AI chatbots are being used by growing numbers of US voters to research candidates and ballot measures ahead of the midterm elections, prompting concerns that the technology may reinforce existing political views while appearing neutral.
Adam Johnson, 40, spent hours discussing his ballot with ChatGPT from his home in Morgantown, West Virginia. He asked the chatbot to list the races he would be voting in before examining candidates and issues in greater detail.
Johnson, a graphic designer studying to become an illustrator, already uses ChatGPT for meal planning, job searches and household tasks. He said the chatbot had become a significant part of his life.
When Johnson questioned whether ChatGPT would discuss his political principles, the chatbot told him it could compare candidates and evidence according to criteria he chose, while avoiding attempts to manipulate his views or tell him how to vote.
About half of American adults say they have used AI-powered chatbots, while 42 per cent of users say they use them to search for information. Voters have described using the tools to investigate lesser-known races, compare candidates and check claims circulating online.
But Rafael Batista, a fellow at John Hopkins University who studies how AI influences people’s experience of the world, said chatbots could make users more confident in views they already held without necessarily improving their understanding.
“It might select some things that would reinforce and persuade you even more towards the way that you were leaning already,” he said. “So you leave more confident, without necessarily learning more about the world.”
Batista said the lack of public information about how commercial AI models are trained, and how they select the sources used in their answers, made it difficult to assess what biases might influence their responses.
How voters are using AI chatbots
Johnson told ChatGPT that he generally voted Republican, but would support Democrats or a third-party candidate where Republican contenders belonged to the Make America Great Again movement.
As he considered a Senate race, he asked about immigration and refugee policy. The chatbot provided a detailed comparison of border enforcement, illegal immigration, asylum, legal immigration and refugee resettlement before acknowledging that it had turned a brief request into an “immigration policy seminar”.
Johnson said he understood that chatbots could be overly agreeable. “ChatGPT is just kind of a people pleaser,” he said. “Whatever you lean towards, it will speak positively about it.”
Lisa Veldran, 65, a former Madison City Council staffer, used Google’s Gemini to compare two candidates in Wisconsin’s primary for governor after Lieutenant Governor Sara Rodriguez withdrew from the contest.
Ms Veldran said the chatbot was useful for research, but not something that would determine her decisions. “You also have to keep in mind what’s running these chatbots and AI in general,” she said. “Who’s paying to build these bots and what is their ultimate goal.”
Lakshmi Iyer, an author in Exton, Pennsylvania, used Claude to create Ballot Lookup, a volunteer-built, non-partisan tool intended to help Pennsylvania voters identify candidates and their positions.
Ms Iyer said she had not used a chatbot to research her own ballot, but believed human oversight remained essential. To limit the risk of bias, she instructed Claude to draw on candidates’ own speeches and official websites.
Tyler Black, a software company vice-president from Nashville, Tennessee, said he was using chatbots to learn about local candidates, although he had not relied on them for the House or Senate races.
Mr Black said he checked answers from one chatbot against another because he knew the systems could produce false or fabricated information. Batista said that approach was preferable to asking the same model to check itself, but there was no evidence it was consistently effective.
Batista advised voters who use chatbots to conceal their personal preferences when framing questions and consider using incognito mode, so the system has less prior information about them.
OpenAI said it continued to monitor bias in its models in an effort to keep ChatGPT’s political responses neutral. Anthropic, which owns Claude, and Google, which owns Gemini, did not respond to requests for comment.
Johnson remains focused on researching the county commission races on his ballot. He said the chatbot had helped him feel prepared to vote, adding: “I wouldn’t feel too bad if I had to do the election today.”
