A study discussed in connection with the Nature Medicine publication suggests that properly designed AI systems can reach a very high standard in tasks related to cognitive behavioral therapy (CBT). In the experiment, researchers analyzed responses generated by an AI model and compared them with responses written by therapists for a set of clinical scenarios.
The results showed that 74.3% of AI-generated sessions received higher scores than the top 10% of therapy sessions prepared by human therapists. Evaluations were conducted using the Cognitive Therapy Rating Scale (CTRS) – a standard framework used to assess the quality of CBT sessions.
The study also examined the impact of adding a dedicated layer of so-called clinical reasoning to a large language model. Systems equipped with this layer scored on average 43% higher on the CTRS scale than language models operating without such specialization.
Clinical experts also compared the quality of responses generated by different AI systems. In those comparisons, specialized therapeutic agents were preferred by clinicians in 82.7% of cases over standard language models – particularly in areas such as therapy structure, clinical justification and the ability to avoid potential harm to patients.
The authors stress, however, that the experiment did not involve real therapy sessions with patients. Instead, researchers analyzed clinical scenarios and therapeutic responses generated under controlled conditions. The goal was to determine whether AI could correctly apply CBT techniques and whether the responses met clinical standards.
Cognitive behavioral therapy is one of the most extensively studied forms of psychotherapy and is commonly used to treat conditions such as depression, anxiety disorders and stress-related problems. The authors note that AI could eventually serve as a supporting tool for therapists – for example by assisting with case analysis, preparing therapy materials or guiding exercises between therapy sessions.
At the same time, the researchers emphasize that current systems are not designed to replace human therapists. In clinical practice, the therapeutic relationship, the patient’s personal context and the responsibility of the treating professional remain essential.