A Third of Young Adults Use Chatbots for Drug Information, Study Finds

More than 32 percent of young adults surveyed by Rutgers Health researchers said they turn to ChatGPT and similar chatbots for information about drugs and alcohol, and roughly half of those users do it at least once a month. The study, published in JMIR Formative Research, drew on survey data from nearly 1,000 participants ages 18 to 29 and is one of the first to measure how generative AI has slipped into the space once occupied by school health classes, friends and a Google search.

The most common reason people gave was unglamorous: they wanted to know about risks and consequences. That is also the category where a confident wrong answer does the most damage.

A familiar shortcut, with a new failure mode

"Young adults are increasingly using AI to get medical advice, so we set out to clarify what that looks like specifically in the context of substances," said Mireia Triguero Roura, a postdoctoral researcher at the Rutgers Institute for Health, Health Care Policy and Aging Research and the study's lead author.

Her concern is not that chatbots exist, but that they answer in complete sentences with no uncertainty markers. "Using these tools can lead to misinformed or otherwise risky substance-use behavior among young people, who are the age group most affected by adverse effects of substance use," she said.

That risk is well documented outside this survey. Health chatbots have been shown to invent dosage ranges, minimize withdrawal danger and produce reassuring responses to leading questions — the sycophancy problem that makes a general-purpose assistant a poor fit for a question like "how much fentanyl is safe to mix with alcohol." The Rutgers authors treat that as the core public health concern: an answer delivered with confidence and no clinician in the loop.

Trust outruns accuracy

The Rutgers team also asked participants how they felt about the answers. Young adults in the sample reported high levels of trust, reliability and helpfulness — a perception researchers say is not matched by the technology's actual performance on substance questions.

The survey found differences in who uses AI this way. Black participants and male participants were more likely to report using chatbots for substance-related information. Triguero Roura said the pattern could reflect differences in health care access, stigma, trust and the ways people seek health information — the same forces that push people toward emergency departments instead of primary care, and toward the internet instead of a clinician.

That reading fits what is known about substance use disorder treatment gaps. People who cannot get a timely appointment, or who expect to be judged when they ask, look for an answer somewhere. A chatbot is available at 2 a.m., does not bill insurance and does not ask for a name.

What clinicians are supposed to do with this

The study does not argue that AI use is harmful on its own. It argues that the information channel changed faster than the clinical response, and that screening conversations should account for it.

In practice that means asking patients where they get drug information, the same way clinicians now ask about supplements and social media. Someone who says a chatbot told them a combination was fine is describing a belief that needs correcting, not a preference. Telehealth services are another place the question belongs: they reach the people most likely to rely on screens for health advice and they have a clinician in the conversation who can redirect it.

For treatment programs, the operational angle is narrower. Intake staff who hear "I looked it up" should not assume the person found a reliable source. A short, non-judgmental correction — and a plain explanation of overdose risk and naloxone — covers ground that a chatbot may have already muddied.

The next question

Triguero Roura and coauthor Pia M. Mauro, an associate professor at the Rutgers School of Public Health, said the next step is understanding how young adults interpret what the tools tell them, and what a new dominant information source means for public health surveillance. Surveys typically assume people form beliefs from clinicians, schools, peers and advertising. Add an always-available assistant that answers in the user's own register and the model of where misinformation starts needs revision.

The study's limitations are the ordinary ones for survey work: nearly 1,000 respondents, self-reported behavior, and a single snapshot in a technology that changes by the quarter. It also cannot say whether AI use preceded substance use or followed it.

What it does establish is scale. Nearly a third of the young adults surveyed had already used a chatbot as a drug-information source. That is no longer an edge case to be handled by a warning label, and the researchers involved treat it as a baseline that clinical guidance and public health messaging have yet to catch up to.