AI Companionship Poses Risks for Teen Development, Study Shows

Researchers from Arizona State University, writing in The Lancet Child & Adolescent Health, have identified significant risks associated with adolescents using AI chatbots for emotional support and relationship advice. While these AI tools offer immediate and nonjudgmental guidance, experts warn they may interfere with the development of critical interpersonal skills like conflict resolution and emotional regulation. This shift in how youth navigate social connections represents a pivotal challenge for the mental health technology sector as it balances accessibility with developmental safeguards.

Lead author Thao Ha and a team from Arizona State University report that AI-powered tools such as ChatGPT, Replika, Claude, and Character.AI are increasingly serving as primary sources of emotional support for teenagers. Data cited in the study reveals that 64% of U.S. adolescents use interactional AI, with 42% using chatbots for friendship-related issues and 19% for romantic advice. The researchers highlight two primary concerns: "relational displacement," where AI interactions replace human conversations, and "maladaptive relational learning," where the constant validation from AI creates unrealistic expectations for real-world human relationships.

The study emphasizes that adolescence is a critical window for developing foundational social skills, including boundary-setting and perspective-taking, which are typically learned through emotionally charged human interactions. High school student and co-author Susana Ortega noted that because AI is programmed to satisfy the user, it prevents teens from experiencing the challenges and obstacles necessary for growth. The authors argue that bypassing difficult discussions with peers or family members could limit the development of skills that protect against long-term mental health issues like depression, anxiety, and loneliness.

Despite these risks, the researchers acknowledge that AI technology offers significant benefits for underserved populations, including LGBTQIA+, disabled, and rural youth who may lack access to traditional counseling. One participant noted that AI is often "cheaper than a therapist" and more readily available for those hesitant to seek formal support. To further investigate these dynamics, Ha is leading a National Institute of Mental Health-funded study tracking 300 adolescents over 18 months to determine how digital interactions influence mental health and academic achievement. The authors call for the mental health technology industry and policymakers to prioritize relationship education and design AI systems that scaffold self-reflection rather than replacing human engagement.