Pew finds Americans trust chatbots with their lab results, not their loneliness

The interesting number in Pew's data is not the third who have asked a chatbot about a symptom. It is the ratio between the two questions Pew asked separately. People who have used AI for health information rate it almost uniformly useful: 47% extremely or very helpful, 48% somewhat, and only 5% dismissive ^[1]. But when the population is asked whether a chatbot helps someone dealing with loneliness, 39% say it does more harm and 19% say it helps ^[7]. For depression the split is 36% harm against 19% help ^[8]. That is roughly two-to-one against, in a population that otherwise reports the tool working.

Those are not the same respondents answering the same question, and the gap is instructive rather than contradictory. Helpfulness was rated by users, retrospectively, about information. The mental health questions asked everyone to judge an outcome for a third party. So what the product data actually says is: retrieval gets graded on whether the answer arrived, and companionship gets graded on what it does to a person. The second bar is much harder to clear and much harder to instrument.

The use cases people admit to line up with that. Quick information sits at 28%, symptom checking at 25%, cost avoidance at 22%, understanding a diagnosis a doctor already gave at 22% ^[3]. Only 18% cite getting information they are uncomfortable discussing ^[3], the item closest to the confessional mode that consumer AI marketing likes to gesture at. The demand is mostly for a search box with better manners, and it skews young: 44% of 18-to-29s versus 17% of those 65 and over ^[4].

Comfort with the data is thinner than the helpfulness scores imply. Among those using chatbots for medical help, 29% are extremely or very comfortable sharing personal health details, 26% are not, and 42% only somewhat ^[5]. PCMag attributes the risk here to two things: platforms differ considerably in how they handle data, and Congress has not acted on privacy ^[14]. Nobody selling a general-purpose assistant can point to a rule that resolves that for the user.

Pew's second survey, published the same day, sets the disclosure bar these products have not been asked to meet. Seventy-two percent call it extremely or very important that a provider disclose any use of AI at all, rising to 81% for scan analysis and diagnosis and 80% for explaining lab results ^[9][10]. A general-purpose chatbot explaining a lab result is doing the thing 80% of people want flagged in a clinic, without the white coat that triggers the expectation. Meanwhile 46% do not know whether AI has been used in their own care, and of those only 22% say they understand how well ^[11].

One structural gap in the published results: Pew did not break attitudes out by platform ^[6]. So a company shipping a health feature cannot tell from this whether the mental health skepticism attaches to its product or to the category, and the survey population includes people whose only reference point is coverage of chatbots implicated in self-harm cases, which PCMag describes as a horribly mixed record ^[15]. The findings rest on 3,488 panel responses collected June 22 to 28 ^[12].