757. Conversational Artificial Intelligence (AI) Use and Adolescent Mental Health in Hong Kong Chinese Youths: A Mixed-Methods Study and Clinical-Community Comparison

Abstract

Abstract Background Conversational artificial intelligence (AI) tools are being adopted rapidly by adolescents and are increasingly used for emotional support, advice, and problem-solving, yet evidence on how young people engage with these tools—and how engagement co-occurs with mental health symptoms—remains limited, particularly in Asian settings. Aims & Objectives This mixed-methods study aimed to characterise conversational AI engagement among Hong Kong Chinese adolescents and examine how engagement co-occurs with symptom burden, while centring adolescents’ perspectives on benefits, boundaries, and concerns. Objectives were to: (i) compare conversational AI engagement (use time and reliance) and mental health symptoms between clinically-recruited and school-based community adolescents; and (ii) explore adolescents’ motivations, perceived benefits, and concerns about conversational AI use through qualitative interviews to contextualise quantitative findings. Method Adolescents aged 12–18 were recruited from clinical services (N = 511) and school-based community settings (N = 513) in Hong Kong (total N = 1,024). Symptoms were assessed using Patient-Health-Questionnaire-9 (PHQ-9), Generalised-Anxiety-Disorder-7-Scale (GAD-7), and Insomnia-Severity-Index (ISI). Conversational AI engagement was assessed via self-reported usage time and reliance indicators. Group differences were examined using Mann–Whitney U tests with rank-biserial correlations, and associations were assessed using Spearman correlations. Semi-structured 1:1 interviews (n = 19) underwent qualitative content analysis, with integration focused on explaining quantitative patterns using adolescents’ lived accounts. Results Clinically-recruited adolescents reported higher symptom levels than community peers (PHQ-9 mean 9.21 vs 6.93; GAD-7 7.26 vs 5.82; ISI 8.62 vs 6.28). The clinical group also reported higher conversational AI reliance (median 6 vs 2; Mann–Whitney p < .001; rank-biserial r = 0.546). Across both groups, conversational AI engagement (self-reported use time and reliance) was positively associated with depressive symptoms and also showed associations with anxiety and insomnia (ps < .01), noting that temporal ordering cannot be inferred from cross-sectional data. Qualitative interviews contextualised these patterns with three themes: (1) “safe but soulless”—AI was experienced as accessible and non-judgemental for emotional offloading yet limited in emotional depth; (2) “the therapist versus the therabot”—some participants reporting seek self-initated mental health advice from AI when they lacked trust in, or felt hesitant to approach, mental health professionals, while still viewing clinicians as distinct from AI support; and (3) perceived concerns and guardrails—participants articulated concerns about overreliance, privacy, and possible displacement of real-life relationships and help-seeking. Mixed-method integration suggested that higher reliance may reflect a preference for immediate, low-stakes support during distress alongside active awareness of limits and risks. Discussion & Conclusions In a large Hong Kong Chinese adolescent sample, clinically-recruited youth reported greater conversational AI engagement alongside higher symptom burden. Conversational AI engagement co-occurred with higher depression, anxiety, and insomnia symptoms, though causal interpretations are not warranted. Adolescents’ accounts highlighted both perceived utility (accessibility, non-judgement) and perceived concerns (overreliance, privacy, potential displacement of human support), supporting the need for future developmentally-informed guidance and longitudinal research that can clarify temporal relationships and identify which adolescents may benefit—or be vulnerable—from conversational AI use.