Researchers and technologists are exploring whether generative artificial intelligence and large language models can predict the success or failure of psychotherapy early in treatment, potentially allowing therapists and clients to adjust course before investing significant time in ineffective approaches.
Early Prediction Timelines
A research study entitled “Predicting Treatment Success And Failure Using Routine Outcome Data: The Role Of Therapist Effects In Dynamic Predictive Modelling” by Daryl Mahon, Takuya Minami, and Jeb Brown, published in Counseling and Psychotherapy Research on September 24, 2024, found that therapy outcomes can be predicted as early as the second session with a good degree of accuracy of 65%. The study analyzed data from 1,020 therapists and 68,690 clients and found that “predictions made as early as the second session remain valid even for clients taking up to 10 sessions to complete treatment.” Various traditional statistical approaches have previously been used for this prediction task, and the latest avenue involves using generative AI and large language models.
The AI Approach
One method involves creating “digital twins”—detailed profiles of both the therapist and the client that allow AI to simulate the therapeutic relationship and predict outcomes. Four primary variations exist: using digital twins of both a real therapist and real client; mimicking a real therapist with a synthesized AI client; using an AI persona therapist with a real client’s digital twin; or synthesizing both therapist and client as AI personas.
Using an LLM, a therapist’s profile and a client’s circumstances can be inputted, and the AI can make predictions about treatment success based on early session descriptions. For example, when given details about a client with mild symptoms, openness to therapy, and no acute risk factors, one LLM predicted successful resolution within six weeks. When presented with a resistant client with unresolved trauma, the same AI predicted limited improvement even after 15 sessions and recommended specialized trauma-focused therapy.
Challenges and Concerns
Significant issues arise with relying solely on LLM predictions for clinical decision-making. Privacy and confidentiality concerns emerge when using AI in this manner, along with potential clinical liability and regulatory difficulties. There are also worries about self-fulfilling prophecies—if therapists and clients believe a prediction about treatment duration, it could unconsciously shape the actual therapeutic process.
This article is an AI-assisted summary. All facts and figures are drawn from the original report: https://www.forbes.com/sites/lanceeliot/2026/04/29/using-generative-ai-to-predict-mental-health-treatment-success-and-psychotherapeutic-trajectories/