A cross-sectional study published in JAMA Pediatrics surveyed 39,761 students in grades 4 to 12 across four Ontario school boards between May 2025 and March 2026. 21% — 8,408 students — reported using AI chatbots for emotional support or advice. Within that group, 57.7% met the study's threshold for serious emotional problems, against 29.2% among students who did not. After adjustment for demographics, loneliness, a sense of mattering and academic AI use, a gap of 26% higher adjusted prevalence remained.
What the design can and cannot support
This is the part that will be lost first. The study is cross-sectional: everything is measured at one point in time, so it cannot establish direction. The authors say so explicitly — the data cannot show whether distress drives young people to AI or whether using AI worsens distress. Reverse causation is at least as plausible as the headline reading: a struggling teenager seeks out something available at 3am that does not tell their parents.
Three ways the numbers get mangled
"57.7% of AI users are mentally unwell" is wrong — the figure covers students who go to a chatbot for emotional support, not everyone who uses AI. The study deliberately separates emotional from academic use, and the association holds on the emotional variable; merging them destroys the finding. And the sample is four school boards in one Canadian province, which does not extrapolate to adolescents generally.
Why the adjusted figure is the interesting one
The raw contrast, 57.7 against 29.2, is close to double and will carry the coverage. The number that survives controls is 26%. That drop is itself informative: most of the raw gap is explained by loneliness, mattering and demographics — variables that describe who reaches for a chatbot in the first place.
What it will be used for
This is one of the first population-scale samples on affective chatbot use among school-age children, and that scale is what regulators and school boards cite. It will be quoted in support of restrictions. Whether the design supports that use is a separate question from whether it will be put to it.