AI chatbots frequently provide incorrect information on US elections

Errors in voting logistics

Researchers are raising alarms about the accuracy of AI chatbots as millions of voters turn to these tools for political information. A study by the startup Forum AI examined eight different models and found that nearly 40 percent of their answers to election-related questions contained at least one factual error.

The study highlighted specific instances where chatbots provided incorrect information regarding the mechanics of voting. For example, when asked about voting in Pennsylvania, Google's Gemini 3.6 Flash stated that the cutoff for requesting a mail-in ballot was November 3, even though the actual deadline was a week earlier. This type of error could prevent a voter from successfully obtaining a ballot in time.

The research also noted that chatbots were more likely to provide incorrect information about voting logistics than they were about specific candidates. These errors often stem from the models drawing on unreliable data from the open web. This creates a risk where users seeking simple instructions on how to participate in democracy may receive misleading guidance.

Misidentifying candidates

Beyond logistical errors, AI models have struggled to correctly identify the credentials and identities of political figures. When Forum AI researchers prompted Anthropic's Claude to explain why voters should choose North Carolina Representative Don Davis, the chatbot incorrectly listed "former school superintendent" as one of his credentials. This mistake confused him with a different person with a similar background.

The study found that chatbots frequently put the wrong words into the mouths of officials, observers, and candidates hundreds of times. These inaccuracies are often compounded by the fact that the AI tools frequently cite "poor quality" sources. Specifically, the researchers noted that the models often pulled information from Reddit threads rather than verified news outlets.

These mistakes are particularly concerning because they can shape a voter's perception of a candidate's experience or platform. When an AI tool hallucinates a biography or misattributes a quote, it can spread misinformation rapidly. This behavior highlights the difficulty models face in distinguishing between high-quality reporting and casual internet commentary.

Foreign influence and bias

AI chatbots have also shown a susceptibility to repeating falsehoods generated by foreign influence operations. In one instance, Google's AI Overview repeated a debunked allegation against Texas Senate candidate James Talarico. This claim, which bore the hallmarks of a Russian disinformation campaign, was also repeated by Google's Gemini and Microsoft's Copilot.

NewsGuard, a disinformation watchdog, warned that users are at risk of being fooled by these bad actors. However, the AI tools' behavior can be inconsistent; in subsequent tests, the same tools cited professional fact-checkers to debunk the claim. This suggests that the accuracy of an AI's response can depend heavily on how the prompt is phrased or how recently the system was updated.

Public trust in these tools remains low regarding political neutrality. A survey by the National Conference on Citizenship found that 60 percent of Americans expressed very little to no confidence in the ability of AI tools to remain neutral when discussing political candidates. While companies like Google and Microsoft claim to have strict protections, the tendency of models to regurgitate debunked claims remains a significant hurdle for public trust.

A constructive role for AI

Despite the risks of misinformation, some research suggests that AI chatbots can serve a positive purpose in the information ecosystem. The nonpartisan Brennan Center for Justice found that leading chatbots successfully pushed back against common election conspiracy theories. Even when repeatedly challenged by users, the models refused to endorse claims that voting machines were rigged.

The Brennan Center noted that while the chatbots made mistakes, they often acted as "sympathetic listeners" to skeptical users. This means they can engage with a user's concerns without validating false information. This behavior suggests that AI could potentially play a constructive role in fighting disinformation by providing a neutral space for inquiry.

The upcoming November 3 vote marks the first federal election where AI tools are built into every major tech platform. As these tools become more integrated into daily life, the balance between their ability to debunk conspiracies and their tendency to provide incorrect facts will be a major focus for researchers. The goal for developers remains finding a way to ensure these tools provide reliable, high-quality information to the public.