One in Three American Adults Now Uses AI Chatbots for Health — Pew

On August 25, the Pew Research Center published one of the most detailed portraits yet of how ordinary Americans are folding AI chatbots into one of the most sensitive corners of their lives: their health. The survey of 3,488 U.S. adults, conducted June 22–28, 2026, finds that 34% of American adults now use AI chatbots for at least one health-related task — and the picture that emerges is one of enthusiastic adoption shadowed by deep, unresolved privacy and trust questions.

The report, “From Diagnoses to Treatments, Why Americans Use AI Chatbots for Health,” by Emma Kikuchi, Giancarlo Pasquini and Eileen Yam, lands at a moment when tech companies are actively marketing chatbot tools for healthcare while regulators and researchers sound alarms about data handling. The numbers suggest the public has already moved faster than the debate.

What people actually ask chatbots about

The eight use cases Pew measured span convenience, cost, and genuine medical triage. The leading motivations:

• 28% — to get health information quickly
• 25% — to figure out what’s causing their symptoms
• 22% — to get health information at little to no cost
• 22% — to get health information about treatments
• 22% — to learn more about a diagnosis from a doctor
• 20% — to help understand lab results
• 18% — to get health information they’re uncomfortable talking about
• 15% — to help decide whether to go to a doctor

Read together, the list describes a population using chatbots as a pre-clinical filter and a post-clinical explainer — before the appointment (“what is this rash?”) and after it (“what does this lab value mean?”). A quarter of Americans are effectively using AI to self-diagnose, and roughly one in five uses it to decode what their doctor told them — a quiet indictment of how overburdened and time-poor conventional healthcare communication has become. Nearly all chatbot health users cited multiple reasons, meaning this is habitual, not experimental, use.

Users find it helpful — overwhelmingly

If you expected a backlash, the satisfaction data points the other way. Among chatbot health users:

• 47% say the information is extremely or very helpful
• 48% say it is somewhat helpful
• Only 5% say it is not too or not at all helpful

That is a 95% “at least somewhat helpful” rating for tools that were niche curiosities three years ago. Pew notes this aligns with its earlier finding that convenience rates highly among chatbot health users. Caveats apply: satisfied users answering a survey can’t see the errors they never caught — the classic failure mode of health misinformation isn’t measured here. But as a measure of perceived value, it is about as lopsided as survey data gets.

Who is using chatbots for health

The demographics track with overall chatbot adoption, with a few sharp edges:

• Asian Americans: 56% (English-speaking respondents) — the standout group
• Ages 18–29: 44%, ages 30–49: 40%, ages 50–64: 32%, 65+: 17%
• College graduates: 44% vs. high school or less: 23%
• Upper income: 48%, middle income: 33%, lower income: 28%
• Hispanic adults: 39%, Black adults: 35%, White adults: 29%

Two readings compete here. The optimistic one: adoption is broad-based across race and ethnicity. The sobering one: adoption skews toward the young, educated, and affluent — the people least likely to be medically underserved — while the 65+ population, the heaviest consumers of healthcare, lags at 17%. If chatbots genuinely improve health literacy, the groups that most need that boost are getting it least.

Attitude is also a powerful predictor: 60% of adults who say they feel more excited than concerned about AI are chatbot health users, versus just 21% of those who feel more concerned.

The privacy paradox

The report’s sharpest tension sits in the data-sharing numbers. Among chatbot health users asked how comfortable they’d be sharing personal health information with a chatbot:

• 29% — extremely or very comfortable
• 42% — somewhat comfortable
• 26% — not too or not at all comfortable

So a population that happily types symptoms into a chat window is nearly as likely to be deeply uncomfortable handing over personal health data as it is to be fine with it — with the largest share caught ambivalently in between. Pew contextualizes this with current controversies: platforms that have faced scrutiny for storing or selling health data, and for using user conversations to train large language models. The survey can’t say whether users understand what happens to their prompts after they hit enter — but the 26% who are uncomfortable appear to be voting with their typing anyway.

This is the gap regulators are circling. Health data enjoys special legal protection in clinical settings (HIPAA in the U.S.), but a symptom description typed into a general-purpose consumer chatbot generally does not. The survey quietly documents millions of Americans routing medically significant information through a regulatory gray zone.

The mental health red line

A companion Pew analysis, published the same day, draws the boundary of public tolerance more starkly. Asked whether chatbots do more to help or hurt people who use them for support with loneliness, depression, or stress, Americans tilt negative:

• Loneliness: 39% say hurt, 19% say help (14% neither, 27% not sure)
• Depression: 36% hurt, 17% help (14% neither, 33% not sure)
• Stress: 29% hurt, 22% help (18% neither, 31% not sure)

Younger adults — the heaviest chatbot health users — are largely pessimistic about chatbots’ social and mental-health value, while sizable shares of older adults simply say they’re not sure. The pattern carves out a clear public distinction: AI as an information tool for health is accepted and even embraced; AI as an emotional companion or therapeutic substitute is viewed with suspicion, at a time of high-profile cases involving chatbot use for social and mental health support.

The trajectory matters more than the snapshot

Pew’s own trendline shows overall chatbot use at 44% of U.S. adults in June 2026, up from 34% a year earlier and more than double the share of two years prior. Health use is riding that curve. A separate recent estimate puts monthly AI health-information use near 29% of adults — nearly double two years ago. Barring a shock, the 34% “ever used” figure will keep climbing.

The unresolved questions are the ones the survey exposes rather than answers: What accuracy standard should apply when a quarter of the country self-diagnoses with AI? Who owns the symptom data, and what may be trained on it? And where is the line between a chatbot that demystifies lab results — a genuine public good — and one that quietly becomes the cheapest mental-health option for people who have no other?

The public, for now, has drawn its own line: information yes, intimacy no. Whether the industry respects it is the story to watch.