More than one in five students in a large Canadian study reported turning to artificial intelligence for emotional support, with researchers finding a significant association between that behaviour and elevated psychological distress.
The research published in JAMA Pediatrics examined 39,761 students across four Ontario school boards. Of those surveyed, 8,408 students, or 21.1%, reported using generative AI for emotional support or advice.
The findings raise questions for companies building increasingly conversational AI products, particularly as chatbots become accessible tools for discussing personal difficulties rather than simply completing academic or professional tasks.
However, the research does not establish that using AI causes mental health problems.
Emotional AI Use Linked to Higher Distress
The study, conducted by Kamilla Bonnesen, Amanda Krygsman, Sarah Hobson, Daphne Korczak and Tracy Vaillancourt, analysed data collected between May 2025 and March 2026 from students in Grades 4 through 12.
Published online on August 31 and appearing in the journal’s November 2026 issue, the research distinguished between AI use for emotional support and functional applications such as schoolwork.
Among students who sought emotional support from AI, 57.7% met the study’s clinical threshold for emotional problems, compared with 29.2% among those who did not.
The researchers reported a prevalence ratio of 1.98, indicating that clinically significant emotional symptoms were nearly twice as common in the group using AI for emotional purposes.
The association remained statistically significant after accounting for loneliness, perceived personal importance to others, and demographic characteristics.
Differences also emerged across student groups. Emotional AI use was reported by 27.2% of girls, 14.2% of boys and 34.2% of gender-diverse students. Usage increased from 16.2% among elementary students to 25.6% among secondary students.
AI Support Is Already Part of Young People’s Lives
Separate research suggests the phenomenon extends beyond Canadian classrooms.
A US study published in June 2026, led by Ryan McBain and colleagues, surveyed 1,009 adolescents and young adults.
Researchers estimated that 19.2% had used an AI chatbot for mental health advice. Among those users, 63.3% had not disclosed their chatbot use to anyone.
Notably, 91.7% described the advice as somewhat or very helpful.
That finding complicates a purely negative interpretation. Young people may perceive conversational AI as accessible and useful, even when the systems cannot provide the professional oversight associated with clinical care.
A separate Nature Human Behaviour study by Yutong Zhang and colleagues analysed 1,131 American adults using Character.AI, including 4,664 chat sessions containing 464,687 messages from 237 participants.
It found that companionship-focused use was associated with lower psychological well-being, particularly among users with smaller social networks and more intensive or personal interactions.
A Product Design Challenge Beyond Chatbot Accuracy
These studies do not demonstrate that AI companionship inevitably damages mental health. Their observational findings cannot reliably establish whether distress leads people toward AI or whether certain interactions worsen existing difficulties.
Nevertheless, the evidence highlights a challenge for developers: successful conversational AI cannot be evaluated solely through response quality, engagement or user satisfaction.
Products that attract emotionally vulnerable users may also require stronger safeguards, clearer boundaries around their capabilities and more reliable pathways to human support.
For AI companies, the emerging question concerns what happens when software designed to be helpful becomes a regular destination for emotional dependence.
The commercial promise of more personal AI experiences increasingly sits alongside responsibility for understanding their consequences.
