Will voters actually use AI chatbots for election information?

Change Research's survey of 1,892 respondents weighted to a registered-voter population, fielded May 5–10, 2026 and recruited through targeted digital ads and text-to-web SMS, asked Americans which sources they would use to learn about 2026 candidates. News articles led (68% very or somewhat likely), followed by friends, family, or colleagues (62%), candidate websites (57%), and social media (53%); 62% said they would rely on Google Search. Only 15% said they were likely to use an AI chatbot, and a majority (58%) said they were "very unlikely" to do so.

The survey's authors argue the headline number understates AI's reach in two ways: "even a small segment of active AI information-seekers is large enough to be decisive in close elections," and voters who search Google may see AI-generated summaries without seeking them out — passive exposure the survey cannot quantify. Chatbot-seeking is also unevenly distributed: voters under 35 are nearly three times as likely as those over 65 to say they'll consult a chatbot (21% vs. 8%), voters of color are likelier than white voters (21% vs. 12%), and men slightly outpace women (17% vs. 13%). On accuracy, 36% of all respondents say neither AI nor candidates are accurate sources, but trust tracks usage: 31% of under-35 voters and 37% of voters of color (nearly half of Black voters) rate AI-generated information as equally or more accurate than candidate-generated information.

This is self-reported intent to seek information, not a measured persuasion effect, which puts it in a different register from Can generative AI scale personality-targeted political persuasion?, which shows AI-personalized ads can move opinion at scale regardless of whether a voter sought them out, and from Can a simple warning reduce how much LLMs persuade people?, which measured actual belief change under experimental conditions rather than asking people what they plan to do.

The survey measures stated intent before the election, not November behavior, what chatbots actually tell voters who ask, or the accuracy of those answers; the online-recruited sample may also undercount voters without reliable digital access. The data support a narrower claim than "AI will decide the election": a demographically concentrated minority of self-identified chatbot users, layered on an unmeasured amount of passive exposure through AI search summaries, in a political landscape where effective AI persuasion has separately been shown to work at scale on receptive audiences.