Too Caring to Be Real? Autonomy‐Supportive AI, Intimacy Gains, and Authenticity Loss

Abstract

Artificial intelligence-mediated communication increasingly incorporates autonomy-supportive and empathic language to foster emotionally meaningful consumer relationships, yet relational outcomes are not uniformly positive. This research introduces Perceived Autonomy Support in AI Communication (PAS-AI) to explain why empathically framed AI simultaneously enhances perceived intimacy and, beyond a threshold, undermines perceived authenticity. Across a staggered field rollout, a longitudinal user-level panel, and a between-subjects experiment (N = 207), we estimate causal and dynamic effects of PAS-AI using event-study Difference-in-Differences, System GMM, and experimental manipulation. Results show that moderate PAS-AI reliably increases relational intimacy, reflected in greater conversational elaboration and voluntary re-engagement. Yet when autonomy-supportive cues become overly consistent or fluent, users report authenticity decline, describing AI responses as scripted or “too caring to be real.” Human-in-the-loop escalation and transparent AI disclosure attenuate this decline, indicating that relational trust can be preserved through calibrated design. The findings reframe empathic AI as a matter of relational calibration rather than imitation. We distinguish throughout between PAS-AIattr (message-level linguistic features measured via NLP classifier) and PAS-AIperceived (the user-side psychological experience of autonomy support).