Self-Compassion As a Neurotype-Specific Lever for AI Self-Disclosure In Adolescents: Implications for Stepped-Care Digital Mental Health

Background and Purpose: Generative AI chatbots are emerging as informal mental health entry points for adolescents because they offer private, low-stakes, always-available interaction. Yet adolescents may not disclose to AI for the same reasons. A key intrapersonal resource may be self-compassion, which can reduce shame and support clearer expression of difficult experiences. However, self-compassion likely operates within the broader context of adolescents’ self-generated thought. Patterns of self-generated thought, operationalized here as daydreaming styles, may shape what becomes available for disclosure: positive-constructive thoughts may support reflective sharing, whereas guilt- and fear-laden or poorly controlled thoughts may prompt disclosure for relief or offloading. These pathways may differ by ADHD status because adolescents with ADHD often experience more spontaneous mind-wandering, poorer attentional control, and lower self-compassion than their peers. This study therefore examined whether self-compassion functions as a neurotype-specific resource for AI self-disclosure, and whether its role differs depending on adolescents’ daydreaming styles.

Methods: We analyzed cross-sectional online survey data from 2,115 adolescents ages 13 to 17 in the United States and Hong Kong. Overall, 38.2% self-reported an ADHD diagnosis or ADHD traits. AI self-disclosure was coded dichotomously from open-ended responses about what participants would discuss with generative AI. Daydreaming styles were measured using the Short Imaginal Processes Inventory, including positive-constructive daydreaming, guilt/fear-of-failure daydreaming, and poor attentional control. Self-compassion was assessed using the Self-Compassion Scale for Youth. Multigroup structural equation modeling with a robust weighted least squares estimator tested whether pathways to AI self-disclosure differed between adolescents with and without ADHD, including interaction terms between self-compassion and each daydreaming style, while adjusting for age, gender, region, and meaning-centered coping.

Results: Approximately 30% of adolescents in both groups reported self-disclosing to AI, with no significant difference by ADHD status (29.0% with ADHD vs 30.1% without ADHD). The key difference was not whether adolescents disclosed to AI, but what predicted that disclosure. Among adolescents without ADHD, positive-constructive daydreaming was associated with greater AI self-disclosure (β = .236, p < .001), whereas this pathway was not significant among adolescents with ADHD (β = .055, p = .173; group difference Δχ²(1) = 4.26, p = .039). Self-compassion further distinguished the groups: among non-ADHD adolescents, higher self-compassion strengthened the link between positive-constructive daydreaming and AI self-disclosure (interaction β = .124, p < .001), suggesting that self-compassion supported more intentional, constructive disclosure. Among adolescents with ADHD, self-compassion did not strengthen disclosure in any comparable way. Across both groups, higher self-compassion attenuated disclosure associated with guilt/fear-of-failure daydreaming and poor attentional control, indicating that its function was more protective than facilitative for distress-driven disclosure.

Conclusions and Implications: Equal rates of AI self-disclosure did not indicate equal mechanisms of engagement. Self-compassion supported constructive AI disclosure among adolescents without ADHD, but not among adolescents with ADHD. Step 0 AI tools should therefore not be designed as neurotype-neutral entry points; adolescents with ADHD may need structured and executive-function-sensitive supports to translate inner experience into safe, help-oriented disclosure and connection to care.