KFF poll: weekly AI users more likely to endorse vaccine myths

A new poll puts numbers on something that has been mostly anecdotal, that the crowd going to AI chatbots for health answers looks meaningfully different from the crowd that trusts a doctor, and their beliefs about vaccines follow that split. The Guardian covered findings from KFF's latest Tracking Poll on Health Information and Trust, which surveyed 2,480 U.S. adults between May 7 and 31, 2026 with a ±3 point margin.

The headline number: among the roughly 20% of adults who say they use AI tools or chatbots at least weekly for health advice, 35% say it is probably or definitely true that MMR vaccines have been proven to cause autism. Among adults who never use AI for health advice, that share is 20%. A similar pattern shows up for the claim that mRNA vaccines alter DNA and that measles vaccines are more dangerous than measles. Weekly social media users show an even larger split on the MMR-autism claim, 37% versus 16%, per KFF's writeup.

Why this matters if you work on models or on public health messaging: it is the first survey I have seen that ties frequency of chatbot use for health specifically to endorsement of specific, testable vaccine myths, at a sample and margin worth taking seriously. The trust picture around it is uncomfortable in both directions. KFF also reports that 56% of adults are not confident they can tell true from false chatbot information, and only 29% say they trust chatbots for health information at all.

The honest caveat, which KFF makes itself, is that this is a correlation. The poll does not tell us whether the chatbots are seeding these beliefs or whether people already inclined to distrust vaccines are more likely to ask a chatbot instead of a physician. Adults without a trusted health care provider endorsed the MMR-autism claim at 39%, versus 24% for adults who have one, which points at the provider-trust gap as at least part of the story. What the reporting does not give you is a breakdown by which chatbot the weekly users are actually querying, or the wording those users see back.

The forward-looking piece is that the intervention points here are unusually concrete. Model builders can tighten grounded answers on vaccine queries and benchmark them against the four myths KFF measures, and public health teams can meet users inside the chatbot flow instead of trying to pull them back to a portal they are not visiting.