Dartmouth College

Mental health treatment faces significant challenges with scalability, user engagement, and retention among digital therapeutics. There is a growing need for personalized, effective interventions that can reach patients at scale, particularly those with clinically significant symptoms of depression, anxiety, and eating disorders.

Researchers at Dartmouth College developed and tested Therabot, an expert-fine-tuned generative AI chatbot for mental health treatment. A national randomized controlled trial enrolled 210 adults with clinically significant symptoms of major depressive disorder (MDD), generalized anxiety disorder (GAD), or at clinically high risk for feeding and eating disorders (CHR-FED), randomizing them to a 4-week Therabot intervention or waitlist control.

Therabot users showed significantly greater reductions in MDD symptoms (Cohen's d=0.845–0.903), GAD symptoms (d=0.794–0.840), and CHR-FED symptoms (d=0.627–0.819) compared to controls at both 4-week and 8-week assessments. Average usage exceeded 6 hours, and participants rated Therabot's therapeutic alliance as comparable to human therapists. This was the first RCT demonstrating effectiveness of a fully generative AI therapy chatbot for clinical-level mental health symptoms.

• Fine-tuned generative AI chatbots can deliver clinically meaningful symptom reductions for depression, anxiety, and eating disorder risk comparable to effect sizes seen in human therapy
• High user engagement (>6 hours average use) and positive therapeutic alliance ratings suggest Gen-AI chatbots can overcome common retention challenges in digital mental health tools
• Scalable AI-based interventions show promise for expanding access to mental health treatment, though larger clinical samples are needed to confirm generalizability