One in three Americans now ask AI chatbots for health advice
One in three US adults has used an AI chatbot for health advice, Pew finds — and most rate the answers as helpful. A Pew Research Center survey of 3,488 US adults fielded June 22–28 found 34% have used chatbots for at least one health reason: 28% for fast answers, 25% to figure out what's causing symptoms, 22% because it's free or nearly free. Adoption runs far ahead of the average among Asian Americans (56%), adults under 30 (44%) and upper-income users (48%). Nearly all chatbot health users find the information at least somewhat useful — yet only 29% say they'd be very comfortable handing over personal health data.
The pattern is less about trust than about access. People aren't choosing chatbots over doctors on medical merit; they're choosing speed and zero cost over waiting rooms and bills, and 18% go to chatbots for topics they'd be embarrassed to raise with a person. The demographic split is the warning: the people leaning hardest on chatbot medicine are also the ones with the least slack in the system. When a quarter of the country self-triages symptoms in a chat window, the question shifts from whether AI gives good medical advice to who gets blamed when it doesn't — and health systems still don't have that answer.
We've tracked this drift before — Nearly 80% of consumers now ask AI before they buy showed the same substitution happening in shopping; health is a heavier domain to lose to defaults.
An unverified demo shows Unitree and AgiBot robots sharing one "brain" — rival hardware cooperating without instructions. A 10-minute single-take video circulating in China's robotics scene shows a Unitree G1 and a 1.7-meter AgiBot robot doing chores in a cramped apartment: cleaning windows, sorting laundry, resuming interrupted tasks, and at one point collaborating — the AgiBot robot drapes a scarf around its Unitree counterpart's neck so both hands stay free. The claim drawing attention is cross-embodiment: one model controlling two competing robots with different sensors and body plans, trained reportedly on only tens of hours of video.
The details, if they hold up, matter more than the chore list. Long-horizon tasks without error buildup, mid-task interruption and recovery, and improvised tool use — kicking a box into position to reach a high shelf — are exactly the failure modes that VLA models (the dominant vision-language-action approach) struggle with. But no team has claimed the video, the footage hasn't been independently verified, and QbitAI's own report reads as awed speculation from unnamed insiders. Treat it as a signal of where embodied AI is heading, not proof someone has arrived.
What to watch: whether any lab steps forward to claim the mystery model — a verified paper or product page would turn this from rumor into the year's biggest robotics story.
Are you one of the third of Americans taking health questions to a chatbot — and would you share your symptoms data with one? Tell us in the comments.
Sources: Pew Research Center · QbitAI · 36Kr
