More Than 8,400 Students Turned to AI for Emotional Support, and a Major Study Found a Troubling Pattern in Their Mental Health

An August 31 report from the New York Post highlighted a growing concern about children and teenagers turning to artificial intelligence chatbots for help with their feelings and relationships. The report centered on a study published the same day in JAMA Pediatrics, which examined the responses of nearly 40,000 students in Ontario, Canada. Researchers found that students who used generative AI for emotional support were substantially more likely to report clinically significant emotional symptoms.

The findings stand out because chatbots are increasingly being used for much more than schoolwork and simple questions. Young people can ask them about difficult friendships, disagreements with family members, loneliness, and other personal concerns without having to approach another person. However, the research does not establish that using AI caused the emotional problems observed among the students.

What the Researchers Actually Investigated

The research team examined data from the Ontario Health and Peer Relations Study, involving students in grades four through 12 across four school boards in Ontario. The relevant survey data were collected between May 2025 and March 2026, allowing researchers to investigate how young people were using generative AI and how that use related to their emotional well-being. The study distinguished between seeking emotional support and using AI for functional purposes, such as completing school-related tasks.

The final analysis included 39,761 students with an average age of 12.8 years. Researchers assessed emotional symptoms using the Ontario Child Health Study Emotional Behavioral Scales, which include a validated threshold for clinically significant problems. They also considered other factors, including loneliness, students’ sense that they mattered to other people, and demographic characteristics.

More Than One in Five Students Had Sought Emotional Help From AI

Of the students included in the analysis, 8,408 reported using generative AI for emotional support or advice. That represented 21.1% of the sample, meaning slightly more than one in five students had turned to AI for this purpose. The category included emotional concerns and personal advice, rather than simply asking a chatbot to explain a school assignment.

The finding also shows why discussions about AI and children increasingly extend beyond classroom performance. The sample covered younger children as well as teenagers, capturing a wide range of developmental stages. Because the study examined reported behavior rather than experimentally assigning students to use chatbots, it provides a snapshot of how AI use and emotional symptoms appeared together in this population.

The Difference in Emotional Symptoms Was Striking

The researchers found a substantial gap between students who reported using AI for emotional purposes and those who did not. Before adjusting for other factors, the average emotional-problem score was 1.04 among affective AI users, compared with 0.67 among students who did not report that type of use. This difference suggested that emotional AI use was associated with noticeably higher reported symptom levels.

The percentages crossing the study’s clinical threshold made that contrast even clearer. Approximately 57.7% of students who reported affective AI use met the threshold, compared with 29.2% of students who did not. The unadjusted comparison means that students in the first group were nearly twice as likely to have scores indicating clinically significant emotional problems, although the figures alone cannot explain why the difference existed.

Crossing the Clinical Threshold Does Not Mean Receiving a Diagnosis

The study’s use of a validated clinical threshold is important because it distinguishes elevated symptoms from ordinary differences in mood. Such a measure can help researchers identify students whose reported difficulties may warrant attention, but it does not establish that each student has a diagnosed mental health condition. The results therefore should not be interpreted as showing that nearly six in ten young people who use chatbots for emotional support have a particular psychiatric disorder.

The researchers were examining emotional-problem scores, not conducting individual clinical evaluations of every participant. Nor did the study test each student’s chatbot conversations to establish whether the advice they received was helpful, harmful, or appropriate. Its principal contribution was identifying a meaningful relationship between a particular type of AI use and a measure of emotional well-being.

Loneliness Did Not Fully Explain the Association

Young people who used AI for emotional support were also more likely to report loneliness and a weaker sense that they mattered to others. These factors offer important context because someone who feels isolated may be especially inclined to seek a response from a readily available chatbot. Researchers therefore considered whether those experiences could account for the relationship between affective AI use and emotional problems.

The association remained statistically significant after the researchers adjusted for loneliness, perceived mattering, and demographic characteristics. The adjusted relationship was smaller than the initial, unadjusted comparison, but it did not disappear. This suggests that those measured factors did not fully explain the pattern, although it still does not establish whether chatbot use contributes to emotional difficulties or whether distressed students are simply more likely to seek AI support.

AI Use for Schoolwork Showed a Different Pattern

The researchers also examined functional AI use, including using the technology for school-related tasks. That behavior showed a more modest, dose-related association with emotional-problem scores, with the relationship increasing slightly as reported use became more frequent. The findings therefore distinguish emotional support from other common reasons young people interact with generative AI.

This distinction matters because not every chatbot interaction serves the same purpose. Asking a system to explain a math problem is different from relying on it to interpret a painful disagreement with a friend or decide how to manage distress. The study does not establish that either kind of use causes emotional problems, but it suggests that researchers should examine why children use AI rather than treating all interactions as equivalent.

Who Was More Likely to Seek Emotional Advice From Chatbots?

The study identified differences in affective AI use across demographic groups. Girls, gender-diverse students, and students who identified as Black, Indigenous, or members of other racial and ethnic groups besides White were more likely to report such use than the respective comparison groups identified by the researchers. Older students and students who used AI for school-related tasks also reported higher levels of affective use.

Those findings require careful interpretation because a higher rate of chatbot use is not the same thing as evidence of greater harm. The researchers’ interaction analysis found that the association between affective AI use and emotional problems was attenuated among gender-diverse students compared with boys and girls. Consequently, the findings should not be simplified into a claim that every demographic group using AI more frequently necessarily experienced a stronger relationship with emotional symptoms.

Separate U.S. Research Suggests the Behavior Is Widespread

The Ontario results fit into a broader pattern emerging from research on young people’s use of AI for mental health advice. A separate nationally representative study published in JAMA Pediatrics examined a survey of 1,009 people ages 12 to 21 in the United States in November 2025. After population weighting, the researchers estimated that approximately 19.2% of the represented youth population had used AI chatbots for mental health advice.

That U.S. study also found that 42.8% of those who sought advice from chatbots did so at least monthly, while 91.7% described the advice as somewhat or very helpful. At the same time, 63.3% said they had not disclosed their AI use for mental health advice to anyone. These findings do not establish that AI advice was clinically beneficial, but they illustrate how frequently young people turn to chatbots and how often that behavior remains unknown to people around them.

Why Chatbots May Feel Easier to Approach Than People

An October analysis published by JAMA examining the Ontario findings explored why young people may find AI appealing for emotional support. Chatbots are available around the clock, often cost nothing to use, and can be approached without scheduling an appointment or explaining a concern face-to-face. For a teenager embarrassed about a relationship problem or worried about being judged, those characteristics may make a chatbot seem like a less intimidating first step.

That accessibility can be valuable for someone who needs help putting a difficult experience into words, particularly when access to human support is limited. Yet an always-available conversation can also create a risk if it becomes the main place a young person turns when distressed. Clinicians discussing the findings raised an important question: whether these interactions help children prepare to seek human support or gradually become a substitute for it.

A Supportive Reply Is Not the Same as Mental Health Care

The distinction between emotional reassurance and clinical support is central to understanding the concern. In a guidance article republished by the University of Massachusetts Chan Medical School, psychologists and a child psychiatrist explained that general-purpose chatbots are designed to produce engaging conversations, not necessarily to deliver validated mental health care. Their responses can sound empathetic while still giving inaccurate or inappropriate advice.

Some purpose-built therapeutic chatbots have been studied under controlled conditions, including tools designed around cognitive behavioral therapy. Those systems should not automatically be treated as equivalent to general-purpose chatbots that teenagers use for companionship or personal advice. The medical guidance also highlights weaknesses in crisis responses and the possibility that a confident answer could reinforce an unhealthy belief instead of helping someone find appropriate support.

The Research Cannot Determine Which Came First

Despite the striking statistical differences, the Ontario study cannot resolve the most important question of causation. Its cross-sectional analysis compared students’ reported AI use with their emotional symptoms, rather than following an experimental group to determine how its mental health changed because of chatbot interactions. That means researchers cannot establish from these results whether emotional distress preceded AI use, whether the interactions contributed to distress, or whether both processes occur for different students.

The data form part of a broader research effort, and further analysis of students over time may help clarify how the relationship develops. Researchers also need studies capable of examining particular kinds of interactions and measuring what happens to young people after they receive AI-generated advice. Without that evidence, claims that chatbots inevitably cause emotional harm would go beyond what this particular study demonstrates.

A Child Psychiatrist Warned About Replacing Human Support

The New York Post also reported concerns from Consuelo Cagande, a child and adolescent psychiatrist who was not part of the research team. Cagande warned that young people may increasingly substitute chatbot conversations for traditional support networks and established mental health services. She also raised concerns about inadequate safeguards, clinical follow-up, and crisis protections when a young person in distress relies on an AI system.

Those concerns reflect a practical problem beyond the statistical relationship itself. A chatbot may respond to a difficult disclosure, but it cannot automatically provide the same evaluation, accountability, or ongoing care as a qualified professional. If a young person receives misleading guidance or experiences a worsening crisis, the absence of a reliable connection to human support could become particularly consequential.

Parents and Educators Can Start by Asking Open Questions

The evidence gives parents, teachers, and school counselors a reason to understand how young people use AI without automatically assuming that every conversation is dangerous. The University of Massachusetts Chan Medical School guidance recommends approaching the subject with curiosity and without judgment, rather than beginning with an accusation or an immediate ban. Adults can ask what young people discuss with chatbots, whether any responses have confused or surprised them, and whether they turn to AI when they feel overwhelmed.

Those conversations can also help families establish sensible boundaries around chatbot use, including during moments of intense distress. Young people should understand that AI responses can be wrong, that simulated empathy is not a substitute for human care, and that sensitive conversations with commercial services should not automatically be assumed to be private. When emotional difficulties arise, trusted adults, school counselors, healthcare professionals, and other appropriate human supports should remain accessible options.

The Next Step Is Better Evidence, Not Simple Assumptions

The questions raised by the findings are also reflected in an accompanying JAMA Pediatrics editorial, “Do No Harm This Time—The Urgency of AI Chatbot Research”, by Dimitri Christakis and Ron Keren. The authors argued that rigorous evidence remains essential, but that waiting for complete certainty can carry costs when technologies with potential risks spread rapidly. Their discussion highlights the challenge of developing appropriate safeguards while research into children’s experiences with AI continues.

For now, the Ontario study establishes a clear association between using generative AI for emotional support and elevated emotional-problem scores among children and adolescents. It does not prove that chatbots caused those symptoms, nor does it show that every young person who uses AI for personal advice will be harmed. The most useful conclusion is that adults and researchers need to pay attention to why young people seek emotional support from AI, how they respond to its advice, and whether those interactions lead them toward meaningful human help or away from it.