72% of Teens Use AI Chatbots; Mental-Health Risks Grow

Adolescent use of artificial intelligence for emotional support has reached a scale that clinicians, parents, and technology developers cannot ignore. According to research cited in an August 2026 syndicated health report, 72% of teens have interacted with AI companion chatbots, and between 13% and 20% turned to AI for emotional concerns within a two-month window in 2025. Equally important, 63% of teens and young adults who used an AI chatbot for mental health advice did not tell anyone. These figures signal a profound shift in help-seeking behavior: a generation is testing mental-health support through general-purpose large-language-model chatbots that were neither designed nor validated for that purpose. The behavior is not hard to understand. At 2 a.m., an anxious teenager can get an immediate, nonjudgmental response from a chatbot without waking parents or waiting for a therapy appointment. ChatGPT, Claude, and Gemini can simulate empathy and sustain engagement, which makes them feel supportive. Unlike social media, which connects teens to peers, these tools are private simulated companions. However, the comfort they provide exists in a clinical vacuum. A small number of purpose-built therapeutic chatbots have received research attention, but most teens are using general-purpose consumer products. The risks are layered. First, general-purpose models can produce plausible but harmful or invalid mental-health advice, and they are not built to reliably detect suicidality, self-harm, or other crisis indicators. Second, the private nature of these conversations, reinforced by the 63% nondisclosure rate, means parents, pediatricians, and mental-health professionals may have no visibility. Third, teen users may develop dependence or substitute chatbot interactions for human connection or formal care. Fourth, data generated in these conversations may be stored, analyzed, or reused in ways families do not understand. The authors, who describe themselves as clinical psychologists and a child psychiatrist working with adolescents, wrote 2025 guidelines for clinicians on addressing teen chatbot use safely; they are now turning to parents and trusted adults. From a health-system perspective, this is not a fringe phenomenon. The 13% to 20% two-month utilization rate suggests millions of adolescents may use AI for emotional concerns each quarter in the U.S. alone. Yet most pediatric primary care and school health workflows do not routinely screen for AI chatbot use. The authors recommend that clinicians ask about chatbot use alongside screen time, sleep, mood, and self-harm. For parents, the guidance is likely to emphasize recognizing patterns, asking nonjudgmental questions, and staying involved rather than banning tools. For healthcare and technology leaders, the development raises urgent questions about accountability. These AI platforms market general-purpose assistance, so manufacturers can disclaim medical use. But the data show teens are repurposing them as mental-health tools regardless of terms of service. That creates a regulatory and ethical gray area for digital health, with unvalidated mental-health support being delivered outside HIPAA, outside FDA oversight, and outside clinician supervision. It may invite scrutiny from regulators and class-action attorneys if harms emerge. Looking forward, three paths are probable. AI developers will face increasing pressure to embed age-aware safety features, crisis referral pathways, and clear limitations on mental-health claims. Health systems and schools will need protocols to screen for and address hidden chatbot use. Researchers will need longitudinal data on whether AI emotional support improves outcomes, worsens isolation, or changes the trajectory of formal mental-health care. The trajectory is likely upward, and 2025 established the baseline; 2026 and beyond will determine whether the response catches up. For now, the clearest risk is not that teens talk to machines, but that they do so without validation, without disclosure, and without the safeguards that human support systems exist to provide.