AI Chatbots Fuel US Election Disinformation

Mashiur Rahman, New York – Concerns are intensifying over inaccurate and misleading information generated by artificial intelligence (AI) chatbots ahead of the upcoming U.S. midterm elections. Researchers warn that an influx of logistical voting errors, candidates’ identity confusion, and the recycling of Russian-backed propaganda is systematically corrupting the digital information landscape.

The November 3 federal election marks the first time that major tech platforms are deploying direct, embedded AI features at scale. This rollout comes at a critical time as millions of American voters are actively turning to AI tools for election-related instructions, procedures, and political research. However, multi-platform evaluations indicate that these chatbots frequently rely on low-quality, open-internet sources—including public forums like Reddit—to formulate responses, thereby polluting an online ecosystem already saturated with disinformation.

A recent study conducted by Forum AI, an artificial intelligence evaluation firm, revealed that nearly 40% of election-related responses from major chatbots contained at least one factual error. Surprisingly, the models were significantly more prone to hallucinating voting administrative procedures and logistics than information regarding candidates’ platforms.

For example, Anthropic’s chatbot, Claude, confused North Carolina Congressman Don Davis with another individual, incorrectly describing him as a former school superintendent. Furthermore, researchers have documented instances where major AI engines completely miscalculated voter deadlines, providing advice that could cause a voter to request a ballot too late to cast a legitimate vote.

Beyond localized errors, researchers are heavily alarmed by how vulnerable AI models are to foreign information operations. In late October, Russian influence operations, notably Storm-1516, launched a deepfake campaign targeting the Texas Senate race involving Democrat James Talarico. The operation circulated an AI-generated video under a fraudulent account named “Zyon Lewis,” in which a fictionalized former student baselessly accused Talarico of sexual misconduct.

When the independent watchdog organization NewsGuard queried major search engines and tools, Google’s AI Overview, Gemini, and Microsoft’s Copilot all repeated the fabricated allegation as factual, drawing content from unverified online echo chambers.

Microsoft later addressed the error, explaining that the situation was developing rapidly amid a localized shortage of reliable data, causing Copilot to initially hallucinate the narrative. Microsoft noted that the model has since been updated to identify the claim as entirely fabricated. Google also maintained that it enforces strict quality-control parameters and specific guidelines to filter out false election-related claims across its AI features.

Despite the systemic errors, research shows that AI chatbots are not entirely counterproductive. According to the non-partisan research body Brennan Center for Justice, popular AI platforms have generally refused to endorse widespread electoral conspiracy theories, such as claims that voting machines are pre-rigged. This indicates that chatbots can serve as a baseline defense against certain institutional rumors.

Voter trust, however, remains shaky. A recent survey conducted by the National Conference on Citizenship indicated that while two out of every five Americans utilize AI tools to check ballot measures and candidate information, only 40% trust the AI to deliver correct times, locations, and procedures for physical voting. Furthermore, 60% of respondents stated they have little to no confidence that AI tools can provide neutral information about political candidates.