To examine associations between young people’s use of generative AI for emotional support and emotional issues, compared with AI use for schoolwork.
Approach:
Study and participants: Cross-sectional analysis of Ontario Health and Peer Relations Study data collected from May 2025 through March 2026 across four Ontario school boards; 39,761 students in grades 4–12 participated (mean age, 12.8 years).
Measures and analysis: Researchers assessed AI use for emotional or interpersonal support and for schoolwork. Emotional issues were measured using seven items from the Ontario Child Health Study Emotional Behavioral Scales. Models adjusted for demographic characteristics, loneliness…
Key Findings:
About 21% of students reported using AI for emotional support or interpersonal advice; 43% reported using AI for school-related tasks at least once.
Clinically significant emotional issues were reported by 58% of students who used AI for emotional support, compared with 29% of nonusers.
After adjustment, affective AI use remained associated with a 26% higher prevalence of clinically significant emotional issues and with higher continuous emotional-problem scores.
Functional AI use had a smaller association with emotional-problem scores, which increased from 0.01 among students reporting a little use to 0.04 among those reporting a lot of use, compared with students who never used AI for schoolwork.
Affective AI use was reported more often by female students (27%) and gender-diverse students (34%) than by male students (14%). It increased across elementary, middle, and secondary school levels (16%, 21%, and 26%, respectively), and was more frequent among …
The association between affective AI use and emotional issues was attenuated among gender-diverse students; it did not differ between female and male students. School level did not modify the association.
Interpretation:
The study authors reported that AI use for emotional support was more strongly associated with emotional issues than AI use for schoolwork. Lead author Kamilla Bonnesen and colleagues wrote that clinicians should normalize screening for affective AI use as a marker for underlying mental health concerns.
Limitations:
The cross-sectional design cannot establish temporal sequence or causality.
Affective AI use was measured as use versus nonuse, without information on frequency, duration, or dose-response.
Measures were self-reported; self-selection and residual confounding were possible.
Data on socioeconomic status and mental health treatment history were unavailable.
Findings about general-purpose AI may not generalize to specialized clinical AI agents.
Conclusion:
The study found an adjusted association between using generative AI for emotional support and emotional issues among students, but could not determine whether distress preceded or followed that use.
