40% of chatbot election answers had an error, study finds

Forum AI, a startup that evaluates AI models, tested eight chatbots on election questions and found almost 40 percent of answers contained at least one factual error — with logistics answers wrong more often than candidate answers.

Claude misidentified a North Carolina congressman's credentials, Gemini gave the wrong Pennsylvania mail-in ballot deadline, and Google's AI Overview, Gemini and Microsoft's Copilot repeated a debunked AI-altered smear clip about a Texas Senate candidate when prompted — two days before the November 3 midterms.