A U.S. survey found 32.3% of respondents ages 18 to 29 had used generative AI for substance-related information, showing how chatbots have entered health searches without testing the answers’ accuracy.
Published Oct. 19 in JMIR Formative Research, the study surveyed 995 people recruited through Prolific in April 2026. Researchers excluded five people who did not pass quality checks. The survey asked about whether participants had ever used AI for substance questions, rather than whether they used it recently.
In all, 321 respondents reported having done so. Among that group, 48.2% said they used AI for this purpose at least monthly: 32.1% monthly, 12.1% weekly and 4% daily or almost daily.
The most common reason was to learn about the consequences of substance use, reported by 92.8% of AI users. Another 48.6% sought information about specific substances, while 34.9% asked about substance use disorders. Participants could select more than one reason.
Google searches were also common. Across the full sample, 572 people, or 57.5%, reported searching Google for substance-related information. Of the 321 AI users, 298, or 92.8%, also reported using Google for that broad purpose.
Among those 321 respondents, the average trustworthiness rating for AI was 3.66 out of five for substance-related questions and 3.51 for general use. Their helpfulness ratings were 4.12 for substance questions and 4.44 for general use. These figures record participants’ views; the researchers did not assess the answers themselves.
The study’s adjusted model found that the odds of reporting substance-related AI use were 2.52 times higher for Black respondents than for white respondents, and 2.04 times higher for male respondents than for female respondents. The comparisons account for demographic variables in the model and describe odds, not differences in usage rates. Study lead author Mireia Triguero Roura said the gaps “could reflect differences in healthcare access, stigma, trust, and health information-seeking behaviors.”
The authors said online volunteer recruitment may have drawn more people who spend substantial time online or already know AI tools. They cautioned that this could raise the estimate and limit how well it represents young adults overall. The survey did not measure whether respondents acted on chatbot answers; the authors called for further research to check the answers and examine how young adults use the information.
