Constructing Trust and Distrust in AI Therapists: A Grounded Theory Study of Public Online Narratives

The global demand for mental health services is increasing, yet access to professional therapists remains limited. In this context, artificial intelligence (AI)-based solutions, especially therapeutic chatbots, are gaining attention as complementary tools. However, user adoption is hindered by trust issues. This study aimed to develop a grounded theory model to explore how trust in AI-driven therapeutic systems is built, maintained, and potentially lost. Trust is conceptualized as a multi-dimensional construct, incorporating system reliability (from human-computer interaction), healthcare credibility (including safety and ethics), and relational trust reflecting therapeutic alliance elements like empathy. A qualitative grounded theory approach was used, analyzing 557 public text units from social media, mental health forums, and related online platforms. Through open, axial, and selective coding, a mid-level conceptual framework was developed, resulting in the "Credential Reappraisal" model. This model outlines three stages: initial evidence appraisal, interactional testing based on empathy and accuracy, and trust stabilization or collapse. Social feedback and the system’s error management were found to be central to strengthening or undermining trust. The study reveals that trust in AI-based therapists is not static but dynamic, continuously renegotiated through interactions with the system and feedback from others. Transparency, accuracy, and error management are key factors in influencing trust. The study emphasizes the need for trust-by-design in developing AI-driven therapeutic systems, ensuring transparency, explainable responses, and active user feedback mechanisms for continuous improvement.