JMIR Publications

ABSTRACT

Background:

The prevalence of mental distress and health risk behaviors among adolescents and young adults (AYAs) has emerged as a pressing public health concern. AI Chatbots have been increasingly recognized for their potential to provide scalable, accessible mental health support and health education; however, questions remain about their effectiveness in addressing the unique challenges faced by AYAs.

Objective:

To synthesize evidence on the effectiveness of AI chatbot interventions in alleviating mental distress and promoting health behaviors among AYAs.

Methods:

Eight databases (PubMed, PsycINFO, Cochrane Library, CINAHL, Embase, Web of Science, Scopus, IEEE Xplore) were searched for randomized controlled trials (RCTs) published in English between January 1, 2014, and January 26, 2025. Eligible studies assessed the effects of AI chatbots on mental distress and health behavior outcomes among AYAs aged between 15 and 39 years. Data were synthesized narratively and meta-analyzed as appropriate; subgroup and meta-regression analyses were performed to explore moderators of chatbot effectiveness. Risk of bias was evaluated using the revised Cochrane risk-of-bias tool for randomized trials (RoB 2). Evidence quality was evaluated using the GRADE approach.

Results:

Out of 2495 records retrieved, 31 RCTs were included comprising 29637 participants; 26 studies were eligible for meta-analysis. Overall, AI chatbots demonstrated small-to-moderate effects in mitigating mental distress (SMD = -0.35; 95% CI: -0.46 to -0.24; p < 0.001) and promoting health behaviors (SMD = 0.11; 95% CI: 0.03 to 0.19; p = 0.006) in AYAs. Significant improvements were observed for depressive (SMD = -0.43; 95% CI: -0.62 to -0.23; p < 0.001), anxiety (SMD = -0.37; 95% CI: -0.58 to -0.17; p = 0.0003), stress (SMD = -0.41; 95% CI: -0.50 to -0.31; p < 0.001) , and psychosomatic symptoms (SMD = -0.48; 95% CI: -0.82 to -0.14; p = 0.006); negative affect (SMD = -0.27; 95% CI: -0.53 to -0.01; p = 0.04); and self-ambivalence and appearance distress (SMD = -0.25; 95% CI: -0.34 to -0.17; p = 0.01). While AI chatbots contributed to modest enhancements in life satisfaction and well-being, their impacts on positive affect and self-efficacy were limited. The effectiveness of AI chatbots varied depending on target samples, control conditions, and design features such as dialogue system methods, deployment formats, and the use of reminders. User engagement emerged as a critical factor for success, with repetitive content and technical issues noted as primary barriers to adherence.

Conclusions:

This systematic review and meta-analysis highlights the potential of AI chatbots to address mental health challenges and promote health behaviors among AYAs. Retrieval-based dialogue systems demonstrated consistent and reliable effects, while generative systems showed promise but their overall effectiveness was inconclusive. Future research should prioritize developing safety protocols and evaluation frameworks for generative systems, and validating their long-term impacts on mental health and behavior change in AYAs.