Dartmouth Health

Mental health providers are severely outnumbered, with roughly 1,600 patients per available provider for depression and anxiety in the United States. Millions of people lack regular or timely access to mental health professionals, and traditional outpatient therapy cannot scale to meet demand. Generative AI chatbots had not yet been validated in a rigorous clinical trial for safety or efficacy.

Dartmouth researchers developed Therabot, a generative AI-powered therapy chatbot trained on evidence-based psychotherapy and cognitive behavioral therapy best practices. Participants accessed the app via smartphone, engaging in open-ended text conversations initiated by Therabot or by the user. The system personalized its responses based on prior conversations and escalated high-risk content (e.g., suicidal ideation) to crisis resources automatically.

After eight weeks, participants with depression experienced a 51% average reduction in symptoms, those with generalized anxiety saw a 31% reduction, and those at risk for eating disorders showed a 19% reduction in body image concerns—significantly outpacing the control group. Participants reported therapeutic alliance (trust and collaboration) comparable to in-person mental health providers, and engaged with Therabot for an average of six hours, equivalent to roughly eight therapy sessions.

• Generative AI therapy can produce clinically meaningful outcomes comparable to gold-standard outpatient cognitive behavioral therapy, but requires continuous clinician oversight to manage safety risks.
• Users formed genuine emotional bonds with the chatbot, frequently initiating conversations outside of prompted check-ins—including late at night—suggesting AI can extend mental health support beyond office hours.
• Rigorous benchmarking (90%+ responses consistent with therapeutic best practices) was a prerequisite before proceeding to a clinical trial; rushing to deploy without this validation creates serious safety risks.