Move by Move: Measuring and Steering How LLMs Conduct Psychotherapy
Researchers have developed an ontology of ten therapeutic moves that large language models (LLMs) use in psychotherapy interactions. They analyzed human clinician sessions and LLM-led sessions to identify how the models differ from humans. The study found that LLMs overuse inquiry, neglect psychoeducation, and are strongly context-anchored. However, by exposing the ontology as a set of tools, the researchers were able to improve turn-level alignment with human therapists by 7
Researchers have developed an ontology of ten therapeutic moves that large language models (LLMs) use in psychotherapy interactions. They analyzed human clinician sessions and LLM-led sessions to identify how the models differ from humans. The study found that LLMs overuse inquiry, neglect psychoeducation, and are strongly context-anchored. However, by exposing the ontology as a set of tools, the researchers were able to improve turn-level alignment with human therapists by 7-9 percentage points without fine-tuning.
---
Why it matters: This research matters because it sheds light on how LLMs conduct psychotherapy interactions and identifies areas for improvement. Understanding these differences is crucial for developing more effective and empathetic AI-powered therapy tools.
Source: https://arxiv.org/abs/2608.21325
This article was originally published at: https://arxiv.org/abs/2608.21325