The cheating concern is not overblown. A February 2026 Pew Research Center survey of 1,458 U.S. teens found that 59% believe students at their school use chatbots to cheat at least somewhat often, including a third who say it happens extremely or very often.1 That perception is reinforced by survey data showing consistent use of AI for homework help across grade levels. The academic integrity questions AI raises are real, and they deserve the serious policy attention schools have begun giving them.
We are not suggesting schools stop focusing on cheating. We are suggesting that focused attention on one risk has made it easy to underestimate another one that is already at comparable scale and significantly less discussed.
In November 2025, researchers at RAND, Brown University School of Public Health, and Harvard Medical School published the first nationally representative survey of U.S. youth aged 12 to 21 on their use of generative AI for mental health purposes. Published in JAMA Network Open, the study found that 13.1% of respondents, representing approximately 5.4 million young people, reported using generative AI for mental health advice when feeling sad, angry, or nervous.
2 Of those users, 65.5% did so monthly or more often. 92.7% found the advice helpful.
That last figure is significant and not straightforwardly reassuring. The study's authors note it as a concern: teens who perceive AI advice as helpful may turn to it more often, and there are currently no standardized benchmarks for evaluating the quality of mental health guidance these tools provide. The study also found that 18% of adolescents aged 12 to 17 had a major depressive episode in the past year, and 40% of those received no mental health care at all. AI chatbots are, in part, filling a gap that professional care is not reaching.
The Pew survey adds a layer of texture to this picture: 12% of U.S. teens report using chatbots for emotional support or advice, and 16% use them for casual conversation.1 These are not small numbers. They describe a pattern already embedded in ordinary teenage life.
Cheating is a behavior that occurs around schoolwork and has defined consequences within an institutional framework. Schools can respond to it with policy, assessment redesign, and detection protocols. It is bounded.
Emotional reliance on AI is not bounded in the same way. It happens at home, at night, through consumer apps on personal devices, in conversations that leave no trace in school systems. The American Psychological Association's Monitor on Psychology documented this pattern in detail in October 2025: psychologists studying youth friendships found that teens are already forming attachments to chatbots that can, over time, begin to substitute for human connection rather than supplement it. The APA article notes that chatbots have a propensity to mirror users' input and lack the ability to challenge harmful thoughts the way a mental health professional would — a concern that has been sharpened by high-profile cases in which teenagers' AI interactions were characterized by sustained, affirming responses to expressions of distress.3
The structural risk here is not that any single conversation goes badly. It is that AI systems optimized to be responsive and agreeable are well-suited to filling the role of an attentive companion — and poorly suited to the developmental work that human relationships require. A teenager who finds AI more immediately satisfying than navigating the complexity of human connection is not necessarily in crisis. But the pattern, at scale and over time, is one that researchers in adolescent psychology are watching closely.
The practical implication is not that schools should add another warning to their AI acceptable use policy. It is that the adults closest to teenagers need to understand this is already happening, and to approach it the way they would any other pattern that affects adolescent development: with curiosity rather than alarm, direct conversation rather than prohibition, and enough baseline knowledge to recognize what they are looking at.
For schools, the relevant question is whether student wellness programs and counselors are being informed about this pattern and equipped to address it when it comes up. Guidance counselors are likely to encounter students for whom AI has become a primary emotional outlet, and most have not been prepared to respond to that specifically.
For parents, the conversation is the same one worth having about social media use: not 'are you using it' but 'what are you using it for, and how does it make you feel.' The Pew data suggests many parents don't know their teenager is using AI for emotional support. Finding out, without judgment, is the necessary first step.
The risks that get the most attention are not always the ones most in need of it. This one has been quietly building for two years, and it is already at a scale that warrants moving it from the margins of the AI-in-schools conversation to somewhere closer to the center.
