One in Three Young Adults Turns to AI for Drug and Alcohol Information, Study Finds

Digital trust is reshaping how young people seek health advice.

NEW BRUNSWICK, NEW JERSEY.

Nearly one in three young adults surveyed in the United States has used artificial intelligence chatbots to obtain information about drugs and alcohol, according to a new Rutgers University study. The findings reveal an emerging dimension of digital health behavior in which conversational AI systems are becoming sources of information about substance use, addiction and potential health risks, raising questions about accuracy, trust and the growing influence of automated advice on personal decisions.

Published in JMIR Formative Research on October 19, the study examined responses from 995 Americans aged 18 to 29 who participated in an online survey conducted in April 2026. Researchers Mireia Triguero Roura and Pia M. Mauro found that 32.3% of respondents had used generative AI tools for substance-related information. The platforms considered included ChatGPT, Google Gemini and Meta AI, reflecting the expanding availability of conversational technologies capable of responding to questions about sensitive health topics.

The research provides an important distinction between occasional experimentation and habitual use. Among participants who had consulted AI about substances, approximately half reported doing so at least monthly. Twelve percent used these systems weekly, while 4% turned to them daily or almost daily. These figures suggest that, for a portion of young adults, chatbots are becoming recurring information resources rather than technologies used only out of curiosity.

The motivation behind these interactions is particularly revealing. Nearly 93% of respondents who used AI for substance-related questions reported seeking information about risks and consequences. Approximately half requested information about particular substances, including questions involving dosage, while 35% explored topics associated with substance use disorders, such as addiction. The findings challenge any assumption that conversations about drugs necessarily indicate an intention to consume them. Many users may be attempting to understand potential dangers, evaluate unfamiliar situations or obtain information they hesitate to request elsewhere.

This distinction matters for public health. Young adulthood is a period during which experimentation, social pressures and changing personal circumstances can influence decisions about alcohol and other substances. Access to reliable information can support safer decisions and earlier recognition of harmful patterns. Inaccurate or misleading guidance, however, can create risks when users mistake automated responses for individualized medical advice.

The study also examined how participants perceived the reliability of generative AI. Respondents who used chatbots for substance-related information assigned them an average trustworthiness score of 3.66 out of five, compared with 3.51 for AI use more generally. Although the numerical difference was modest, it indicates that users may place considerable confidence in systems when discussing subjects with direct implications for their health.

That confidence deserves scrutiny because perceived trustworthiness is not equivalent to demonstrated accuracy. Generative AI models produce responses by identifying patterns learned during training and subsequent optimization. Their conversational fluency can make explanations appear authoritative, even when particular statements are incomplete, unsupported or incorrect. In health-related contexts, the consequences of such errors may extend beyond misinformation to decisions affecting physical safety.

The researchers did not independently evaluate the medical accuracy of chatbot responses in this survey. Nor did they establish whether participants subsequently changed their substance use behavior because of the information they received. The findings therefore document reported usage patterns and perceptions, not proof that AI advice caused harmful or beneficial outcomes.

Another significant result concerns the relationship between AI and conventional internet searches. Among respondents who had used generative AI for substance-related information, 92.8% also reported searching Google for similar topics. This overlap suggests that chatbots are functioning primarily as an additional information channel rather than replacing established search engines.

The distinction has implications for how young adults evaluate information. Traditional search engines generally direct users toward external sources that can be compared and examined individually. Conversational AI systems often deliver synthesized answers within a single interaction, reducing the effort required to obtain an explanation. That convenience can improve accessibility, but it may also make it less obvious where particular claims originated or whether competing interpretations exist.

The appeal of conversational systems may be especially strong when questions involve embarrassment, stigma or fear of judgment. Someone uncertain about a substance’s effects or worried about a pattern of consumption may find an automated conversation less intimidating than discussing the subject with a family member, healthcare professional or counselor. The study identifies anonymity as a plausible factor influencing use, although it does not establish the specific motivations of every participant.

Researchers also identified differences across demographic groups. After accounting for the other variables included in their statistical model, Black respondents had 2.52 times the adjusted odds of reporting substance-related AI use compared with white respondents. Men had approximately twice the adjusted odds of women. These figures describe associations within the survey sample and should not be interpreted as evidence that demographic characteristics directly cause particular patterns of technology use.

The authors suggested that differences in healthcare access, stigma, trust and information-seeking behavior may contribute to the observed variation. Those explanations remain hypotheses requiring further investigation. Understanding why different groups turn to AI could help researchers and public health institutions identify unmet information needs without assuming that all users share the same circumstances or expectations.

The study’s methodology also warrants attention. Participants were recruited through Prolific, an online research platform, using demographic quotas intended to approximate the United States young adult population. Although the researchers implemented quality controls, the sample was not selected through probability-based national sampling. Individuals who are more comfortable using digital services may consequently be overrepresented, limiting the extent to which the 32.3% figure can be generalized to every American aged 18 to 29.

These limitations do not diminish the significance of the observed behavior. They establish the boundaries within which the findings should be interpreted. The study provides evidence that a substantial proportion of respondents are already incorporating generative AI into decisions about where to seek substance-related information, while leaving open questions about the wider population and the practical consequences of those interactions.

For technology developers, the findings reinforce the importance of designing AI systems that distinguish general educational information from circumstances requiring professional medical assessment. A chatbot can explain established health concepts, but it may lack the clinical history, examination findings and contextual information necessary to assess an individual’s condition. Reliability also depends on how systems respond when evidence is uncertain or when a user’s question suggests an immediate health risk.

Public health organizations face a related challenge. If younger adults increasingly obtain information through conversational interfaces, educational strategies designed primarily for websites, brochures or conventional search results may reach only part of their intended audience. Institutions may need to consider how authoritative information can remain accessible and recognizable within an environment where automated systems increasingly mediate communication.

The issue is not simply whether artificial intelligence should provide information about drugs and alcohol. Restricting access to reliable educational material could discourage legitimate questions, while unrestricted but poorly calibrated responses may expose users to misleading claims. The more consequential objective is to ensure that information remains accurate, appropriately qualified and clear about the limits of automated assistance.

This creates responsibilities extending beyond individual chatbot providers. Researchers can evaluate the accuracy of generated responses, health agencies can publish accessible evidence-based guidance, and educational institutions can strengthen the ability of students to assess sources critically. Healthcare professionals may also benefit from understanding that patients could arrive at consultations having already discussed sensitive topics with AI systems.

The Rutgers findings illuminate a broader transformation in the relationship between people and information. Conversational systems do more than retrieve material. They organize explanations, respond to follow-up questions and create an experience that can resemble personalized guidance. That interaction may encourage users to disclose concerns or ask questions they would otherwise avoid, even when the system cannot provide the accountability associated with professional care.

The central risk is therefore not necessarily the act of consulting artificial intelligence. It arises when users assign more authority to an answer than its evidence, context or limitations justify. This gap between perceived understanding and verified reliability becomes particularly consequential in decisions involving substance use and personal health.

The study identifies an emerging public health challenge rather than demonstrating a crisis caused by chatbots. Its authors call for further research into response accuracy, demographic differences and the ways AI-generated information may influence behavior. Those investigations will be essential for distinguishing beneficial access to information from situations in which automated guidance contributes to misunderstanding or harm.

As generative AI becomes embedded in everyday information seeking, the boundary between receiving an explanation and trusting a recommendation may become increasingly difficult for users to recognize. The findings from Rutgers suggest that this transition is already taking place among young adults discussing drugs and alcohol, subjects where uncertainty can carry substantial consequences.

The larger question is how health institutions, researchers and technology companies will respond to an environment in which access to information is becoming easier while evaluating its reliability remains demanding. The future of AI-mediated health communication will depend not only on the sophistication of the technology, but on whether users can understand what its answers establish, what they leave unresolved and when human expertise remains indispensable.

Where Innovation Meets Responsibility.