One in three respondents used AI for substance questions in a US young-adult survey

In a survey of 995 US adults aged 18–29, 321 respondents—32.3%—said they had used generative AI for information about substances. Among those users, almost half reported asking at least monthly. The study gives a detailed view of a health-information habit within the sampled group, including what people wanted to know and how they judged the answers.[2]

Researchers Mireia Triguero Roura and Pia Mauro recruited participants through the Prolific online panel in April 2026, using quotas for age, sex and race. The paper appeared in JMIR Formative Research on 19 October and was highlighted by Rutgers on 23 October. Quotas help structure the sample, but they do not make it a probability sample of all young adults in the United States.[1][2]

Keeping the denominators visible

The questions concerned information seeking, rather than a measurement of substance-use prevalence or clinical consultations. Asking about consequences does not establish that a person used a substance, followed an answer or changed a health decision. Those outcomes would require different questions or follow-up.[2]

AI was also part of a wider information environment. The paper reports that 57.5% of the full sample used Google for substance-related information, and nearly all respondents who used AI for this purpose also reported Google use. That overlap suggests people may move between sources rather than replace conventional search entirely.[2]

Trust is a reported attitude

Participants generally rated AI as useful and trustworthy, and use was associated with respondent characteristics, including higher adjusted odds among male and Black participants in the study. These are patterns within the sample. They do not explain why a particular individual chose AI or establish a single cause for differences between groups.[2]

Confidentiality, convenience and a conversational interface are plausible reasons for seeking information this way, but the survey cannot determine which explanation caused the reported behaviour. Those possibilities are questions for further research, particularly when considering how reliable services could become easier to reach.

The study did not collect and clinically evaluate the actual answers users received. Trust scores therefore describe perceptions, not verified accuracy. A separate assessment would need realistic questions, expert review and attention to whether an answer recognises uncertainty or situations requiring human care.[2]

That distinction gives health researchers two complementary tasks: understand which questions people bring to AI, and test how well current systems answer them. The survey contributes to the first. Its topic and frequency data can help shape an evaluation grounded in actual information-seeking behaviour rather than an arbitrary list of benchmark prompts.

The finding is specific enough to matter without projecting it unqualified onto an entire generation. A substantial share of this online sample already used AI for substance questions, often alongside search. Evidence about those habits can help health services decide which information needs deserve clearer, more accessible answers.[1][2]

Report an error or suggest a correction