Introduction: The growing demand for mental health care and limited access to specialized services have driven the development of AI-based digital solutions. Conversational chatbots emerge as promising diagnostic and therapeutic support tools, yet their clinical effectiveness and safety requirements remain under investigation.
Purpose: To analyse the current state of AI-powered chatbots in mental health, examining their effectiveness, technical requirements, and user/professional perceptions as complementary tools in clinical practice.
Method: A narrative review with exploratory elements was conducted across PubMed, PsycINFO, and Web of Science databases. Studies published between August 2019 and August 2025 addressing AI applications in mental health, specifically chatbots for diagnostic or therapeutic support, were included. Search terms combined artificial intelligence, conversational agents, mental health, and clinical processes. After duplicate removal and screening, 26 studies were analysed.
Results: The review included 11 randomized controlled trials, 6 systematic reviews, 7 observational studies, and 2 technical development studies. Evidence shows modest but consistent positive effects on depressive and anxiety symptoms, with small-to-moderate effect sizes in short-term interventions. Rule-based systems demonstrated good adherence and structured outcomes, while generative models showed greater conversational flexibility but required enhanced safety protocols. Critical limitations include lack of robust clinical validation, interpretation errors, and inability to manage crisis situations autonomously. Users valued accessibility and privacy, while professionals emphasized the need for complementary rather than substitutive use.
Conclusions: Chatbots show potential for transforming mental health care access but require rigorous evaluation, safety protocols, and clear regulatory frameworks. Implementation demands technical safeguards, clinical supervision, and ethical guidelines ensuring responsible integration into healthcare systems while maintaining human clinical judgment as fundamental.