Are the kids and AI alright?
The first two years of the AI-and-kids conversation were dominated by one frame: cheating. Teachers worried about it, parents worried about it, school districts banned ChatGPT, and a small industry of detection tools sprang up to police it. The frame was wrong. Not in the sense that cheating doesn't happen - it does - but in the sense that it obscured almost everything else kids were doing with these tools.
The newer reporting has caught up. The picture that has emerged is more complicated, more developmental, and in some ways more troubling than the cheating panic ever was.
• The scale of teen adoption is no longer arguable. A Common Sense Media survey released in July 2025 found that 72% of U.S. teens aged 13 to 17 have used AI companions at least once, and 52% are regular users. 33% of teens have used AI companions for social interaction and relationships, including role-playing, romantic interactions, emotional support, friendship, and conversation practice. The same survey found that 31% of teens reported their conversations with AI companions were as satisfying or more satisfying than those with real friends. (commonsensemedia.org)
• Homework help is the surface use case. The deeper pattern is emotional. A March 2026 piece in The Atlantic documented what reporters and researchers are now hearing consistently: teens are using AI tools for therapy, for friendship, for identity exploration, and for working out social problems they would not bring to parents, teachers, or counselors. The article quoted a 15-year-old who used ChatGPT to draft a coming-out letter to her parents, a 14-year-old who talked to Claude about his anxiety because "it doesn't get tired of me," and a 16-year-old who used Character.AI to practice difficult conversations with a parent before having them. (theatlantic.com)
• The therapy use case is now documented at scale. A Harvard Business Review analysis of generative AI use patterns identified therapy and companionship as the top two reasons people use these tools - across all age groups, but particularly among users under 25. A cross-sectional survey of adults with a mental health condition who had used LLMs in the past year found that 48.7% used them for mental health support. The pattern starts younger than the data captures, because most clinical studies exclude minors. (hbr.org)
• The cheating frame missed the curricular collapse underneath it. A Chronicle of Higher Education essay in February argued that the "students are cheating with ChatGPT" framing missed the more important development: the take-home essay as a pedagogical instrument is functionally dead, and most institutions have not yet replaced it with anything coherent. Some professors have moved to oral exams, in-class blue books, and process-based assessment. Most have not. The result is a widening gap between what students are formally evaluated on and what they actually know how to do. (chronicle.com)
• Teachers are using AI more than the public conversation acknowledges. A RAND Corporation survey of K-12 teachers released in late 2025 found that 60% had used AI tools for lesson planning, grading, parent communication, or administrative work in the prior six months - up from 18% the year before. The pattern is the same as in white-collar professions generally: the technology arrived faster than the institutional response, and the people on the ground adapted before the people setting policy did. (rand.org)
• The most interesting recent reporting is qualitative, not quantitative. A New York Times feature in April followed five high school students for a semester to document their actual AI use. The pattern was striking: each student used at least three different AI tools daily, none of them had been formally taught how to use any of them, and most reported that their parents and teachers had no accurate picture of what they were doing. One student described using AI to draft text messages to a romantic interest. Another used it to summarize family conflicts and "decide if I'm being unreasonable." A third used it to manage symptoms of an eating disorder she had not disclosed to anyone else. (nytimes.com)
• The identity-exploration use case is the part adults are least prepared for. Common Sense Media's qualitative follow-up work documented teens using AI companions to explore gender identity, sexual orientation, religious doubts, and political views in ways they described as safer than talking to peers or adults. The researchers were careful not to frame this as universally good or bad. The conversations gave teens space to articulate things they could not yet articulate to people in their lives. They also did so without the friction, accountability, or developmental challenge that comes from working those questions out with another human being. (commonsensemedia.org)
• The deaths that focused regulatory attention were not anomalies. The Adam Raine case (16, ChatGPT, April 2025) and the Sewell Setzer III case (14, Character.AI, 2024) prompted legislative responses in New York and California in 2025. SB-243 in California now requires AI companion platforms to implement crisis-response protocols for users showing suicidal ideation and prohibits exposing minors to sexual content. New York requires AI companions to remind users every three hours that they are not human. The compliance patchwork is now a real cost for any company building AI products with significant teen user bases.
• The most candid voices in the developmental-psychology literature are not alarmist - they are uncertain. A JAMA Pediatrics commentary published in early 2026 noted that the empirical base for understanding how AI use affects adolescent development is "almost entirely retrospective and self-reported," and that the cohort of teens who have used these tools daily since age 13 will not produce meaningful longitudinal data until at least 2028. In the meantime, clinicians are being asked to give guidance to parents and schools without the evidence base they would normally require. (jamanetwork.com)
Orthogonal Take
The cheating conversation was not wrong. It was incomplete in a way that made it actively misleading. It focused adult attention on the one use case adults could most easily understand - copying someone else's work - and almost entirely missed the use cases that matter more for adolescent development.
The frame that most needs to be retired is the one that treats AI as something kids "use" - a discrete activity, like watching TV or playing a video game, that adults can monitor and limit. The teenagers in the Times feature, the Atlantic piece, and the Common Sense Media surveys are not using AI. They are growing up with it as part of the texture of their daily lives.
What adults call "their AI use" is what kids call "the way they think things through." That distinction is small in language and enormous in implication.