Can AI Conduct Psychodynamic Therapy?

Consumers use large language models (LLMs) to provide low-cost psychotherapy. While researchers have identified the dangers of using LLMs for psychotherapy by testing their safety features (e.g., Santos et al., 2025), no one knows the quality of psychotherapy process conducted by LLMs. This session will report on the results of a study seeks to address this gap by assessing typical psychotherapy process conducted by three LLMs (ChatGPT, Claude, Copilot), which implemented three treatment models (psychodynamic therapy, cognitive-behavioral therapy, and supportive-expressive therapy) in 54 two-session modules.