AI Chatbots Are Making Up US Election Races: Voters Could Pay the Price

Millions of Americans are turning to artificial intelligence for election information, but a new investigation has found that popular chatbots can invent political races, misidentify candidates and repeat inaccurate information from AI-generated websites.

A Washington Post analysis of more than 3,000 voter guides produced by six popular AI chatbots found significant errors in responses about the 2026 US midterm elections. The investigation tested ChatGPT, Anthropic's Claude, Google's AI Mode, Microsoft's Copilot, Perplexity and Amazon's Alexa+ using randomly selected addresses in every US county.

The findings come as more voters report using AI to research candidates. A National Conference on Citizenship survey found that 25% of US adults had used an AI chat programme to learn about Senate, House or gubernatorial candidates while deciding how to vote, up from 14% two weeks earlier.

The proportion who said they had used AI for at least one of four election-related tasks, including checking election stories and finding out how to vote, rose from 28% to 40%. The figures are based on self-reported use, and the survey's authors cautioned that the absolute levels should be treated as approximate.

Chatbots Invent Races and Misidentify Candidates

The Washington Post found that ChatGPT included a US Senate race in Missouri that was not on the 2026 ballot. Claude incorrectly identified Wyoming politicians Curt Meier and Ana Cordova as Senate candidates, even though Meier was running for re-election as state treasurer and Cordova was running for superintendent of public instruction.

Google's AI Mode also made mistakes when asked to generate a voter guide for a Maryland address. It placed the address in the wrong congressional district and incorrectly suggested that Rep. Steny Hoyer was facing a Republican challenger, despite his announcement that he would not seek re-election.

The district-level findings raise further concerns about the reliability of these tools. The chatbots failed to identify an address's congressional district in 48% of cases. Among the voter guides that did specify a district, 22% identified the wrong one. Claude and Alexa+ returned incorrect district information for more than one in five addresses.

ChatGPT, Google AI Mode and Perplexity appeared to perform better, but largely because they often did not provide a district at all.

'Right now, they're far from reliable,' Jevin West, co-founder of the University of Washington's Center for an Informed Public, told the Washington Post.

AI Chatbots Are Repeating Other AI's Mistakes

The investigation also exposed a troubling feedback loop: chatbots can repeat misinformation published by websites that rely on AI to compile election information.

In one example, ChatGPT cited myvotega.com when generating a voter guide for an address in Burke County, Georgia. The website incorrectly placed Burke County in the wrong congressional district, an error that ChatGPT picked up and repeated.

Another website, The Ballot Brief, became a frequently cited source for several AI systems despite errors in its voter information. The site uses AI tools to compile candidate profiles and election details. The Washington Post found that it had incorrectly mapped some ZIP codes to congressional districts, potentially passing those errors on to chatbots that relied on its information.

A follow-up investigation published on October 30 showed how widely the website had spread through AI-generated answers. In the newspaper's tests, ChatGPT cited The Ballot Brief in 22% of its answers to election-related questions, while Microsoft's Copilot cited it in 75%. After the findings were shared with OpenAI, ChatGPT cited the website in just 4% of answers in a second round of testing.

The follow-up also found errors in the website's candidate profiles. These included crediting Sen. Jon Husted with supporting a bill that had passed before he took office, as well as merging the profiles of two California politicians named Robert Garcia.

Tech Companies Respond to the Findings

The companies have pointed to safeguards and partnerships intended to improve election-related responses. OpenAI and Anthropic have partnered with Democracy Works to provide users with official voting information.

However, the investigations show that these measures have not eliminated the problem. Amazon said it had identified the cause of the errors found in Alexa+ and paused the chatbot's voter-guide feature. OpenAI said it takes action to improve responses when issues are brought to its attention.

Microsoft initially said it was looking into the findings. Following the investigation, it said it had updated Copilot to prioritise authoritative sources, including official state election websites.

Anthropic said it had pressure-tested its models and strengthened safeguards to provide reliable voting information. However, the follow-up investigation found that Claude, which had initially relied on established non-profit voter guides, began frequently linking to The Ballot Brief in the second round of testing. Google said its systems corroborate AI-generated claims against trusted data sources.

West also questioned whether disclaimers advising users to verify information would be enough to prevent them from accepting incorrect chatbot responses. 'They will claim something's true because they Googled it, or a chatbot told them,' he said.

Voters Should Check Official Election Sources

The findings highlight the risks of relying on AI chatbots for information that can affect how people participate in elections. Even detailed chatbot-generated voter guides can contain invented races, misidentified candidates or incorrect district information.

Voters should verify candidates, polling locations, registration requirements and ballot details through official state and local election websites rather than treating chatbot-generated voter guides as authoritative.