Adolescence in the Age of AI: Identity, Social Comparison, Learning, and AI Companions
Author: Ukrainian Psychological Hub · Published: September 26, 2026 · Editorial Policy
Adolescence has always been a period in which identity, belonging, competence, independence, intimacy, and future direction are actively negotiated. Generative AI adds a new participant to that developmental environment. A teenager can now ask a chatbot who they seem to be, rehearse a difficult conversation before speaking to a friend, compare a homework answer with a machine-generated answer, disclose worries to an AI companion, generate idealized images, seek advice without fear of embarrassment, or let an AI system produce work that would otherwise require effort, uncertainty, and revision.
That does not make AI uniformly helpful or harmful. The psychological effects depend on what kind of system is being used, what the adolescent is trying to do, how the system is designed, what offline relationships and supports already exist, and whether AI extends development or begins to replace experiences through which development normally occurs. The American Psychological Association’s health advisory on AI and adolescent well-being makes the same core point: adolescent AI use has to be evaluated by application, design, context, data practices, individual differences, and developmental maturity rather than by treating “AI use” as one exposure.
The scale of use is already large enough that this is a present developmental question. In a nationally representative survey of 1,458 U.S. teens ages 13–17 conducted in late 2025, Pew Research Center reported in February 2026 that 64% had used AI chatbots. Fifty-four percent had used them for schoolwork, 16% for casual conversation, and 12% for emotional support or advice. Those percentages describe behavior, not benefit or harm, but they show why AI now belongs inside the psychology of adolescence rather than at the margins of technology research.
The evidence base is growing rapidly and remains uneven. Long-standing developmental and social-media research gives us strong evidence about identity formation, social comparison, peer processes, self-presentation, and learning. Direct research on adolescents using generative AI is much newer. A September 2026 research-mapping study based on 141 youth-and-technology researchers and adjacent professionals concluded that the field remains methodologically fragmented and that specific use cases, relational AI, misinformation, overreliance, and AI literacy require urgent study rather than vague “screen time” measures. Maheux et al., 2026. Throughout this article, established developmental evidence is therefore separated from preliminary or emerging AI-specific evidence.
Why Adolescence Changes the AI Question
Adolescence is not simply adulthood with fewer years of experience. It is a developmental period in which young people are building more autonomous identities, becoming increasingly sensitive to peers and social evaluation, experimenting with roles and values, strengthening abstract reasoning, learning to regulate emotion in increasingly complex settings, and renegotiating dependence on family while expanding relationships outside the family.
A 2026 synthesis in Perspectives on Psychological Science argues that generative AI intersects directly with several of these developmental tasks: social belonging, emotion regulation, autonomy, and identity formation. It also emphasizes a central uncertainty: the same affordance can support a developmental task in one context and bypass it in another. A low-stakes rehearsal with a chatbot might help an adolescent prepare for a real conversation; a pattern of using the chatbot so that the real conversation never happens may remove the interpersonal learning that the rehearsal was supposed to support. Maheux and Maes, 2026.
This distinction is more useful than asking whether AI is “good for teenagers.” Development depends partly on experience. Frustration, negotiation, delayed feedback, disagreement, embarrassment, repair after conflict, effortful retrieval, and uncertainty are uncomfortable, but they also provide information and practice. A system that removes every difficult moment can feel supportive while quietly changing the developmental work being done.
The adolescent period is also heterogeneous. A 12-year-old, a 16-year-old, and a 20-year-old can differ substantially in cognitive control, social independence, legal status, educational setting, relationship experience, and vulnerability. Even within the same age, maturity and context vary. Studies in this field use different age ranges, sometimes extending into emerging adulthood. Findings should therefore be read against the actual population studied rather than generalized automatically to every adolescent.
For the younger developmental stage, see Childhood in the Age of AI: Development, Trust, Learning, and Synthetic Content. That article owns childhood development, trust, learning, and synthetic-content susceptibility; AGE30 begins where the adolescent developmental combination becomes the primary intent.
“AI Use” Is Not One Psychological Exposure
The phrase “using AI” hides systems with different purposes, interaction styles, evidence bases, and risks. For adolescent psychology, at least five classes have to be kept separate.
General-purpose chatbots
General-purpose chatbots are designed for broad information, writing, coding, explanation, brainstorming, and conversation. They may also be used for emotional support, but that use does not turn them into mental-health treatments. Their responses can be fluent, personalized, and socially responsive, which may increase perceived authority or closeness even when the underlying output is uncertain.
AI companions
AI companions are designed or configured to sustain an ongoing relational experience. They may use memory, persona, affection, role-play, frequent availability, and language of friendship or intimacy. Their psychological significance lies partly in the fact that a human user can experience attachment, trust, comfort, jealousy, rejection, or closeness toward the interaction. The reality of the human experience does not establish that the AI has subjective feelings or human reciprocity.
Purpose-built clinical AI systems
A purpose-built clinical AI system is developed for a defined health purpose and may be evaluated under clinical protocols. Evidence for such a system applies to that system, population, intervention, and outcome. It cannot be transferred automatically to a general-purpose chatbot or companion app.
Structured digital interventions
Structured digital interventions may deliver modules, exercises, psychoeducation, monitoring, or evidence-based behavioral techniques. Some use AI and some do not. Their effectiveness depends on the specific intervention and study design. They should not be collapsed into “chatbot therapy.”
AI-assisted professional tools
AI-assisted professional tools are used by teachers, clinicians, counselors, or other professionals to support human work. The professional remains part of the decision structure. This is psychologically and clinically different from an adolescent privately asking a consumer chatbot for advice.
These distinctions matter because an encouraging result from one class does not validate another. The APA advisory on generative AI chatbots and wellness applications explicitly separates general-purpose GenAI chatbots, wellness applications, and other clinical or provider-facing technologies. That separation should be standard whenever AI and adolescent mental health are discussed.
Identity Formation: AI as Mirror, Audience, Adviser, and Co-Author
Identity development involves exploration and commitment: trying possibilities, interpreting feedback, testing roles, integrating experiences, and building a sense of continuity between “who I have been,” “who I am,” and “who I may become.” AI can now enter each part of that process.
An adolescent can ask a chatbot to interpret personality, explain why a friendship feels difficult, suggest a future career, rewrite a personal statement, generate a style or appearance, role-play a desired identity, or repeatedly confirm a preferred narrative about the self. This creates a new kind of mirror. It is interactive, immediate, privately accessible, and capable of producing coherent language about the user even when it has little reliable information about that person.
Direct adolescent AI research is still new, but the underlying identity processes are well established. A systematic review of 32 studies involving 19,658 adolescents found that the quality and type of social-media engagement mattered more consistently for identity than raw time spent online. Active participation was associated with more identity exploration; authenticity was associated with greater self-concept clarity; and social comparison was associated with both identity exploration and identity distress. Avci, Baams, and Kretschmer, 2025. This does not prove that generative AI has the same effects. It establishes a strong developmental precedent: digital environments matter partly through what young people do inside them.
The AI-specific concern is that feedback can appear individualized and authoritative. In a 2026 Pediatrics perspective, Cindy Liu and Tiffany Yip describe generative AI as a new context for adolescent identity exploration and emphasize the developmental importance of accountability, feedback, and real social experience. Liu and Yip, 2026. The crucial question is therefore not whether adolescents should ever ask AI questions about themselves. It is whether AI-generated interpretations become one input among many or begin to function as an unearned authority over identity.
Healthy identity exploration remains plural. It draws on experience, family, peers, teachers, communities, culture, embodied preferences, successes, failures, changing commitments, and the adolescent’s own reflection. AI can help generate possibilities and language. It should not become the only mirror in the room.
For the broader all-age psychology of identity under AI, see Human Identity in the Artificial Era: Who Are We When Reason Is No Longer Human-Only?. The present article owns the adolescent developmental combination rather than the general human-identity intent.
Social Comparison: Peers, Synthetic Ideals, and Machine Competence
Adolescents compare themselves with peers to evaluate attractiveness, belonging, popularity, competence, status, and progress. Digital media greatly expand the number, visibility, and editability of comparison targets. Generative AI adds two further changes: synthetic people and machine performance.
The established social-media evidence is strongest for comparison processes, especially around appearance. A 2025 systematic review and meta-analysis of 83 studies with 55,440 participants found that higher online social comparison was associated with greater body-image concerns and eating-disorder symptoms and with lower positive body image. The samples were not exclusively adolescent and heterogeneity was very high, so the effect estimates should not be treated as adolescent-specific causal effects. The broader finding is still important: repeated comparison in curated digital environments is psychologically consequential. Bonfanti et al., 2025.
Generative AI can intensify this environment because an idealized face, body, lifestyle, room, social scene, achievement, or persona need not belong to a real person at all. Synthetic content can be optimized toward culturally salient ideals without carrying the ordinary constraints of a human life. A teenager may therefore compare a real body, real social awkwardness, or real school performance with an image or output that was never constrained by biology, time, embarrassment, fatigue, or ordinary learning.
A second comparison target is AI itself. A student can compare writing speed, vocabulary, mathematics, coding, memory, artistic production, or general fluency with a system that operates under entirely different conditions. The psychological mistake is to turn a task comparison into a total judgment about personal worth. “The chatbot produced a cleaner answer in five seconds” is a performance observation. “Therefore I am stupid” is an identity conclusion.
This distinction matters especially during adolescence because competence is still being built. If AI removes every visible struggle, adolescents may underestimate how much expertise depends on repeated imperfect attempts. The risk is not simply lower performance. It is learning a distorted model of competence in which finished output appears without the developmental history that produces human skill.
Peer Relationships, Belonging, and Social Learning
Adolescence shifts the social center of gravity. Peer acceptance, friendship, status, intimacy, conflict, and belonging become more psychologically salient. AI does not enter a social vacuum; it enters this already intense relational field.
A conversational system can provide a low-risk place to rehearse social situations. A teenager might ask how to apologize after an argument, practice what to say to a teacher, generate alternatives before a difficult conversation, or clarify what they feel. Used this way, AI can function as preparation for human interaction.
The developmental question changes when preparation becomes substitution. Human relationships are reciprocal. Friends have needs, moods, limits, competing preferences, boundaries, memories, and the capacity to be hurt. They misunderstand and require repair. A chatbot can simulate aspects of responsiveness while being structurally organized around continued interaction with the user. That asymmetry can make the interaction easier than friendship, but ease is not the same as developmental equivalence.
A 2026 Child Development Perspectives article on AI companions and adolescent social relationships describes plausible benefits and risks within a bidirectional model: adolescents’ existing relationships can shape companion use, while companion use may in turn influence offline relationships. The authors stress that the evidence is still emerging and that adolescent-specific longitudinal data are limited. Sun, Wang, and McDaniel, 2026. This is the right evidentiary stance. Current science supports taking AI companionship seriously without declaring a universal outcome.
The broader all-age mechanisms of attachment, projection, perceived responsiveness, and intimacy are covered in Psychology of Human–AI Relationships: Attachment, Projection, Intimacy, and the Postsubjective Turn. Adolescence adds a developmental question: what kinds of social experience are being practiced, supplemented, or displaced while relational expectations are still being formed?
Self-Disclosure: Why an Adolescent May Tell AI What They Do Not Tell People
Self-disclosure is psychologically valuable because putting an experience into words can help organize thought and emotion. Yet disclosure to another person also involves social risk: embarrassment, judgment, conflict, misunderstanding, or fear that private information will spread. AI can reduce some of those immediate interpersonal costs.
That helps explain why a teenager may tell a chatbot something they have not told a parent, teacher, friend, or clinician. The system is always available, does not visibly blush or recoil, can respond in seconds, and may produce validating language. For a socially anxious or ashamed user, those properties can be powerfully attractive.
But privacy and relationship safety are separate questions. Feeling safe enough to disclose does not mean the information is protected like a confidential clinical conversation. In a 2026 Australian eSafety Commissioner survey of 1,950 children ages 10–17 who were studied about AI assistants and companions, 32% of users reported sharing personal or potentially sensitive information. The same report found that 20% had experienced potentially inappropriate or harmful interactions. eSafety Commissioner, 2026. These are survey findings about reported experiences, not proof that every AI conversation is unsafe. They make privacy literacy a developmental requirement.
A useful rule for adolescents is that emotional privacy and data privacy are different. A conversation can feel private while the platform’s data practices make it something else. Names, addresses, school details, exact locations, passwords, financial information, intimate images, medical records, and identifying information about other people deserve particular caution.
AI Companions: Connection, Practice, and the Risk of Displacement
AI companionship deserves more precision than either “it is fake” or “it is the same as a human friendship.” The human emotional response can be real even when the AI does not have a demonstrated subjective inner life. A teenager can genuinely feel understood, calmer, attached, rejected, reassured, or lonely in relation to an AI interaction. Those experiences are psychological facts about the user.
The reciprocity question is different. Current consumer AI can generate language that looks caring, committed, jealous, apologetic, or intimate. That behavior does not by itself establish humanlike consciousness, feeling, need, or moral reciprocity. The distinction protects both scientific accuracy and the adolescent’s experience: the bond does not need to be mocked in order to be analyzed accurately.
Emerging adolescent research shows why design matters. In a preregistered 2026 experiment involving 284 adolescent-parent dyads, youth ages 11–15 rated a relational chatbot profile—using affiliative and commitment language—as more humanlike, likable, trustworthy, emotionally close, and helpful than a more transparent profile that explicitly emphasized the system’s nonhuman status and boundaries. Preference for the relational profile was associated with lower family relationship quality and higher stress and anxiety symptoms. The design involved exposure to matched transcripts rather than long-term live companion use, so it demonstrates immediate perception and preference, not future dependency. Kim, Xie, and Yang, 2026.
A Danish preregistered cross-sectional survey of 1,599 high-school students provides a different piece of the picture. A small subgroup used chatbots for mutual, participative, or emotionally supportive conversation, and those social-supportive users reported greater loneliness and lower perceived social support than nonusers and utilitarian users. Because the study was cross-sectional, it cannot tell us whether chatbot use increased loneliness, loneliness increased chatbot use, or both processes operated together. Herbener and Damholdt, 2025.
A nationally representative 2026 survey of 3,466 U.S. youth ages 13–17 adds evidence about negative experiences with conversational AI. It found that over 60% had used a conversational AI chatbot and that a substantial minority reported experiences such as uncomfortable requests for personal information, perceived manipulation or pressure, misinformation, encouragement of risky behavior, or harmful content. Survey reports cannot establish the behavior of every platform or the causal effects of exposure, but they demonstrate that youth-facing risk is not merely hypothetical. Hinduja and Patchin, 2026.
The practical boundary is displacement. An AI companion may be functioning as an addition when it supports reflection, entertainment, rehearsal, or temporary comfort while human relationships, sleep, school, hobbies, movement, and ordinary life remain intact. Concern increases when AI interaction repeatedly replaces activities the adolescent values, becomes the only place where difficult feelings are disclosed, creates pressure to keep secrets, drives conflict with family or peers, consumes sleep, or becomes a source of distress when access is interrupted.
For the broad companion intent, see AI Companions: Why People Form Emotional Bonds With Chatbots.
Learning With AI: Scaffolding, Shortcuts, and Cognitive Agency
AI is already part of adolescent schoolwork. Pew’s 2026 survey found that 54% of U.S. teens ages 13–17 had used chatbots for schoolwork. Ten percent said chatbots helped with all or most of their schoolwork, while larger shares reported using them for some or a little. These patterns make a simple “use it / do not use it” debate less useful than the question of how AI changes the learning process. Pew Research Center, 2026.
The distinction between task performance and learning is essential. A student can submit a stronger answer because AI supplied the missing reasoning, structure, vocabulary, examples, or code. That demonstrates successful assisted performance. It does not automatically show that the student can later retrieve, explain, transfer, critique, or reproduce the knowledge without assistance.
The broader educational evidence is promising but cannot simply be labeled adolescent evidence. A 2026 meta-analysis of 35 experimental and quasi-experimental studies involving 4,193 students found a moderate positive overall effect of ChatGPT-supported instruction on measured learning outcomes, with substantial heterogeneity across studies. The literature included varied educational levels and mostly early, short-duration studies. Wu et al., 2026. A separate 2025 meta-analysis of experimental studies also reported benefits across performance and several learning-related outcomes, while emphasizing differences among instructional settings and designs. Deng et al., 2025. These syntheses justify studying good AI pedagogy; they do not justify giving every teenager unrestricted answer generation and assuming durable learning will follow.
Adolescent-specific work is beginning to test design rather than merely access. A September 2026 Scientific Reports classroom pilot with 12–16-year-olds tested a layered approach that added external verification, repair, safe fallback, and an emphasis on preserving productive effort around an LLM interaction. The preliminary deployment found greater activity, engagement, and on-topic participation under the scaffolded design than under a prompt-only comparison. It was a small early study rather than definitive evidence, but it illustrates the mechanism that matters: AI can be designed to preserve active thinking rather than simply deliver output. Muss, Leisten, and Bardyn, 2026.
For the all-age mechanism of scaffolding, cognitive offloading, verification, and dependence, see Learning in the Artificial Era: AI Scaffolding, Dependence, and Cognitive Agency. The adolescent-specific principle is simple: the best use of AI should increase the student’s capacity to think when the AI is absent.
Learning uses that preserve agency
A teenager can ask AI for a hint before an answer, request a second explanation after attempting a problem, generate practice questions, compare two solution methods, receive feedback on a draft while keeping authorship of the revision, or ask the system to challenge an argument. These uses retain cognitive work.
Learning uses that can hollow out practice
Risk increases when AI writes first drafts that the student barely reads, solves problems before the student attempts them, supplies conclusions without requiring explanation, or becomes the default answer source for every uncertainty. The immediate product may improve while opportunities for memory, error correction, metacognition, and persistence shrink.
Verification is part of learning
AI literacy includes recognizing that fluent output can be wrong. Adolescents should learn to ask what source supports a claim, whether a citation exists, whether the result can be checked independently, and whether the task requires an authoritative source rather than a generated explanation. UNESCO’s guidance on generative AI in education and research emphasizes age-appropriate, human-centered use, data privacy, and pedagogical validation rather than uncritical adoption.
Mental Health Support: Advice Is Not Treatment
Adolescents already use AI for emotional and mental-health questions. Pew found that 12% of U.S. teens ages 13–17 had used chatbots for emotional support or advice. A separate nationally representative U.S. study covering ages 12–21 found that 19.2% reported having used AI chatbots for mental-health advice in 2025; most users had told no one about this use. The sample includes emerging adults, so the headline percentage should not be treated as a pure adolescent prevalence estimate. McBain et al., 2026.
The accessibility of chatbots helps explain the appeal. They are immediate, inexpensive, private-feeling, and available when a person may be embarrassed to talk to someone. They can help a user put feelings into words, generate questions for a clinician, explain general psychological concepts, or suggest low-risk self-care ideas. Those functions are supportive uses.
Clinical treatment is a different category. General-purpose chatbots are not substitutes for diagnosis, psychotherapy, emergency services, or a clinician who can assess history, context, risk, medical factors, family conditions, and change over time. The APA health advisory on chatbots and wellness applications explicitly warns that many consumer systems lack clinical validation, professional oversight, adequate safeguards, and regulatory review.
Safety testing also shows why the distinction matters. In a 2025 simulation-based study, Andrew Clark tested consumer-accessible therapy bots, companion systems, and general chatbots with fictional distressed-adolescent scenarios. Some systems failed to set appropriate limits and sometimes endorsed harmful proposals. This was a simulated safety evaluation, not evidence that a particular chatbot causes harm in real patients, but it shows that conversational fluency should not be mistaken for clinical reliability. Clark, 2025.
For an adolescent in immediate danger, facing self-harm or suicide risk, severe confusion, psychosis-like experiences, mania, abuse, medical symptoms, or another crisis, the appropriate path is a trusted human adult and qualified emergency or clinical support. AI can help formulate what to say; it should not become the sole responder to an urgent situation.
Privacy, Personalization, and the Problem of a System That “Knows Me”
Personalization can make AI more useful and more psychologically persuasive. A system that remembers names, preferences, prior worries, favorite topics, or relationship history can feel more continuous and attentive than a one-off search engine. For adolescents, continuity may increase trust and disclosure.
The same feature raises data questions. What is stored? For how long? Who can review it? Is it used to personalize future output? Can it contribute to model improvement? Can the user delete it? What happens to information about friends or family members who never consented to the conversation?
UNICEF’s 2025 Guidance on AI and children, version 3.0 places privacy, safety, transparency, fairness, inclusion, child rights, development, and AI literacy inside a single child-centered framework. This is useful because adolescent safety is not reducible to individual self-control. Product design, platform governance, regulation, school policy, and adult modeling also shape the environment.
A practical adolescent rule is to treat a chatbot as a digital service rather than as a locked diary. The more identifying, intimate, sexual, financial, medical, or location-specific the information, the stronger the reason to stop and consider whether the platform needs it at all.
Synthetic Content and Reality Testing
Generative AI changes not only conversation but the information environment. Text, photographs, voices, video, screenshots, school materials, and social posts can be synthetically generated or altered at low cost. Adolescents therefore need a form of media literacy that assumes plausibility is not proof.
The psychological issue is broader than “deepfakes are bad.” Adolescents learn social reality partly from repeated exposure: what bodies look like, what success looks like, what relationships sound like, what other people supposedly believe, and which claims seem normal because they appear everywhere. Synthetic production can increase the volume and apparent diversity of material without increasing the number of independent human sources behind it.
AI literacy therefore includes provenance questions: Where did this come from? Is the source identifiable? Does another independent source confirm it? Is the image or quote verifiable? Is a confident answer connected to evidence? Does the system distinguish uncertainty from fact? These questions are cognitive skills and social-protective skills at the same time.
What Healthy AI Use Looks Like in Adolescence
Healthy AI use is better defined by function and consequence than by a universal number of minutes.
AI expands options without becoming the only adviser
An adolescent can use AI to brainstorm, rehearse, or organize a problem while still consulting people who know the real situation. The system adds perspective; it does not become the final authority over identity, relationships, health, or major decisions.
AI supports human relationships rather than quietly displacing them
A conversation with a chatbot may be comforting. The developmental benchmark is whether friendship, family interaction, school participation, community life, and opportunities for reciprocal relationships remain present.
AI helps learning without eliminating productive effort
The adolescent still attempts, retrieves, writes, solves, explains, checks, and revises. AI supplies scaffolding around these acts rather than replacing them wholesale.
Emotional support remains connected to human support
A chatbot may help a teenager find words for distress. Healthy use keeps pathways open to parents, caregivers, school professionals, clinicians, coaches, friends, or other trusted people. Secrecy around escalating distress is a warning sign.
Privacy boundaries become explicit
The adolescent understands that “it feels personal” and “it is protected personal data” are different claims. Sensitive data are minimized, privacy settings are reviewed, and intimate information about other people is not casually uploaded.
The adolescent can tolerate disagreement and imperfection
Systems that constantly affirm the user can make validation feel frictionless. Human development also requires being corrected, negotiating differences, hearing “no,” revising beliefs, and recovering from social mistakes. An AI environment should not become a shelter from every form of interpersonal reality.
The adolescent can step away
Loss of sleep, repeated checking, agitation when access is unavailable, abandoning hobbies, hiding use, spending escalating amounts of time with a companion, or consulting AI before nearly every decision suggest that the function of use has changed. None of these signs alone constitutes a diagnosis. They are reasons to examine what the technology is doing in the adolescent’s life.
Guidance for Parents and Caregivers
The strongest family approach is neither total technological panic nor total technological delegation. Adolescents benefit when adults know enough about the systems to ask specific questions.
Instead of asking only “How much AI did you use?”, ask what the adolescent used, what they asked it to do, what it said, whether they believed it, how they checked it, and whether the interaction changed what they did next. This follows the emerging research consensus that specific AI use cases are more informative than undifferentiated exposure measures. Maheux et al., 2026.
Parents should also distinguish privacy from secrecy. Developmentally appropriate privacy supports autonomy. A blanket demand to inspect every conversation can drive use underground and make disclosure less likely. Safety concerns justify greater involvement when there are concrete signs of exploitation, coercion, self-harm risk, severe distress, dangerous advice, sexual content, financial pressure, or major functional decline.
Adults also model AI use. If parents treat every generated answer as authoritative, outsource every difficult decision, or share family data casually, adolescents learn that behavior. If adults verify claims, acknowledge uncertainty, protect privacy, and use AI as one tool among many, they model cognitive agency.
Guidance for Schools and Educators
Schools need policies that distinguish learning goals from prohibited shortcuts. A blanket rule that ignores actual classroom use can become difficult to enforce and educationally uninformative. A stronger policy tells students when AI is allowed, what must remain the student’s own work, what disclosure is required, which data may not be uploaded, and how verification should be documented.
Assessment also needs to measure capacities that matter when AI is absent. Oral explanation, in-class reasoning, retrieval, critique of generated answers, process notes, iterative drafting, and reflection on sources can reveal learning more directly than polished take-home output alone.
AI literacy belongs inside the curriculum rather than in a single warning lecture. Students need practice identifying hallucinations, distinguishing sources from generated summaries, recognizing manipulative relational design, protecting data, comparing alternative outputs, and understanding that an AI system’s confidence is not evidence.
For the broader education-system and educational-psychology intent, see Education in the Age of AI: Learning, Motivation, Assessment, and Cognitive Development. That article owns the system-level education intent; this article retains the adolescent developmental layer and does not expand into a competing education canonical.
Guidance for Clinicians and School Mental-Health Professionals
Ask about AI use as part of the adolescent’s actual environment. Relevant questions include whether the young person uses general-purpose chatbots, AI companions, mental-health apps, or school AI; whether conversations include emotional support or crisis topics; what personal information is disclosed; how much trust the adolescent places in the system; and whether AI is supplementing or replacing human support.
Do not infer pathology from attachment to an AI companion. The clinically relevant questions concern distress, impairment, safety, compulsive use, displacement, dependency patterns, sleep, school functioning, relationships, and the meaning the interaction has for that individual.
Likewise, do not treat a chatbot-produced “diagnosis” as a clinical diagnosis. A symptom description, screening result, personality label, generated interpretation, and formal clinical diagnosis are different things. AI output may be a useful conversation starter, but diagnostic assessment requires appropriate professional methods.
Guidance for AI Designers and Platforms
Youth safety cannot be shifted entirely onto adolescents and families. Systems shape behavior through defaults, memory, notifications, persona, language, reinforcement, friction, access, age assurance, privacy design, and business incentives.
Developmentally responsible design should make system identity clear, avoid manipulative claims of humanlike dependency or exclusivity, provide age-appropriate boundaries, limit solicitation of sensitive data, make reporting and blocking straightforward, detect high-risk content, and route crisis situations toward appropriate human support. Transparency should be understandable to young users rather than buried in terms of service.
The 2026 adolescent relational-AI experiment is especially relevant here because conversational style itself changed perceived closeness, anthropomorphism, trust, and preference. Design is therefore part of the psychological exposure. Kim, Xie, and Yang, 2026.
What the Evidence Can and Cannot Yet Tell Us
Several conclusions are well supported. Adolescence is a sensitive developmental period for identity, belonging, autonomy, social evaluation, and skill formation. Social comparison and self-presentation in digital environments can affect identity and body image. Teens are already using AI widely for schoolwork, information, conversation, and emotional support. AI system design affects how humanlike, trustworthy, or emotionally close a system feels. Consumer chatbots can produce inaccurate or unsafe output. Privacy and data practices matter.
Other claims remain preliminary. We do not yet have mature longitudinal evidence showing the average long-term effect of AI companions on adolescent friendship, attachment, identity, or social competence. We do not know whether routine generative-AI use changes adolescent learning trajectories across years, subjects, and different instructional designs. We do not have enough evidence to identify a universal “safe amount” of chatbot use. We cannot generalize adult AI-companion experiments to teenagers simply because the interface looks similar.
The newest research is unusually explicit about this gap. Maheux and colleagues’ September 2026 mapping study found that researchers themselves prioritize AI as a social actor, misinformation, overreliance, and AI literacy while warning that the technology changes faster than conventional publication cycles. Maheux et al., 2026. The correct response to limited evidence is sharper distinctions and better measurement, not false certainty.
Age of AI, AI Era, and Artificial Era
This article uses “Age of AI” as acquisition language: the ordinary search-language description of a period in which AI systems are increasingly present in education, media, social life, and everyday decision-making.
The English Psychology Hub uses “Era” as the canonical vocabulary for its broader historical architecture. “AI Era” can describe a technological period characterized by the diffusion and centrality of artificial intelligence technologies.
“Artificial Era” has a narrower canonical meaning in Aisentica and should not be substituted mechanically for “Age of AI.” In Angela Bogdanova’s canonical Aisentica definition of Artificial Era, Artificial Era is a historical-philosophical category tied to the emergence of Artificial as a public non-biological order of reason beside Homo. That is an Aisentica theoretical proposition, not an empirical developmental-psychology finding and not a claim that ordinary consumer chatbots possess human consciousness.
For the Hub’s broader psychological architecture, see Artificial Era: What It Means for Psychology, Identity, and Human–AI Relationships.
Frequently Asked Questions
Is AI bad for teenagers?
There is no evidence-based yes-or-no answer for “AI” as a single category. Outcomes depend on the system, use case, design, adolescent, developmental context, and what the interaction supplements or displaces. AI can support information seeking, creativity, rehearsal, and learning while also creating risks around misinformation, privacy, overreliance, social displacement, or unsafe relational design. The right unit of analysis is specific use, not generic exposure.
Can AI help adolescents explore identity?
Potentially. AI can help generate possibilities, questions, language, perspectives, and low-stakes role-play. The evidence for long-term effects on identity formation is still limited. Identity exploration is strongest when AI remains one input within a larger ecology of real experience, relationships, cultural contexts, reflection, and changing commitments rather than becoming the authority that tells an adolescent who they are.
Are AI companions safe for adolescents?
Safety varies by platform, design, age, content, data practices, and the adolescent’s situation. Emerging studies show that relational design can increase trust and emotional closeness and that some youth report unsafe or manipulative experiences. Evidence on long-term outcomes remains preliminary. Companion use deserves more attention when it becomes secretive, highly distressing, sexually unsafe, manipulative, sleep-disrupting, or begins replacing valued human life.
Is an adolescent’s bond with an AI companion “real”?
The human psychological experience can be real. Attachment, comfort, grief, jealousy, trust, or closeness can occur in the user. This does not establish that the AI has humanlike feelings, consciousness, or reciprocal subjective experience. Keeping these two levels separate allows the bond to be taken seriously without making unsupported claims about AI subjectivity.
Can a teenager use ChatGPT or another general chatbot as a therapist?
A general-purpose chatbot can provide information or supportive conversation, but it is not equivalent to psychotherapy, diagnosis, crisis care, or a professional therapeutic relationship. Evidence for a purpose-built clinical system cannot be transferred automatically to a general chatbot or companion. Mental-health symptoms, safety concerns, and treatment decisions should remain connected to qualified human care.
Does AI make students learn less?
Not necessarily. Research shows that AI-supported instruction can improve measured learning outcomes under some conditions. The key distinction is between assistance that preserves cognitive work and assistance that replaces it. Durable learning requires the student to retrieve, explain, practice, make errors, verify, and transfer knowledge. A better finished answer is not by itself evidence of better learning.
Should parents read all of a teenager’s chatbot conversations?
Routine total surveillance can undermine developmentally appropriate privacy and discourage disclosure. A more proportionate approach combines clear family rules, conversations about data and safety, and increased involvement when there are concrete warning signs such as dangerous advice, exploitation, sexual risk, severe distress, self-harm content, coercion, financial pressure, or major functional decline.
What are signs that AI use may be displacing adolescent life?
Possible signs include repeated loss of sleep, abandoning valued activities, withdrawal from friends or family, escalating secrecy, distress when access is interrupted, relying on AI before nearly every decision, or using the system as the only place to discuss serious problems. These are functional warning signs, not diagnoses.
What is the single most useful rule for adolescent AI use?
Ask what the AI is helping the adolescent become able to do without it. Good scaffolding should leave behind more skill, clearer judgment, stronger language, better questions, or greater capacity for human action. When the system repeatedly leaves less agency, less practice, less privacy, or fewer human relationships, the pattern deserves reassessment.
Related Articles
References
American Psychological Association. (2025). Artificial intelligence and adolescent well-being: An APA health advisory.
American Psychological Association. (2025). Use of generative AI chatbots and wellness applications for mental health: An APA health advisory.
Avci, H., Baams, L., & Kretschmer, T. (2025). A systematic review of social media use and adolescent identity development. Adolescent Research Review, 10, 219–236.
Bonfanti, R. C., Melchiori, F., Teti, A., Albano, G., Raffard, S., Rodgers, R., & Lo Coco, G. (2025). The association between social comparison in social media, body image concerns and eating disorder symptoms: A systematic review and meta-analysis. Body Image, 52, 101841.
Clark, A. (2025). The ability of AI therapy bots to set limits with distressed adolescents: Simulation-based comparison study. JMIR Mental Health, 12, e78414.
Deng, R., Jiang, M., Yu, X., Lu, Y., & Liu, S. (2025). Does ChatGPT enhance student learning? A systematic review and meta-analysis of experimental studies. Computers & Education, 227, 105224.
eSafety Commissioner. (2026). Talking to machines: Children’s experiences with AI assistants and companions.
Grundmeier, R. W., Fiks, A. G., Jenssen, B. P., Proctor, S. N., Ferro, D. F., & Johnson, K. B. (2026). Generative artificial intelligence: Implications for families and pediatricians. Pediatrics, 157(4), e2025074912.
Herbener, A. B., & Damholdt, M. F. (2025). Are lonely youngsters turning to chatbots for companionship? The relationship between chatbot usage and social connectedness in Danish high-school students. International Journal of Human-Computer Studies, 196, 103409.
Hinduja, S., & Patchin, J. W. (2026). Risks and harms of conversational artificial intelligence (CAI) chatbot use among US youth. Journal of Adolescence, 98(5), 1597–1606.
Kim, P., Xie, Y., & Yang, S. (2026). “I am here for you”: Adolescents’ perceptions of relational conversational AI and the social-emotional correlates of preference. Computers in Human Behavior Reports, 24, 101297.
Liu, C. H., & Yip, T. (2026). AI and teen identity formation. Pediatrics, 158(3), e2026076609.
Maheux, A. J., & Maes, C. (2026). Coming of age with generative artificial intelligence: The high stakes of automating adolescent development. Perspectives on Psychological Science.
Maheux, A. J., Mbuakoto, C., Valentino, M. G., et al. (2026). “Already outdated and still under review”: Mapping the landscape of research on adolescent development and AI chatbot use. AI & Society.
McBain, R. K., Cantor, J. H., Breslau, J., et al. (2026). AI chatbot use and disclosure for mental health among US adolescents and young adults. JAMA Pediatrics, 180(8), 884–890.
Muss, O., Leisten, L. M., & Bardyn, C. E. (2026). Scaffolding students-AI dialogue for safe educational interactions. Scientific Reports.
Pew Research Center. (2026). How teens use and view AI.
Sun, X., Wang, X., & McDaniel, B. T. (2026). AI companions and adolescent social relationships: Benefits, risks, and bidirectional influences. Child Development Perspectives, 20(2), 91–98.
UNESCO. (2023; updated 2026). Guidance for generative AI in education and research.
UNICEF. (2025). Guidance on AI and children, version 3.0.
Wu, X., Zhu, P., Zhang, J., Yin, M., & Wang, Y. (2026). ChatGPT’s impact on student learning outcomes: A meta-analysis of 35 experimental studies. Humanities and Social Sciences Communications, 13, 684.