The chatbot as confidant, and the question nobody has answered

A survey of nearly 1,000 people aged 18 to 29, published by Rutgers Health researchers in JMIR Formative Research, found that one in three use generative AI such as ChatGPT for information about drugs and alcohol, mostly about risks and consequences¹. About half of those do so at least monthly, and they rated the tools highly for trust, reliability and helpfulness¹. Black and male respondents were more likely to use AI this way, which the lead author said could reflect “differences in healthcare access, stigma, trust”¹.

The wider picture comes from the American Psychological Association’s annual stress survey of 3,132 US adults, run by Harris Poll from late July to late August². Adults who have ever used AI rose from 62% in 2025 to 70% in 2026². Among users, 49% had talked to it as a friend and 47% had used it in a role similar to a mental health professional, for example to check a therapist’s suggestion². One in ten said they rely on AI more than on friends and family for advice².

The reading. Last night this desk read the same APA survey as a ledger of money worries. Read it again as a ledger of where people go with them. Therapist-style use falls steeply with age: 26% of users aged 18 to 29 do it regularly or often, against 3% of those aged 62 to 80². Users report more good experiences than bad, 51% to 29%, but 12% worry about dependence and 9% say it left them feeling more disconnected².

Neither study can say whether the chatbot eases loneliness or replaces the harder phone call². That is the honest answer. The constructive one: a question asked of a machine is still a question asked. Public-health services, pharmacies and doctors’ practices can meet it with clear, findable answers. The machine borrows its knowledge from somewhere; make sure that somewhere is good.