AI Chatbots Could Reshape Cannabis Use Support, But Safety Questions Loom

Cannabis is no longer a fringe public health concern. Across the globe, the number of people living with cannabis use disorder and the burden of disease tied to it have climbed notably over recent decades, with young people bearing a disproportionate share of the impact. Yet as the problem has grown, the public health response has lagged behind. In a recent editorial published in the International Journal of Public Health, researcher Yuqi Hu of the Herbert Wertheim School of Public Health and Human Longevity Science at the University of California San Diego argues that artificial intelligence could help close this widening gap, provided the technology is built with safety and privacy as non-negotiable design principles rather than afterthoughts.

The treatment landscape for problematic cannabis use is strikingly thin. There are no established pharmacological treatments for cannabis use and dependence, which leaves psychosocial approaches as the mainstay of care. Interventions grounded in motivational interviewing and cognitive-behavioral principles can improve cannabis use outcomes, helping people examine their ambivalence about use, identify the situations that trigger craving, and rehearse practical coping and relapse-prevention skills. But these approaches share a structural weakness: they depend on trained professionals, scheduled appointments, and the willingness of individuals to walk through a clinic door, and each of those dependencies creates a bottleneck.

Access barriers stack up quickly. Workforce shortages limit how many people can receive care, time constraints make repeated sessions difficult to sustain, and transportation challenges put formal treatment out of reach for many. Stigma compounds the problem. Histories of drug control and enforcement have fallen unevenly on underserved populations, and the shadow of that history discourages formal help-seeking even where services exist. Compounding all of this, existing interventions are rarely delivered at the moments when people need support most, and they are often minimally responsive to shifting risk states and individual circumstances. It is precisely this combination of scarcity, stigma, and poor timing that AI-based tools are positioned to address.

Conversational chatbots are the most immediately promising entry point. In principle, a well-engineered chatbot could deliver the core elements of psychosocial interventions for cannabis use directly through a conversational interface: guiding users as they set goals, helping them identify personal triggers, strengthening motivation to change, and walking them through coping strategies and relapse-prevention exercises. The appeal is not novelty for its own sake. A chatbot can offer private, on-demand support outside conventional clinic settings, reaching people who face geographic, transportation, scheduling, or stigma-related barriers to care. And because software scales differently from human labor, chatbots could expand to large populations at exceptionally low marginal cost, reducing reliance on scarce trained professionals.

The evidence base for this approach, while still developing, is encouraging. A recent systematic review of chatbot-assisted substance use interventions, focused mainly on tobacco and alcohol rather than cannabis, found that experimental studies reported meaningful effects on substance use outcomes. Users, for their part, generally found the interventions useful and acceptable. That combination of measurable benefit and user acceptance matters, because an intervention that people abandon is an intervention that cannot help them. The review’s findings suggest that the conversational format itself, delivered at low cost and without judgment, can carry real therapeutic content, even though direct evidence for cannabis-specific chatbots remains limited.

Beyond chatbots, the editorial highlights a second, more technically ambitious avenue: just-in-time adaptive interventions, or JITAIs. These systems use machine learning methods to identify moments of heightened vulnerability and trigger support in a timely manner, although evidence of their effectiveness for substance use is still limited. For cannabis use, a JITAI could draw on self-reported data, prior engagement patterns, time, location, or other digital signals to detect vulnerable moments, such as craving after work, social exposure to use, boredom late at night, or stress following a conflict. Once a high-risk moment is detected, the system could respond in real time with a motivational message, a coping prompt, or a brief cognitive-behavioral exercise. The logic is rooted in the nature of cannabis use itself, which is often episodic and heavily context-dependent; support delivered at the moment of risk may succeed where scheduled sessions cannot reach.

The two approaches are not mutually exclusive, and the editorial suggests they could be combined. In such a model, the JITAI component identifies moments of heightened risk using self-reported, behavioral, or contextual data, while the chatbot delivers support in a conversational format that adapts to the user’s current state. A recent smoking-cessation intervention that integrated a JITAI with a generative chatbot was evaluated favorably by a domain expert, offering a proof of concept for the architecture. There is also reason to believe psychosocial intervention elements from tobacco research could transfer to cannabis, since the two substances overlap in the psychosocial processes involved: motivation, cravings, triggers, coping, and relapse. Supporting this possibility, cannabis co-users who reduced their cigarette use in a smoking cessation intervention also tended to reduce their cannabis use, even though cannabis was never explicitly addressed in the intervention.

Adaptation, however, would be essential rather than optional. Cannabis use differs from tobacco use in ways that any AI-delivered intervention must account for. The products themselves are diverse, spanning different formulations and routes of administration. Many users report self-medication as a motivation, which changes the calculus of behavior change. Cannabis use is entangled with social identity in distinctive ways, and its legal status varies across jurisdictions, shaping both the risks users face and their willingness to disclose use to a digital system. An intervention that treats cannabis as interchangeable with tobacco would miss these realities, and the editorial is explicit that adaptation to these unique characteristics is a prerequisite, not a refinement.

Safety concerns sharpen the picture considerably. A review cited in the editorial found that over half of health-related chatbots failed to respond adequately to high-risk messages, including expressions of suicidal thoughts. In cannabis interventions, such limitations become particularly consequential, because users may experience psychotic symptoms or other acute crises during the course of their use. A chatbot that handles routine craving prompts well but fumbles a crisis message is not a safe intervention. The editorial argues that AI systems in this space should therefore have a clearly limited clinical scope, minimize data collection and retention, clearly explain any limits of confidentiality to users, and provide reliable escalation pathways to human support when conversations move beyond what the system can safely handle.

Privacy raises a further layer of concern. AI systems may collect sensitive behavioral data, infer vulnerability from patterns in that data, and transmit information across systems that users do not fully understand. Depending on the jurisdiction, intervention data may even be subject to compelled disclosure in legal or administrative proceedings, a risk that carries particular weight for a substance whose legal status remains contested in many places. These are not abstract worries for the populations such tools are meant to reach, many of whom already have reason to distrust institutional data collection. The editorial’s prescription is clear: privacy protections must be built into the architecture of these systems from the start, with data minimization and transparent confidentiality limits treated as core requirements rather than optional features.

Taken together, the editorial makes the case that AI approaches could substantially extend the reach of psychosocial support for cannabis use and warrant sustained research investment. But it also frames the research agenda in a way that resists technological enthusiasm for its own sake. Design, development, and evaluation of AI interventions should look beyond efficacy alone and ask whether these tools genuinely improve engagement and reduce access and stigma-related barriers without introducing safety or privacy harms. Future implementation research, the editorial argues, should identify which populations benefit and which remain underserved, establish reliable pathways to human care, and examine how effective innovations can be integrated into routine care and public health infrastructure. The promise of AI in this field is real, but so is the standard it must meet.

Subject of Research: Artificial intelligence approaches, including chatbots and just-in-time adaptive interventions, for cannabis use interventions

Article Title: Chatbots and beyond: transforming cannabis use interventions with innovative AI approaches

Article References: Hu, Y. (2026). Chatbots and beyond: transforming cannabis use interventions with innovative AI approaches. International Journal of Public Health, 71, Article 1609866. https://doi.org/10.3389/ijph.2026.1609866

Image Credits: AI Generated

DOI: 10.3389/ijph.2026.1609866

Keywords: artificial intelligence, cannabis use, chatbots, just-in-time adaptive interventions, psychosocial intervention, public health, machine learning, digital health, beyond, transforming, cannabis, interventions