On March 27, 2025, NEJM AI published the results of what Dartmouth described as the first-ever clinical trial of a generative-AI-powered therapy chatbot. The study tested Therabot, a text-based smartphone app that, unlike the scripted cognitive-behavioral-therapy bots that preceded it, generated its own conversational responses using a large language model. The work was directed by Nicholas Jacobson and Michael Heinz of Dartmouth’s Geisel School of Medicine, where Therabot had been in development in Jacobson’s AI and Mental Health Lab since 2019.
The randomized controlled trial enrolled 106 participants from across the United States, each diagnosed with major depressive disorder, generalized anxiety disorder, or at clinically high risk for a feeding or eating disorder. Over four weeks of using the app, participants saw an average 51 percent reduction in depression symptoms, a 31 percent reduction in generalized anxiety symptoms, and a 19 percent reduction in body-image and weight concerns. Dartmouth reported the improvements were comparable to what is seen in traditional outpatient therapy, and that participants said they could trust and communicate with the system to a degree comparable to working with a human professional.
The trial matters because it applied the gold standard of clinical evidence - a randomized controlled trial - to the newest and riskiest kind of mental-health chatbot, one that improvises rather than reading from a script. Earlier studies, like the 2017 Woebot trial, tested rule-based bots whose every response was vetted in advance; a generative system can say things its makers never wrote, which is both its promise and its hazard. The researchers stressed that Therabot was closely monitored by clinicians throughout and is not a substitute for professional care.
Why business readers should care: this is the moment generative AI in mental health crossed from demo to measured intervention, and it sets the evidentiary bar - run the trial, monitor the model, do not ship an unsupervised therapist - that regulators and buyers will now expect of anyone making clinical claims.