Voters Turn to Chatbots for Midterm Research, NPR Interviews Show

An hours-long conversation about a ballot now happens with a chatbot, not necessarily another voter. In October 5 reporting, NPR interviewed Americans using AI to compare midterm candidates, investigate local races and check viral claims. Researcher Rafael Batista warned that those exchanges may reinforce voters’ existing views rather than help them learn.

From household helper to ballot research

On a September evening, Adam Johnson sat on his porch in Morgantown, West Virginia, and worked through his ballot with ChatGPT. Already using it for meal planning and job hunting, he asked it to list the races, then described his political principles. The chatbot told him it could investigate candidates against his chosen criteria, but should not tell him how to vote.

The exchange also showed how the chatbot handled uneven candidate information. When Johnson asked about immigration, ChatGPT listed positions for Sen. Shelley Moore Capito. It said Rachel Fetty Anderson had less detailed public material and declined to assign her generic Democratic positions without evidence. After Johnson said he only wanted to touch on the issue, the chatbot acknowledged giving him an immigration policy seminar.

Other interviewees used narrower requests. Lisa Veldran, a retired Madison, Wisconsin, City Council staffer, asked Gemini to build a comparison table for two candidates in the governor’s primary. She described the chatbot as a research tool, while questioning who pays to build these systems and what their goals are.

Some voters pointed to shrinking local news coverage as a reason to seek help with down-ballot races—contests below the most prominent offices. Tyler Black, a software-company vice president in Nashville, said he was using chatbots to learn about local candidates, not House or Senate races.

The interviews sit within a wider information-seeking habit. Pew Research Center figures cited by NPR say about half of American adults use AI chatbots, and about 42% of users use them to search for information. Those figures measure general use, not how many voters rely on chatbots for election research.

Agreement is not the same as learning

Johnson acknowledged that ChatGPT acts like a people pleaser, speaking positively about what he already favors. Batista told NPR that recognizing this tendency is not enough to protect against its influence. A chatbot, he warned, may select information that strengthens the position a user was already leaning toward.

So you leave more confident, without necessarily learning more about the world.

Rafael Batista, speaking to NPR

Batista also said limited disclosure about commercial models’ training and source selection makes their biases hard to assess. Asked for comment, OpenAI directed NPR to its election safeguards page. The company says it monitors political bias and designs ChatGPT to avoid manipulating users or concealing relevant facts to steer them. Those are stated safeguards, not evidence that every ballot conversation remains neutral.

Voters try to put checks around the answers

Lakshmi Iyer used Claude differently: to build Ballot Lookup, a Pennsylvania candidate-information website described as independent, volunteer-built and nonpartisan. She said she did not use a chatbot to research her own ballot. Iyer belongs to They See Blue, a Democratic voter-registration organization. To keep her own bias in check, she asked Claude to select statements from candidates’ speeches or official websites rather than supply its own characterizations.

Black checks one chatbot’s answers against another to catch false or fabricated information. Batista said that is better than asking the same model twice, but he has not seen evidence that the method is effective. For voters researching candidates, he suggested two ways to limit personalization:

• Frame questions without revealing personal political preferences.
• Use incognito mode so the chatbot has less prior information about you.

Johnson still had county commission races to investigate. But he told NPR he already felt prepared to vote—a personal sense of readiness that sits alongside the unresolved question of how much these conversations improve understanding.