JMIR Publications

ABSTRACT

Background:

Generative artificial intelligence (GenAI) is rapidly entering mental health care, supporting both client-facing tools (e.g., chatbots for support and self-management) and clinician-facing systems (e.g., documentation and assessment aids). Whether these tools ultimately help or harm psychotherapists and their clients depends not only on their technical performance but on how psychotherapists trust and distrust them in practice—that is, when they are willing to rely on GenAI, when they withhold reliance, and how they manage clients’ own GenAI use. Understanding how psychotherapists negotiate their trust and distrust is essential for future responsible and ethical integration of GenAI in mental health care, where GenAI’s promising benefits, such as reducing administration burden or enhancing client’ accessibility, must be balanced against risk that requires professional judgement rather than blanket adoption or rejection. Yet little empirical work has examined how practicing psychotherapists actively calibrate trust and distrust in GenAI across tasks and contexts, or how these judgments shape the evolving psychotherapist–client–GenAI relationship.

Objective:

This study aims to examine (1) what are psychotherapists' experiences with, perceptions of, and trust/distrust in GenAI in therapeutic contexts? and (2) how do psychotherapists perceive the role of GenAI within the therapeutic relationship, and how do their perceptions shape their trust and distrust in GenAI?

Methods:

Between January 2025 and May 2025, we conducted an interview study with 18 psychotherapists in the United States. Psychotherapists were recruited.

Results:

Our findings show that psychotherapists' adoption of GenAI was highly individualized and underpinned by “conditional'' trust—confidence that depended on maintaining professional control, aligning GenAI use with specific tasks, and considering who was using the GenAI tools. Trust was sustained when GenAI operated in clinician-supervised, supportive roles, but diminished when control shifted, tasks became high-stakes, or GenAI appeared to encroach on the therapeutic relationship (e.g., forming emotional bonds with clients or replacing core psychotherapist functions). Additionally, participants also voiced distrust towards the broader sociotechnical ecosystem, including developers, commercial incentives, and the absence of clear organizational guidelines.

Conclusions:

Psychotherapists’ perspectives offer critical insights into GenAI's current usages in their professional practices and the conditions under which they are willing to trust and distrust GenAI tools. Their experiences highlight the importance of maintaining clinician control, ensuring contextual appropriateness, and preserving the human connection central to psychotherapy. Future work should further examine how therapeutic orientation, professional experience, and client characteristics shape trust and distrust in GenAI. As GenAI becomes more embedded in mental-health care, research is also needed to explore how specific GenAI system features can be responsibly designed to support clinical workflows and enhance therapeutic relationships. Organizational and policy frameworks will be essential to ensure responsible, ethically aligned, and human-centered GenAI deployment in psychotherapy.