🛰️ Daily AI Frontier
‹ back to 2026-08-23

Move by Move: Measuring and Steering How LLMs Conduct Psychotherapy

Research Medical/Healthcare AI

Ranking

Overall 80
Content 95
Popularity 45

Observed public metrics from 1 member.

Merged summary

TL;DR - This paper introduces a psychologist-validated ontology of ten therapeutic moves to measure how LLMs conduct psychotherapy. Frontier models differ substantially from human clinicians, but tool-based guidance improves their alignment without fine-tuning.

  • Models use inquiry at up to three times the human rate while underusing psychoeducation.
  • LLMs tend to continue strategies introduced by human clinicians rather than initiate those strategies independently.
  • A judge-based annotation approach scales the ontology while matching expert agreement.
  • Exposing therapeutic moves as tools roughly halves deviation from human move distributions and improves turn-level alignment by 7–9 percentage points.

Sources (1)

Move by Move: Measuring and Steering How LLMs Conduct Psychotherapy

arXiv cs.CL Afonso Baldo, Hugo Pitorro, Areti Vassilopoulos, Anabela C. Areias, Maya D'Eon, FabĂ­ola Costa, Ricardo Rei, Nuno M. Guerreiro 2026-08-21 arXiv:2608.21325
Public signals Hugging Face upvotes 0
Providers: Hugging Face · Upvotes 0 OpenAlex · N/A Publisher · N/A Semantic Scholar · N/A X · N/A Fetched 2026-09-22 14:33:06.999161 UTC

TL;DR - This paper introduces a psychologist-validated ontology of ten therapeutic moves to measure how LLMs conduct psychotherapy. Frontier models differ substantially from human clinicians, but tool-based guidance improves their alignment without fine-tuning.

  • Models use inquiry at up to three times the human rate while underusing psychoeducation.
  • LLMs tend to continue strategies introduced by human clinicians rather than initiate those strategies independently.
  • A judge-based annotation approach scales the ontology while matching expert agreement.
  • Exposing therapeutic moves as tools roughly halves deviation from human move distributions and improves turn-level alignment by 7–9 percentage points.
item →