Tests Show Chatbots Backed Opposing Parties in Midterm Races

Studies and reporting ahead of the U.S. midterms raise concerns about voters relying on AI chatbots for political information. In a test of nearly 500 races, Grok recommended Republican candidates in 92% of responses and GPT recommended Democrats in 97%, with the models often drawing on different topics and sources. Separately, an evaluation of eight AI models found that nearly 40% of answers to election-related questions contained at least one factual error, including incorrect voting deadlines that could cause voters to miss their chance to cast a ballot. Researchers also warn that chatbots can offer a definitive-sounding answer based on uneven or low-quality sources, even when users may not know enough to assess its accuracy. Reporting on voter guides and election-information tools underscores the need for voters to verify chatbot answers with reliable sources, particularly for practical details such as deadlines and procedures.