Large language model chatbots in adolescent and young adult mental health information and support: a scoping review

Adolescents and young adults may use digital sources for early mental health information and support, but evidence concerning generative artificial intelligence (GenAI) chatbots remains fragmented. This scoping review mapped GenAI chatbots, predominantly large language model (LLM)–based, used for mental health information and support among adolescents and young adults or evaluated through youth-relevant mental health scenarios. Following the Joanna Briggs Institute (JBI) guidance and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR), six databases were searched for peer-reviewed studies published from 2020 to December 31, 2025, supplemented by citation chasing. Twenty-one studies were included: 13 user studies and eight system-output evaluations, 17 explicitly identified an LLM, whereas four confirmed generative functionality without reporting the underlying model. The evidence addressed informational, emotional, structured therapeutic or self-help, and screening, triage, or crisis-related functions. User studies mainly reported perceived helpfulness, acceptability, engagement, disclosure, and short-term within-person outcomes, but most were brief, uncontrolled, or observational. System-output evaluations identified variable response quality, unsafe or misleading guidance, inconsistent crisis handling, and misclassification in higher-stakes scenarios. Safeguards were rarely evaluated directly. Adolescent-specific evidence, sustained clinical outcomes, referral completion, and real-world safety remained limited. The findings support cautious, function-specific interpretation and the need for age-sensitive controlled and longitudinal research, transparent system reporting, and direct evaluation of safeguards and escalation pathways.