This narrative review examines whether Large Language Model (LLM)–based chatbots can help close the global mental health treatment gap while weighing their public-health risks. We synthesize peer-reviewed studies and relevant case reports to: (1) map the dimensions of the mental health treatment gap, (2) describe how recent LLM advances have changed chatbot capabilities, (3) explore how chatbots can address the dimensions of the gap, (4) evaluate evidence for clinical effectiveness, and (5) outline major safety, ethical, and policy concerns. Findings indicate that chatbots offer scalable, always-available, and low-cost support that can reduce barriers related to stigma, geographic and temporal access, affordability, and mental-health awareness. We found that the evidence supports chatbot interventions’ efficiency in small-to-moderate short-term reductions in depression and anxiety symptoms, while the long-term effects and use in other disorders remain largely unexplored. However, LLM chatbots also present clear risks: hallucinations and clinically inappropriate responses, amplification of stigma or bias, user dependence, and significant data-security vulnerabilities. Importantly, most widely used generalist LLMs lack rigorous clinical validation. We conclude that LLM chatbots are a persistent feature of the mental-health ecosystem whose benefits can be realized only with robust safety guardrails, transparent evaluation, integration into stepped-care pathways, and proactive regulation.
Ufniarski et al. (Fri,) studied this question.