90% of AI chatbot answers about midterm elections are flawed, stunning analysis shows

A study by Forum AI, released on June 5, 2026, found that approximately 90% of answers provided by four popular AI chatbots about the upcoming midterm elections were factually incorrect, biased, or cited state-run media. The researchers described the findings as ‘stunning’ in their assessment of the chatbots’ reliability.

The study evaluated chatbots from unnamed companies, focusing on accuracy, bias, and source credibility. Researchers found that the majority of answers contained factual errors or relied on questionable sources, including state-run media outlets. The specific errors included misstatements about voting procedures, candidate positions, and election dates.

The findings highlight ongoing concerns about the use of AI chatbots for disseminating civic information. Forum AI has called for increased transparency and rigorous testing of AI systems before deployment in contexts that could affect democratic processes. The study’s authors emphasised the need for improved safeguards to prevent the spread of misinformation during election cycles.