One in four U.S. adults have asked AI about candidates, survey finds

AI is becoming part of Americans’ ballot research without winning their confidence. One in four U.S. adults said they had used AI to learn about major-race candidates, up from 14% two weeks earlier. Yet 60% of election-related AI users doubted its impartiality, according to research from the National Conference on Citizenship with researchers at The Carter Center.

The comparison covers October 1–4 and October 15–18, 2026. Across four election tasks, the share reporting any AI use rose from 28% to 40%. Each task showed a statistically significant increase. The candidate question specifically asked about Senate, House and governor races as respondents decided how to vote.

The increase extended beyond those prominent races. By mid-October, respondents reported using AI to:

• Research local candidates or ballot measures: 25%, up from 16% in early October.
• Check the accuracy of election stories seen elsewhere: 25%, up from 16%.
• Find out when, where or how to vote: 21%, up from 11%.

People also reported trying AI for a wider range of election needs. The share saying they had used it for all four tasks rose from 5% to 10%. These figures count tasks tried during the election, not how often someone consulted a chatbot.

Checking a story is not the same as trusting AI

Trust depended on the question. In mid-October, 23% of adults trusted AI a great deal or quite a bit to provide complete election information. Confidence was higher for voting logistics, at 40%. None of the changes in high trust across the survey’s four trust questions was statistically significant.

Open-ended answers offered a different view of what people actually asked. About a third of users described candidate questions, and about a quarter described voting logistics. Fewer than one in ten described checking a claim or story. About 7% mentioned checking the AI’s answer against another source.

Those answers concerned respondents’ most recent election-related AI use. They are not a count of everyone who ever verified an answer. Nor does the survey measure whether AI supplied accurate information: its subject is reported use and trust.

Party differences were smaller than demographic gaps

Reported use crossed party lines. Republicans and Republican-leaning independents stood at 39%, compared with 41% for Democrats and Democratic leaners. The researchers found no statistically detectable party difference in either wave.

Use also varied by race and ethnicity: 51% of Black adults and 51% of Hispanic adults reported an election-related AI task, compared with 32% of White adults. Across both waves, race and ethnicity were associated with use on every task, even after accounting for age, education, income, party and other measured characteristics.

A rise in reported use, not a precise headcount

SSRS surveyed two independent national samples through its probability-based Opinion Panel Omnibus: 1,033 adults in early October and 1,030 in mid-October. Interviews took place online and by telephone, in English and Spanish. Each full sample had a margin of error of ±3.4 percentage points; the difference between waves had a margin of about ±4.8 points.

The questions asked about any use during this election, so some growth is expected as Election Day approaches. The authors describe the levels as approximate. Roughly one in nine people reporting use gave a follow-up indicating no use, no recollection or no usable answer. Some described AI search summaries rather than chat programs; unrecognized AI use could also go unreported.

NCoC’s advice is to treat AI answers as a starting point and confirm important voting details with a state or local election office.