The Uses Cluster Around Translation and Fallback
Laid out across the eight categories, the top uses are not people asking AI to be their doctor. They are two plainer needs. One is translation: making sense of a diagnosis a doctor delivered, or reading the numbers on a lab report — a quarter of people are doing this. The other is fallback: 22% use it because it provides low-cost information, meaning it substitutes when professional channels are unavailable or inconvenient. The category that comes closest to actual diagnosis — working out what is causing symptoms — sits at 25%. That is not a small share, and it differs in kind from the other two. The first two supplement an existing medical process; this one happens before that process, and the consequences of being wrong are different.
Useful and Comfortable Are Not the Same Thing
The gap worth remembering is this one: 47% call the answers extremely or very helpful, while only 29% are very comfortable sharing personal health data with the tools. A substantial group is using these tools while not entirely trusting them — willing to ask, unwilling to hand over the specifics. For teams building health-adjacent AI products, that gap is more instructive than the adoption rate. What limits conversion is usually not answer quality but how much context users are willing to provide. Commitments like keeping data local, excluding it from training and allowing deletion at any time are exactly what determine whether someone actually uploads the lab report.
Demographic Splits and the Other Half of the Findings
Usage varies sharply by group: 56% among Asian Americans and 44% among adults under 30, both well above the 34% overall. The design implication is concrete — younger users and certain demographic groups already treat this as a default entry point, while the broader population does not. Pew released a companion report the same day pointing the opposite direction: Americans on balance say chatbots do more to hurt than help people who use them for loneliness, depression or stress, with younger adults largely pessimistic. Read together, public attitudes are actually well differentiated — AI as an information tool is acceptable; AI as an object of emotional support is not. On method: the sample is 3,488 US adults drawn from Pew's American Trends Panel, representing the full US adult population, surveyed June 22–28. One comparison point: in a survey conducted October 20–26, 2025, 22% said they get health information from AI chatbots at least sometimes — notable growth across eight months, though the question wording differs, so the two figures cannot simply be subtracted.
via: Pew Research Center report, Pew: chatbots and mental health