The real finding of Nevada's study on AI and health: people aren't asking for diagnoses, they're looking for comfort

By Zendoric · October 6, 2026.

A team at the University of Nevada, Reno —led by Information Systems professor Alan Yang, of the College of Business, along with students Jade Wang, Hyun Woo Chung and Giang Do— has analyzed more than 1,600 real conversations between users and generative AI systems focused on physical and mental health. The goal was to understand what people use these tools for when it comes to their body or their mind, beyond the occasional anecdote.

The most frequent reason was health education: questions about illnesses, symptoms, medications, nutrition and lifestyle habits. But the interesting finding is not the list of topics, it is the purpose behind them. According to the researchers, most users were not looking for a definitive diagnosis, but to clear up a doubt or find reassurance about a symptom that worried them. "Most people don't use AI to replace the doctor," Yang sums up. "They use it as a comfortable first step to understand symptoms, learn about conditions, ask about medications or resolve everyday health questions."

The finding that really sets this study apart shows up in the mental health conversations. Unlike the physical health queries, these did not arrive as direct questions but as personal accounts, detailed about the users' own experiences and difficulties. For Yang, that suggests some users do not treat AI as an encyclopedia, but as a space for reflection and support: a non-judgmental interlocutor for subjects that are hard to share.

The study also offers a reassuring nuance. Many conversations already mentioned treatments, medications or diagnoses issued by health professionals, and the chatbots frequently responded by encouraging users to seek medical attention in the face of serious signs. In other words, in the usage observed, AI appears to act more as a complement and a gateway than as a substitute. Yang puts it in writing himself: "While AI can be a useful educational tool, it should not be seen as a substitute for professional medical care. AI systems can give incomplete information, make mistakes or lack important context about a person's medical history."

Broadly speaking, the underlying reason someone turns to a chatbot to talk about anxiety or insomnia is no mystery: professional therapy remains expensive, slow to obtain and, in many areas, scarce, while a chatbot is available at any hour with no appointment fee. That access gap is the one generative AI is filling in practice, with or without clinical design behind it.

Here it is worth separating the data from the easy headline. A chatbot "encouraging someone to see a doctor" is a reasonable design guardrail, not a safety guarantee: it does not replace clinical judgment, it does not detect a crisis with the reliability of a professional trained to do so, and nobody takes responsibility if it fails at the worst moment. The study itself flags that limitation —lack of context, errors, incomplete information— and it should not be lost sight of just because it also brings good news.

But the underlying good news does not disappear because of that caveat. If these tools are designed with doctors and therapists inside the process rather than outside it —with crisis detection, with traceable referral, with a real connection to the healthcare system—, the result is the first rung of health access that many people lack today: something to ask at three in the morning that does not charge for an appointment and does not judge. It is the kind of abundance we champion over the long term at Zendoric: not that AI cures, but that it multiplies access to the first step of care —information, reassurance, guidance— so that the human healthcare system reaches those who genuinely need a doctor better and sooner. The Nevada study does not measure that future, it measures how people behave today. But that snapshot, taken with academic rigor rather than alarm, is exactly the material hospitals, insurers and AI developers should be using to design the next version of these tools: not to replace the professional, but so that fewer people arrive too late at their appointment.