摘要
Introduction Generative artificial intelligence (GenAI) companions increasingly respond to users' negative emotions with validation, reassurance, reframing, and personalized guidance. However, immediate emotional relief does not establish improved independent emotion regulation or emotional dependence. Methods In a two-condition randomized experiment with 380 adults, we compared adaptive regulation-oriented responses with relational reassurance-oriented responses. We assessed immediate negative affect, emotion-regulation self-efficacy, independently coded strategy quality, affective recovery during a subsequent no-AI task, perceived AI emotional necessity, and short-term AI-reliance intentions. Results Both response styles substantially reduced immediate negative affect, with no significant between-condition difference in relief ( b = 0.055, p = 0.323). Adaptive regulation-oriented responses were associated with higher emotion-regulation self-efficacy ( b = 0.500, p < 0.001), better independently coded strategy quality ( b = 0.353, p = 0.009), and greater no-AI affective recovery ( b = 0.279, p < 0.001). Relational reassurance-oriented responses were associated with greater perceived AI emotional necessity ( b = 1.010, p < 0.001) and stronger short-term AI-reliance intentions ( b = 0.628, p < 0.001). Bootstrap indirect-effect models were consistent with an adaptive pathway through self-efficacy [indirect effect = 0.279, 95% CI [0.171, 0.402]] and a reliance-related pathway through perceived AI necessity [indirect effect = 0.375, 95% CI [0.264, 0.501]]. Because the mediator-outcome associations were not randomized, these models do not establish causal mediation. Discussion The findings indicate an immediate-relief-subsequent-divergence pattern: similarly comforting responses can differ in whether they orient users toward independent regulation or continued access to AI. The reliance outcomes are short-term, state-like beliefs and intentions and should not be interpreted as emotional dependence.