Answers are grounded in this paper and cite their sources.
Conversational artificial intelligence now belongs to ordinary adolescent life, but the psychological research that should guide its design spreads over fields which seldom cite each other. Research on adolescent development, on misinformation susceptibility, on trust in automation, on chatbot-supported learning and on companion-chatbot overreliance has grown up separately, and each of these areas answers only part of the question of how a general-purpose chatbot should talk to a teenager. This paper brings together twenty-one sources from developmental psychology, human factors, human-computer interaction and educational technology, in order to show what is now known about adolescent chatbot interaction and what this means for conversational design. Three results appear repeatedly in this research. First, adolescence is a design context of its own, not a smaller version of adulthood, since reward sensitivity, social influence and emotional regulation are still developing, while exposure to persuasive conversational systems has already reached almost everyone. Second, trust in a chatbot changes behavior in ways that can be measured; thus, the proper design target is calibrated trust rather than maximal trust. Third, the failure modes that matter most for adolescent users, that is, hallucination, sycophancy and unwarranted confidence, originate in the model; the user is not their source. From these results this paper consolidates ten conversational design principles into four clusters, namely epistemic calibration, metacognitive scaffolding, bounded emotional support and autonomy preservation. The framework here is a conceptual synthesis and puts forward testable predictions, but it has not been empirically validated.
No takes yet. Share an insight, caveat, or question.
Linxi Chen (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: