The new first listener: Daydreaming styles and self‐compassion predict adolescent disclosure to AI , differently in ADHD

Abstract

AI chatbots are rapidly becoming a common place where adolescents first disclose personal experiences, because the interaction is private, always available, and often feels lower-stakes than telling another person. Recent U.S. evidence suggests that about one in eight adolescents and young adults already use generative-AI chatbots for mental health advice, which makes AI disclosure a realistic candidate for a stepped-care “Step 0” contact-an upstream, pre-service entry point that occurs before any formal assessment or treatment, and that could be bridged to professional help-seeking when need is indicated. The key question is not only whether youth disclose to AI, but which internal processes make AI disclosure more likely, and therefore how Step 0 should be interpreted across different young people. Daydreaming styles provide a tractable map of self-generated thought that should shape disclosure tendencies. Classic work differentiates positive-constructive daydreaming (pleasant imagery and future planning), guilty-dysphoric daydreaming (negative, guilt- and fear-of-failure content), and poor attentional control (fleeting thoughts and difficulty sustaining focus). These styles plausibly influence what becomes salient enough to externalise and the motive for using an AI “listener” (elaboration and meaning-making versus offloading distress or fragmented attention). Self-compassion then acts as a regulator of threat and shame: it is associated with reduced shame-proneness and more adaptive emotion regulation, which can lower the impulse to externalise distress for relief while supporting clearer articulation when inner material is constructive. This regulatory pattern may differ in ADHD, where spontaneous mind wandering and attentional control problems are pronounced. Using a large online survey of youth from the United States and Hong Kong (N = 2,115), we tested these pathways via structural equation modelling with multi-group analyses (ADHD vs peers). Youth with ADHD differed markedly in daydreaming styles and self-compassion, and the daydreaming to AI disclosure pathways diverged by group. Self-compassion was context-dependent: it predicted reduced AI disclosure in ADHD, but in the general population it amplified disclosure linked to positive-constructive daydreaming while constraining disclosure linked to guilty-dysphoric and poor attentional-control daydreaming.