259 - AI-Chatbot Mental Health Interventions for Adolescents: A Systematic Review

Publication Number: 3249.259

Sarah Sabir, University of Kansas School of Medicine, Wichita, KS, United States; Ghulam Mustafa Bajwa, Montclair State University, New York, NY, United States; Aoun Zaib Khan, Metrohealth Medical Center/Case Western Reserve Univerity, Cleveland, OH, United States

Background: Adolescents increasingly face mental health challenges such as depression, anxiety, and body image concerns. Barriers to traditional care include stigma, cost, and provider shortages. AI-powered chatbots (e.g., ChatGPT, Woebot) offer accessible mental health support through therapeutic conversations and psychoeducation. However, evidence about their effectiveness in adolescents remains scattered and inconsistent.

Objective: This systematic review aims to synthesize evidence from randomized and quasi-experimental trials evaluating the effectiveness of AI chatbot interventions on mental health outcomes in adolescents.

Design/Methods: A systematic review was conducted including randomized controlled trials (RCTs) and quasi-experimental studies comparing AI chatbot mental health interventions to standard care, waitlist, or alternative interventions. Studies published in English from 2020 to 2025 were eligible. Primary outcomes evaluated changes in validated mental health measures such as depression, anxiety, psychological distress, and body image. Comprehensive searches were performed in PubMed and Google Scholar. Eligible populations were adolescents aged 13 to 18 years using AI chatbots targeting key psychological symptoms including depression, anxiety, body image, and self-efficacy. Outcome measures included instruments like PHQ-9, GAD-7, GHQ-28, and the Rosenberg Self-Esteem Scale. Data on risk of bias (assessed via Cochrane RoB 2), sample size, retention, adherence, and effect sizes were extracted and compared. Meta-analysis was planned contingent on data sufficiency.

Results: Included studies showed variable but generally positive effects of AI chatbot interventions on depressive and anxiety symptoms, self-esteem, and body image. Effect sizes ranged from small to moderate (Cohen's d ~0.1-0.6), with some sustained improvements at follow-up. Common limitations included small sample sizes, short follow-up durations, reliance on self-reported data, and moderate risk of bias due to lack of blinding and attrition. Retention and adherence rates varied widely across studies.

Conclusion(s): AI chatbot interventions represent a promising, scalable, and acceptable modality for supporting adolescent mental health, with preliminary evidence indicating symptomatic improvement. However, effect sizes remain modest and heterogeneity in study design and quality limits definitive conclusions. Future research should prioritize larger sample sizes, extended follow-up durations, and rigorous randomized designs to confirm efficacy, optimize user engagement, and clarify differential impacts among diverse adolescent populations.

Figure 1: Flow diagram illustrating the study selection process based on PRISMA 2020 guidelines for the systematic review.
Figure 1: Flow diagram illustrating the study selection process based on PRISMA 2020 guidelines for the systematic review.

Table 1. Characteristics of the included studies. Table 2. Summary of statistical analyses of the included studies.
Table 1. Charcteristics of the included studies.
Table 2. Summary of statistical analyses of the included studies.


Figure 2. Forest plot displaying the effect sizes of the included studies.
Figure 2. Forest plot displaying the effect sizes of the included studies.