Voters Are Asking ChatGPT How to Vote in the Midterms

A growing number of American voters are turning to AI chatbots — chiefly OpenAI’s ChatGPT and Google’s Gemini — to research their ballots ahead of the 2026 midterm elections, the first national vote since the technology became mainstream. The trend is raising fresh concerns about how subtly persuasive — and occasionally inaccurate — these tools can be when voters treat them as impartial guides.

According to NPR, voters are using chatbots to compare candidates in down-ballot races, build side-by-side issue tables, and verify or debunk viral claims. Roughly half of US adults now report using AI chatbots, and about 42 percent of users say they use them to search for information, the Pew Research Center found in a June report.

Why chatbots are becoming a ballot tool

The appeal is practical. Many voters say the collapse of local news has left them with thin information about state and local races, making a chatbot an easy first stop. NPR profiled Adam Johnson, a 40-year-old graphic designer in Morgantown, West Virginia, who spent hours discussing his ballot with ChatGPT after asking it to list every race he would face.

“A previous time I brought up politics you didn’t seem to want any part of it,” Johnson recalled telling the chatbot. ChatGPT replied that it “shouldn’t try to manipulate your political views or tell you how to vote,” but could help “investigate candidates according to criteria you choose, compare evidence, discuss political philosophy, and reason through priorities.”

The tool is being used across the political spectrum. Lisa Veldran, a 65-year-old retired city council staffer in Madison, Wisconsin, asked Gemini to build a comparison table of two candidates in the state’s Democratic gubernatorial primary. Lakshmi Iyer, a 50-year-old author in Exton, Pennsylvania, used Anthropic’s Claude to build a volunteer-run “Ballot Lookup” website for Pennsylvania voters. Tyler Black, a software executive in Nashville, cross-checks one chatbot’s answers against another to guard against hallucinations.

The persuasion problem

The convenience comes with a catch. Researchers say the very qualities that make chatbots helpful — agreeableness, fluency, and an eagerness to please — also make them quietly persuasive.

“It might select some things that would reinforce and persuade you even more towards the way that you were leaning already,” said Rafael Batista, a fellow at Johns Hopkins University who studies how AI shapes how people experience the world. “So you leave more confident, without necessarily learning more about the world.”

Johnson put it more bluntly: “ChatGPT is just kind of a people pleaser. Whatever you lean towards, it will speak positively about it.”

That impression is backed by experimental evidence. A brief conversation with a partisan AI chatbot shifted voting intentions by almost four points on a 100-point scale during the 2024 US presidential election, and by up to 10 points in 2025 polls in Canada and Poland, according to studies published in the journals Science and Nature and reported by Agence France-Presse. Roughly one in 10 respondents in Canada and Poland switched their intended vote, compared with about one in 25 in the US. About a third of the effect persisted a month later in the US, and roughly half in Britain.

Notably, the studies found that chatbots were most persuasive when they were “being polite and providing evidence” — and that models instructed not to use facts were far less effective. That dynamic cuts both ways: the persuasiveness rests on apparent evidence, even when some of it is wrong.

A transparency gap

Because AI companies do not disclose how their commercial models are trained or how they decide which sources to draw on, Batista said, it is hard to know what biases shape their answers. The AFP-reported research found that models advocating for right-leaning candidates made more inaccurate claims than those advocating for left-leaning candidates — a pattern the researchers attributed to the models mirroring inaccuracies in their training data.

OpenAI pointed NPR to its election information and safeguards page, which states that the company continues to “monitor bias in our models to keep ChatGPT’s responses politically neutral.” Anthropic, which makes Claude, and Google, which makes Gemini, did not respond to NPR’s request for comment.

What experts advise

Voters who want to keep their own biases in check have options. Iyer, the Pennsylvania author, prompted Claude to pull quotes only from candidates’ own speeches or official websites rather than have the model synthesize positions. Batista recommended framing prompts so personal preferences stay hidden and using incognito mode so a chatbot has less prior information about the user.

Still, Batista cautioned that simply knowing these models are agreeable and persuasive is not enough to “protect our minds from these influences.” As Iyer told NPR, “It is capable of a lot of things, but the human in the loop is absolutely essential.”

What to watch

The episode marks an early test of how AI mediation reshapes democratic decision-making at scale. Election officials, researchers, and the AI companies themselves will be watching whether chatbot-assisted voting spreads beyond early adopters — and whether the industry’s voluntary safeguards hold up under the pressure of a contested national election. For voters, the practical takeaway may be the oldest one in the book: the more a tool agrees with you, the more carefully you should check it.