Affective Generative Artificial Intelligence Use and Youth Mental Health.

Abstract

Importance Adolescents are increasingly using generative artificial intelligence (AI) as an anonymous emotional support system, yet the association between affective use and mental health remains poorly understood. Objective To examine the association between affective generative AI use and emotional problems among a large, diverse sample of youth. Design, Setting, and Participants This cross-sectional study evaluates Ontario Health and Peer Relations Study data (May 2025 through March 2026) among students in grades 4 through 12 across 4 school boards in Ontario, Canada. Data were analyzed in May 2026. Exposures Affective AI use (emotional support or advice) and functional AI use (school-related tasks). Main Outcomes and Measures Emotional problems were assessed via the Ontario Child Health Study Emotional Behavioral Scales using a validated clinical threshold. Covariates included mattering, loneliness, and demographic characteristics. Results The final analytic sample consisted of 39 761 students (mean [SD] age, 12.8 [2.4] years; 19 291 [48.5%] girls). A total of 8408 participants (21.1%) reported using AI for affective purposes. Female students, gender-diverse students, and students who were Black, Indigenous, or of other racial and ethnic groups besides White were more likely to have used AI for affective purposes compared with boys and White students, respectively. Additionally, students who were older and students who reported using AI for school reported incrementally higher affective AI use. In unadjusted models, affective AI use was associated with significantly higher emotional problem scores (mean [SD] score, 1.04 [0.54] with use vs 0.67 [0.51] without use; B = 0.37 [95% CI, 0.36 to 0.39]) and were twice as likely to have clinical emotional problems (57.7% vs 29.2%; prevalence ratio, 1.98 [95% CI, 1.91 to 2.04]). In fully adjusted models, affective AI use remained associated with emotional problems (B = 0.14 [95% CI, 0.13 to 0.15]), whereas functional AI use showed a modest dose-response association with emotional problems (ranging from B = 0.01 [95% CI, 0.00 to 0.02] with rare use to B = 0.04 [95% CI, 0.02 to 0.05] with a lot of use). A significant interaction between affective AI use and gender identity was observed, which was attenuated for gender-diverse students compared with boys (B = -0.07 [95% CI, -0.12 to -0.03]) and girls (B = -0.08 [95% CI, -0.13 to -0.03]). No significant interaction was observed between affective AI use and school level. Conclusions and Relevance This cross-sectional study suggests that seeking emotional support from generative AI is a unique marker of psychological distress in children and adolescents, independent of mattering or loneliness. Clinical frameworks and digital literacy programs should distinguish between functional AI assistance and the use of algorithmic interfaces as a digital refuge for emotional needs.