A clinical trial conducted by Dartmouth researchers has shown the potential of a generative AI-powered therapy chatbot, known as Therabot, in improving the mental health of individuals with conditions like depression, anxiety, and eating disorders
The results, published on March 27 in NEJM AI, revealed significant improvements in the participants’ symptoms after using the chatbot for just eight weeks.
Therapy chatbot
The study involved 106 individuals across the United States, all diagnosed with major depressive disorder, generalised anxiety disorder, or an eating disorder.
Participants interacted with Therabot through a smartphone app, responding to prompts about their feelings or starting conversations when they needed to talk. Throughout the trial, participants experienced marked reductions in their symptoms.
Those who had depression showed a 51% average reduction in symptoms, leading to clear improvements in mood and overall well-being.
People with anxiety reported a 31% reduction, with many participants moving from moderate anxiety to mild anxiety, or even below the clinical threshold for a diagnosis. Participants at risk for eating disorders also showed a 19% reduction in concerns about body image and weight, showing a significant difference between the groups that didn’t have access to AI therapy.
AI supporting mental health
Although the results have been promising, these AI-powered therapies won’t be replacing traditional in-person mental health care.
However, Therabot’s ability to provide support on-demand is seen as a potential solution for those who lack regular access to mental health professionals. With an average ratio of 1,600 patients for every mental health provider in the United States, this technology could help fill a crucial gap in mental health care.
The AI chatbot, developed by Dartmouth’s AI and Mental Health Lab, interacts with users using natural, open-ended text responses, based on evidence-based best practices for psychotherapy and cognitive behavioural therapy.
For example, if a user with anxiety described feeling overwhelmed, Therabot might respond with a calming prompt like, “Let’s take a step back and ask why you feel that way.” If the system detects high-risk content, such as suicidal thoughts, it will immediately provide emergency contact information.
AI for mental health
Participants in the study had four weeks of unlimited access to Therabot, with evaluations conducted before and after the trial period. Nearly 75% of those who used Therabot did not undergo any other form of therapy or take medication during the study.
The participants interacted with the AI an average of six hours throughout the trial, equivalent to eight therapy sessions. Researchers tracked their progress using standardised questionnaires commonly used by clinicians to monitor mental health conditions.
The study also highlighted trust as an important aspect of the therapeutic process. Users reported forming a “therapeutic alliance” with Therabot, a term used to describe the trust and collaboration between a patient and therapist. Many participants initiated conversations with the chatbot, sometimes during heightened distress, such as in the middle of the night. The users treated Therabot almost like a friend, feeling comfortable opening up without fear of judgment.
Although positive findings offer encouraging results, AI therapy could still be risky. There is still a lot to be learned about the potential risks of using generative AI in mental health care, especially in high-risk scenarios. However, with safety protocols and clinician oversight, AI therapy could become invaluable for supporting mental health and offering real-time, accessible care.