Impact of confidentiality assurances and self-disclosure on user satisfaction with AI-assisted emotional support

Purpose While AI conversational agents such as Woebot offer a promising avenue for improving psychological well-being through empathetic conversations with nonclinical populations, growing privacy breach risks of AI are a significant threat to their use for emotional support. Therefore, this study aims to examine how confidentiality assurances from AI agents affect user satisfaction with AI-assisted emotional support, exploring the serial-mediation effects of perceived confidentiality and self-disclosure and the moderation effect of privacy concerns. Design/methodology/approach In a controlled experiment, 350 US adults interacted with a standardized AI emotional support agent in a three-dimensional metaverse and were randomly assigned to either the confidentiality-assurance condition or the control condition. Findings Results showed that confidentiality assurances significantly increased user satisfaction with AI-assisted emotional support. Moreover, we observed the mediation effects of perceived confidentiality and the serial-mediation effects of both perceived confidentiality and willingness to self-disclose. Originality/value This study contributes to the privacy literature by exploring how brief, standardized confidentiality assurances function in an AI-driven emotional support context. Instead of focusing on complex, personalized privacy designs, these findings offer a nuanced perspective on human-AI interactions that a simple structural cue can serve as a sufficient heuristic anchor to strengthen users' willingness to self-disclose across varying user traits.