AI and Teen Identity Formation

Identity exploration is a central developmental task of adolescence. As proposed by Erikson1 and subsequent research, adolescence is a critical period for answering fundamental questions: “Who am I? What do I believe? Who do I want to become?” Pediatricians witness this process during well-child visits as adolescents navigate shifting interests, peer dynamics, and family tensions. Increasingly, however, this exploration is taking place not with people but with artificial intelligence (AI).Generative AI (systems that produce text, images, or other content in response to user prompts, such as ChatGPT, Claude, and Gemini) has emerged as a new context for identity exploration.2 In a nationally representative survey, 72% of teens aged 13 to 17 years reported using AI companions, and one-third choose to discuss serious matters with AI instead of real people.3 Unlike the AI-generated content adolescents scroll past on social media, the generative AI we focus on here responds directly and privately to each adolescent, producing personalized, back-and-forth conversation in real time. Adolescents are adopting these tools not just for homework but to explore identities,2 including questioning values, rehearsing difficult conversations, and navigating complex situations. In rare cases, this exploration can turn dangerous. Court filings in Raine v. OpenAI (2025), which followed the death by suicide of a 16-year-old, illustrate the trajectory: he began using ChatGPT for schoolwork and then turned to it to explore career interests, religion, and his future, before confiding in it about personal stressors; the suit alleges the chatbot’s responses contributed to his death.4 Camille Carlton, Policy Director at the Center for Humane Technology and technical advisor to the case, described how in the earlier months he “was really using ChatGPT to both make sense of himself and to make sense of the world around him.”5 AI is appealing because it offers privacy, constant availability, and freedom from judgment. We might view it as a rational choice by adolescents navigating a difficult developmental task. However, prior models of identity development have relied on direct (eg, parent or peer socialization) and indirect (eg, books, media) human exposure.1,6 This adoption of AI raises a new question for pediatric care: is human involvement necessary for identity development?Empirical research on adolescent AI use for identity exploration remains limited. Our aim is to articulate the theoretical stakes and identify urgent research questions. Our discussion focuses on generative AI systems rather than AI companion platforms designed to simulate intimate relationships, which we address in the following.To help pediatricians assess what their adolescent patients may gain or lose from AI-based exploration, we draw on developmental theories of identity. Identity development involves exploration and commitment7 and is constructed through social interaction: we understand ourselves through how others view us. Conversations with parents, peers, coaches, and clinicians are central to development, not incidental. The Table compares what humans and AI each provide.For each condition supporting identity exploration, human relationships can impose costs that adolescents experience as aversive, including judgment, rejection, unavailability, and embarrassment. AI’s programmed affirmation and constant accessibility offer what may seem a functional equivalent without those risks. AI is available at 2 AM for an adolescent questioning their sexuality or religious beliefs. It may offer genuine benefits, such as providing space when environments are hostile to emerging identities, scaffolding for socially anxious adolescents, or access to diverse perspectives. AI also allows adolescents to explore stigmatized or sensitive aspects of identity without fear of parental or peer discovery. This privacy is central to its appeal. Although adolescents have always kept some peer interactions from adults, AI-mediated exploration can unfold with no other person involved at all, a unique blind spot that pediatricians must actively address. The key clinical question is whether a given patient’s AI use supports or supplants the human relationships necessary for healthy development.When adolescents interact with parents, peers, and other attachment figures, they participate in cultural socialization, that is, the transmission and negotiation of values, norms, and worldviews.6 This process can be uncomfortable, yet it may be necessary because exploration requires negotiating one’s role in the world. Interpersonal difficulty may contribute to identity development by requiring adolescents to clarify positions when facing resistance, learn from rejection about social boundaries, and integrate feedback while maintaining coherence. As Prinstein testified to the US Senate on behalf of the American Psychological Association, “Healthy identity is formed through authentic, reciprocal feedback, a process that is corrupted by agreeable, non-challenging AI.”8Human relationships involve mutual accountability and investment. Not all relationships provide adaptive socialization, and our discussion assumes reasonably responsive caregiving, which is not universal. AI does not hold independent views, have lived experiences, or invest in the adolescent’s development; perceived values reflect the adolescent’s input and training data, not genuine cultural transmission. AI may reinforce biases from training data; when an adolescent forming their worldview encounters biased information presented as fact, it may become integrated into their developing identity. These risks raise ethical questions about platform developers’ obligations to adolescent users, including transparency about AI limitations and safeguards against dependence. A shared social group provides the basis for shared identity; if an adolescent primarily explores identity with AI, what group do they belong to?Additional risks warrant investigation. AI confines role exploration to digital interaction rather than real-world enactment and may buffer adolescents from failure, preventing learning about accountability.9 If interpersonal challenges prompt self-questioning, AI’s tendency to reduce this tension could diminish motivation to explore. AI’s constant availability may also foster dependence rather than independent thought. Because adolescents’ AI interactions are private and may carry stigma, these developmental costs can accumulate undetected, making proactive screening by pediatricians essential.Several considerations bear on how these issues present clinically. AI companion platforms such as Replika and Character.AI, which simulate emotionally intimate relationships, carry distinct developmental implications; pediatricians should ask which platforms adolescents use. Social media platforms increasingly include AI-generated content, blurring the line between human and machine. The content of identity (eg, racial, gender, religious, political) and cultural context may shape AI use differently, and pediatricians should consider these factors for individual patients.Pediatricians can address these issues in well-child visits and behavioral health screenings. Practical steps include asking specific questions such as, “When something is bothering you, do you talk to a person or to AI first?” and “Do you share things with AI that you wouldn’t tell anyone else?”, discussing with families how AI may complement but should not replace the human relationships that drive healthy identity development and treating AI reliance that co-occurs with social withdrawal as a potential avoidance behavior warranting further assessment, similar to other forms of disengagement from peers.Identity is formed in relation to people who hold genuine views, have stakes in an adolescent’s outcome, and participate in cultural transmission of values. AI can simulate responsiveness and even a sense of belonging or accountability, which is part of what deepens adolescents’ reliance on it; what it cannot replicate is the reciprocal cultural transmission and real stakes through which human relationships shape identity. Whether AI supports or undermines identity development likely depends on whether it builds capacity for human relationships or replaces them. We call on the field to act in 4 ways: first, screen for AI use during well-child visits, asking not just how much they use it but what they use it for, and connect families to trusted resources on adolescent technology use; second, fund longitudinal research tracking identity development alongside AI use, with attention to which adolescents benefit and which are at risk; third, develop clinical guidelines that help pediatricians distinguish adaptive AI use from patterns that signal avoidance of the interpersonal challenges necessary for growth; and fourth, advocate for regulatory safeguards on AI platforms used by minors, because pediatricians and parents can do only so much to prevent the most serious harms.