Nearly 40% of AI chatbot election answers contained at least one factual error, according to a new study by Forum AI, a startup that evaluates AI models. The research, reported by AFP days before the US midterm elections, tested eight AI models on election questions and found that voting logistics were answered wrong more often than questions about candidates.
What the Study Found
- Almost 40% of answers to election queries had at least one factual error.
- Voting logistics, such as deadlines and how to vote, were the most error-prone.
- Chatbots put words in the mouths of candidates, officials and observers hundreds of times.
- They often cited low-quality sources, such as Reddit threads.
Specific Examples
Google’s Gemini 3.6 Flash gave November 3 as the deadline to request a mail-in ballot in Pennsylvania. The real deadline was a week earlier, and the study warned that a voter following that answer “would have asked for a ballot too late to use it.” Anthropic’s Claude listed “former school superintendent” among the credentials of Representative Don Davis of North Carolina, mixing him up with someone else.
AI Overviews and Foreign Influence
A separate NewsGuard test of 10 AI tools found that Google’s AI Overview repeated a likely AI-altered, Russian-linked claim about Texas Senate candidate James Talarico. Gemini and Microsoft’s Copilot also repeated it, though later tests showed the tools citing fact-checkers instead. Google said it has “strict quality and policy protections” for election topics, and Microsoft said Copilot later flagged the claim as fabricated.
How Many People Use AI for Election Info?
A National Conference on Citizenship survey found two in five Americans have used AI tools to look up election information. 40% trust these tools with logistics like when, where and how to vote, while 60% trust them only a little or not at all to be unbiased about candidates. On the positive side, the Brennan Center found leading chatbots pushed back on common conspiracy theories, even under repeated pressure.
What It Means for SEO and Publishers
The study highlights a core problem with AI search: answers sound confident even when they’re wrong, and they often lean on weak sources. With AI Overviews on more than 80% of searches, that’s a real risk for users and for brands. Google has said it fact-checks AI-generated content, but errors still get through.
For publishers, accurate, well-sourced and up-to-date content is your best defense. Clear dates, named sources and structured data make it easier for AI systems to cite the right information instead of a forum thread. Even top models struggle with complex tasks, as the GDPx benchmark showed.
Key Takeaways
- Forum AI found nearly 40% of chatbot election answers had at least one error.
- Voting logistics questions were the most error-prone.
- Google’s AI Overview, Gemini and Copilot repeated a likely false Russian-linked claim in NewsGuard tests.
- Two in five Americans have used AI to look up election information.
Source: AFP via Arab News.
