This cross-sectional study surveyed 39,761 students in grades 4 through 12 across four Ontario school boards between May 2025 and March 2026, mean age 12.8 years. It separated two kinds of use: functional AI use for schoolwork, and affective AI use, meaning turning to a chatbot for emotional support or advice.
Affective use was reported by 8,408 students, 21.1% of the sample. It was more common in girls, in gender-diverse students, and in students who were Black, Indigenous or of racial and ethnic groups other than White, and it rose with age.
The association with distress was large. Students reporting affective use scored a mean 1.04 on the emotional problems scale versus 0.67 without, and 57.7% versus 29.2% met the validated clinical threshold — a prevalence ratio of 1.98. After full adjustment, including for loneliness and for whether the child feels they matter to others, the association persisted (B = 0.14, 95% CI 0.13 to 0.15). Functional AI use showed only a weak dose-response association.
The authors' framing is the useful one: affective AI use is a marker of distress rather than, on this evidence, a cause of it. A cross-sectional survey cannot tell you which came first, and the obvious explanation — that distressed children look for help wherever it is anonymous and free — is at least as likely as the reverse.
What it gives a clinician is a screening question that did not exist a year ago. Asking an adolescent whether they talk to an AI about how they feel is quick, non-threatening, and identifies a group with roughly double the rate of clinical emotional problems.
- Ask adolescents whether they use AI for emotional support, not just for schoolwork — the distinction carries the signal
- Treat a yes as a prompt for further mental health assessment, not as a problem in itself
- Note the association survived adjustment for loneliness and for feeling that one matters, so it is not simply a proxy for isolation
- Distinguish functional from affective use when discussing screen habits with families
- This is cross-sectional: it identifies distressed children, it does not show that AI use causes distress
The statistics, in plain English
A prevalence ratio of 1.98 means affective AI users were about twice as likely to meet the clinical threshold. The adjusted coefficient of 0.14 is much smaller than the unadjusted 0.37, which tells you most of the crude association was explained by other factors — but not all of it, and the surviving association is precisely estimated because the sample is nearly 40,000. Precision is not the same as importance: with a sample this large, trivially small effects reach statistical significance, which is why the functional-use coefficients of 0.01 to 0.04 should be read as essentially nothing despite their narrow intervals. And a cross-sectional design fixes the direction of nothing.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for paediatrics, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free