPT-5.5 performed best, while several others produced outdated or incomplete answers. Separately, University of Arizona researchers found that leading AI models can reinforce misinformation during extended conversations, particularly on obscure topics, underscoring the need for users to verify chatbot responses with current official sources.