Voters are using AI to research ballots. What do the studies show?

A familiar tool becomes a ballot-research aid

NPR spoke with Adam Johnson, 40, of Morgantown, who already used ChatGPT for everyday tasks before turning to it for election research. Other interviewees described comparing candidates or investigating local contests that received less news coverage. [1]

The report also raised a concern about agreeable answers: a chatbot can reinforce a user’s existing views while making the user feel better informed. That warning does not establish that every chatbot conversation persuades someone, or that a second chatbot provides an independent factual check. [1]

What the usage numbers measure

Pew Research Center’s June 17 report found that 49% of U.S. adults said they used AI chatbots, including 24% who used them daily. Those figures concern general chatbot use, not election advice or candidate selection. [2]

The survey covered 5,119 adults from February 17–23, 2026. Pew used its randomly recruited American Trends Panel, with online and telephone interviews and weighting to represent the adult population. The full-sample margin of sampling error was plus or minus 1.6 percentage points. [3]

Pew also reported that 42% of all U.S. adults used chatbots to search for information. That denominator is the adult population, not just chatbot users, and the question was not limited to politics. [2]

Accuracy and completeness are different tests

States United’s August 4 report examined 497 responses in six states during fall 2025, then 402 responses focused on Arizona, Michigan and Pennsylvania in January–February 2026. It tested free ChatGPT and Google AI Overview under different account conditions. [4]

Researchers coded no verifiable factual errors in the later round, but found substantial omissions: 88.9% of ChatGPT answers to gubernatorial-candidate questions were incomplete. Naming real candidates can still leave a voter with a partial list. [4]

The study was descriptive, not causal. It did not measure changed votes or represent every state, question or current model. A mid-study Google format change also prevented a clean separation of personalization effects from changes in how answers were displayed. Its results are a dated snapshot, not a permanent score for AI. [4]

Where official information fits

For registration, voting procedures and results, USAGov maintains a directory of state and local election offices. It gives readers a direct route to the officials responsible for their jurisdiction rather than relying on an AI-generated summary alone. [5]

The evidence here addresses three separate questions: how some people research ballots, how widely chatbots are used, and how selected systems answered a limited test. None of those findings supplies a reason to support a particular candidate. [1] [2] [4]