Can Computers Read Emotions in Therapy? New Research Explores AI Analysis of Therapy Sessions

When people go to therapy, emotions are at the heart of the process. Patients share their feelings, work through difficult experiences, and hopefully develop healthier ways of managing their emotional lives. But measuring and tracking those emotions throughout treatment has always been challenging for both researchers and therapists.

A recent study explored whether artificial intelligence could help by automatically analyzing the emotional tone of therapy conversations. Researchers used computer algorithms to examine transcripts from therapy sessions, looking for signs of positive and negative emotions in patients' words. They then checked whether these AI detected emotions matched what patients and therapists actually reported feeling during those same sessions.

Reading Between the Lines

The research team analyzed 85 therapy sessions involving 35 adult patients receiving cognitive behavioral therapy at a German university clinic. They used a sophisticated AI system designed to detect sentiment in text, which had been trained on millions of social media posts and other written content in multiple languages including German.

The AI examined only what patients said during sessions, not therapist statements, focusing on detecting overall positive or negative emotional tone. Meanwhile, both patients and therapists filled out questionnaires after each session rating how the patient felt during that particular meeting, using measures of emotions like sadness, anxiety, contentment, and relaxation.

Promising Connections

The results showed encouraging signs that the AI analysis could capture meaningful emotional information. When the computer detected more positive sentiment in a patient's speech, both patients and therapists were more likely to report positive emotions in that same session. Similarly, when the AI picked up negative sentiment, human ratings of negative emotions tended to be higher.

These connections held up even when researchers used advanced statistical methods to account for the fact that the same patients appeared across multiple sessions and were treated by different therapists. The correlations weren't perfect, but they were substantial enough to suggest the AI was detecting real emotional patterns.

The researchers also found that AI detected sentiments connected to important therapy processes. When patients showed more positive sentiment, they were more likely to report having good coping experiences and feeling like they were mastering challenges during therapy. Negative sentiment was associated with sessions where patients engaged more deeply with difficult emotional material.

Perhaps most importantly, changes in sentiment over time predicted therapy outcomes. Patients whose speech became more positive throughout treatment showed better results on measures of anxiety, depression, and overall psychological functioning when therapy ended. Those whose sentiment became more negative tended to have worse outcomes.

A New Tool in the Toolkit

This research suggests that sentiment analysis could become a valuable addition to existing ways of monitoring progress in therapy. Currently, therapists rely primarily on patient self reports, their own observations, and standardized questionnaires to track how someone is doing. Computer analysis of therapy conversations could provide another perspective, potentially catching emotional patterns that humans might miss.

The approach could be particularly useful because it provides an objective measure that doesn't rely on patients remembering how they felt or being willing to accurately report their emotions. Some people struggle to identify or communicate their emotional states, and others might feel pressured to present themselves in a certain way to their therapist.

If this technology continues to develop, it could help therapists in several ways. They might receive feedback about emotional patterns in their sessions, helping them notice when a patient is struggling even if that person doesn't explicitly say so. The analysis could also flag sessions with sudden increases in negative emotion, potentially alerting therapists to times when extra support might be needed.

For therapy training and supervision, sentiment analysis could help new therapists learn to recognize emotional dynamics in their sessions. Supervisors could review transcripts with trainees, using the AI analysis to discuss how emotional tone shifted throughout a session and what that might mean for treatment.

The technology could also advance therapy research by making it easier to study emotional processes across large numbers of sessions. Currently, analyzing emotions in therapy often requires trained human raters, which is time consuming and expensive. Automated analysis could help researchers study emotional patterns on a much larger scale.

The researchers acknowledge several important limitations to their work. The study was relatively small, involving only 35 patients, and was exploratory in nature. The AI system they used wasn't specifically designed for therapy conversations, so a system trained specifically on therapy data might perform better.

There were also some discrepancies between what the AI detected and what therapists reported about patient emotions. This could mean the AI was missing something important, or it might suggest that therapists sometimes have difficulty accurately reading their patients' emotional states. More research is needed to understand these differences.

The study also required therapy sessions to be transcribed, which is currently time consuming and expensive. For this technology to be practical in real therapy settings, more efficient transcription methods would need to be developed.

The Human Element Remains Central

It's important to note that this research doesn't suggest computers should replace human judgment in therapy. Rather, sentiment analysis might serve as an additional source of information to complement therapists' clinical expertise and patients' self reports.

The researchers emphasize that their AI system only captures basic emotional tone, not the complex nuances of human emotion and experience that skilled therapists are trained to recognize and respond to. The goal is to augment human understanding, not replace it.

Future research in this area might combine sentiment analysis with other automated approaches, such as analyzing facial expressions, voice tone, or physiological measures like heart rate. This multimodal approach could provide an even richer picture of emotional processes during therapy.

Researchers are also working on developing AI systems specifically trained on therapy conversations, which might be more accurate than general purpose sentiment analysis tools. As automatic transcription technology improves, it may become more feasible to implement these approaches in routine clinical practice.

The study represents an early step toward understanding how artificial intelligence might support emotional assessment in therapy. While the technology isn't ready for widespread clinical use, the research suggests it could eventually become a valuable tool for therapists, researchers, and patients working to understand and improve emotional well being.

As mental health care continues to evolve, tools like sentiment analysis might help make therapy more responsive to patients' moment to moment emotional experiences, potentially improving outcomes for people seeking help with psychological difficulties.

Eberhardt, S. T., Schaffrath, J., Moggia, D., Schwartz, B., Jaehde, M., Rubel, J. A., ... & Lutz, W. (2025). Decoding emotions: Exploring the validity of sentiment analysis in psychotherapy. Psychotherapy Research, 35(2), 174-189.