A conceptual model for trust in AI-based psychotherapy assistants

The growing use of AI in mental health care highlights the urgent need to better understand and design trustworthy AI systems, especially in psychotherapy where trust plays a central role. Despite increasing attention to the ethics of AI, little research has systematically examined trust in AI-based Psychotherapy Assistants (AIPAs) from both ethical and psychological perspectives. This paper focuses on AIPAs—including chatbots, robots, and apps—that simulate therapeutic interactions. Our main goal is to explore how AIPAs can be designed to foster trustworthiness. We approach this by integrating existing literature on trust, trustworthy AI, and trust in psychotherapy. To move beyond abstract notions, we aim to translate the concept of trustworthiness into practical design language that developers and designers can apply. We conducted a semi-systematic literature review to identify key factors influencing trust in AIPAs. These were categorized into four domains: trustor-related, trustee-related, interaction, and environmental factors. Based on these insights, we developed a conceptual model of trust in AI-assisted psychotherapy. To validate and refine this model, we conducted a questionnaire study involving interdisciplinary experts in AI ethics, HCI, psychology, and related fields. Finally, we apply our model in a hypothetical case study to illustrate its practical value and provide concrete design recommendations. Our work contributes to a deeper understanding of trust in AI for mental health and offers a foundation for more trustworthy and human-centered AIPA development.