Are we close enough? Impact of stress management chatbot’s self-disclosure and relationship types on user’s self-disclosure, trust, intimacy, and anonymity

Abstract

College students experience elevated stress levels, making AI-based chatbots promising tools for accessible stress management support. For effective stress management interventions, a user’s self-disclosure and trust toward chatbots are critical factors. While prior studies highlight the benefits of self-disclosure to chatbots, it remains unclear how these effects manifest in emotionally sensitive contexts or vary depending on the relationship type established with the chatbot.

The CASA (Computers Are Social Actors) paradigm explains that people apply social rules to computers, while Social Penetration Theory (SPT) emphasizes the reciprocity of self-disclosure in relationship formation. Therefore, this study investigates whether stress management chatbot’s self-disclosure and chatbot-user relationship types influence user’s self-disclosure, trust, anonymity, and intimacy in ways that parallel human-to-human interactions, as suggested by the CASA paradigm.

A 2 × 2 factorial design was employed with the chatbot’s self-disclosure presence and relationship type (mentor or peer) as independent variables. A total of 112 undergraduate students participated in chatbot-mediated stress management sessions, followed by evaluation surveys; after data screening, 100 participants were retained for analysis. Semi-structured interviews were conducted and analyzed with 24 participants across four chatbot types.

Results revealed that the chatbot’s self-disclosure had no significant effect on user outcomes. However, relationship type significantly influenced user anonymity and self-disclosure: mentor-relationship chatbots, mimicking more distant professional relationships, provided greater user anonymity, while peer relationship chatbots elicited greater user’s self-disclosure. These findings notably support the CASA paradigm. By integrating qualitative insights with quantitative findings, this study identifies design implications and key factors for developing effective stress management chatbots.