A generative artificial intelligence chatbot developed at Dartmouth College significantly reduced symptom severity in individuals experiencing depression, anxiety, and eating disorders during a four-week clinical trial, according to findings published March 27 in the New England Journal of Medicine AI.
The smartphone-based software tool, known as Therabot, was evaluated in a trial involving 106 participants across the United States who engaged in text-based conversations with the app. According to researchers at Dartmouth’s AI and Mental Health Lab, the intervention yielded measurable psychological improvements when compared against a control group of 104 individuals with the same conditions who did not use the software.
Clinical Outcomes Across Mental Health Conditions
Participants diagnosed with major depressive disorder who used Therabot reported an average 51% reduction in symptoms over the four-week study period, according to data published in the journal. Users with generalized anxiety disorder showed a 31% reduction in symptom severity. Among participants at risk for eating disorders, concerns specifically related to body image and weight declined by 19%.
The trial evaluated individuals diagnosed with major depressive disorder, generalized anxiety disorder, or eating disorders. Notably, nearly three-quarters of the participants in the Therabot group were not receiving medication or therapy at the time of the study, indicating the software functioned independently of other treatments. The application dynamically adjusted its questions and responses based on user input, incorporating cognitive behavioral therapy and other evidence-based clinical approaches.
User Engagement and Therapeutic Alliance
Throughout the four-week trial, users spent an average of six hours interacting with Therabot, an engagement volume researchers noted as roughly equivalent to eight therapy sessions. Increased user activity frequently occurred during late-night hours, a window frequently associated with psychological distress.
Researchers reported that the software maintained a strong therapeutic alliance with users, establishing a sense of trust and collaboration typically observed between patients and caregivers. In clinical settings, a robust therapeutic alliance is widely considered a critical metric linked to better treatment outcomes.
Safety Protocols and Future Development
Despite the positive outcomes, researchers emphasized that generative AI tools in mental health care require careful oversight. Therabot was programmed with automated detection systems designed to identify high-risk content, including expressions of suicidal ideation. When triggered, the software immediately directed users to seek professional help through emergency services or dedicated crisis lines, while ongoing conversations were monitored to verify adherence to therapeutic best practices.
Development of Therabot began in 2019 at Dartmouth’s AI and Mental Health Lab, with consultation from psychologists and psychiatrists to guide the software’s clinical responses and safety architecture.
Keep reading