Recently, researchers at Dartmouth College, a prestigious Ivy League university in the U.S., conducted the first clinical trial focused on using generative AI for mental health counseling. The results were published in the leading journal NEJM AI. The study demonstrated that an AI tool called Therabot significantly improved mental health symptoms among participants.
The trial included 106 participants from across the United States who had been diagnosed with major depressive disorder, generalized anxiety disorder, or an eating disorder. Through a mobile app, participants interacted with Therabot, and many reported that their sense of trust and communication with the system was comparable to working with a licensed mental health professional.
As part of the randomized clinical trial, participants assigned to the Therabot group were given unlimited access to the system for four weeks. A separate control group of 104 individuals with the same diagnoses did not use Therabot. Nearly 75% of the Therabot users were not receiving medication or any other treatment during the trial period.
Therabot asked users about their well-being and personalized its questions and responses based on their conversations. Participants interacted with Therabot for an average of six hours over the four-week period — the equivalent of about eight therapy sessions. A standardized assessment was administered at the four-week mark, followed by a second evaluation four weeks later.
The results were striking: all participants in the Therabot group showed clinically significant improvement, exceeding researchers’ expectations. Study lead Dr. Jacobson noted, “We didn’t expect people to almost treat the software like a friend. It suggests they actually formed a relationship with Therabot.”
Key findings after eight weeks of use included:
- Depression: Participants diagnosed with depression experienced an average 51% reduction in symptoms, with clinically significant improvements in mood and overall well-being.
- Generalized Anxiety Disorder: Anxiety symptoms dropped by an average of 31%, with many individuals moving from moderate to mild levels, or from mild to below the clinical threshold.
- Eating Disorders: Among high-risk individuals — traditionally a challenging group to treat — concerns around body image and weight decreased by 19% on average, significantly outperforming the control group.
Professor Jacobson explained, “In-person therapy is still irreplaceable, but the demand for mental health services far exceeds the available supply. Our hope is that generative AI can offer mental health support to the many people who are unable to access traditional therapy.”
He emphasized that in the United States, each licensed mental health professional is, on average, responsible for over 1,600 individuals suffering from depression or anxiety. While the use of AI in mental health care still requires careful monitoring and oversight, Jacobson believes it has the potential to deliver immediate support to a large population who lack access to conventional or timely services.
Therabot is not limited by geography and is available anytime, anywhere. It offers users real-time support in coping with everyday challenges and provides helpful prompts or reflections in the moment.
Interestingly, the study found that users developed a surprisingly strong therapeutic alliance with Therabot — a level of trust and collaboration comparable to that seen between clients and human therapists.
“We didn’t expect people to relate to the software almost like a friend,” Professor Jacobson said. “It suggests that users genuinely formed a connection with Therabot.”
(This article is adapted from a Dartmouth College press release. The original study was published in NEJM AI*.*)
Commentary:
Perhaps the most unexpected and significant finding in this research is the development of a therapeutic alliance. In traditional therapy, this bond — the relationship of trust and collaboration between therapist and client — is widely recognized as one of the strongest predictors of positive outcomes.
Until now, many have assumed that such a relationship could only be formed between humans, given its deeply emotional and empathetic nature. However, this study presents a compelling case that AI may have the potential to foster a similar type of connection — one that, while different, still holds therapeutic value.
Since the release of ChatGPT and other generative AI tools, many individuals have experimented with using AI for emotional support, but until now, there has been little clinical evidence to validate its impact. This study represents a breakthrough in our understanding of AI’s role in mental health care and could open the door to more research exploring its full potential.
Curious about AI for emotional support?
Wondering how AI can help you process your emotions and find calm?
If you're not quite ready to step into a therapist's office — or if you just need someone to talk to right now — try Sort AI, an AI-powered mental wellness app developed by licensed therapists and AI engineers.
Trained on extensive psychological research, Sort AI uses proven therapeutic techniques to guide conversations and help you release emotional tension.
More than 85% of users report a positive shift in their emotional well-being after using Sort AI.
Enjoyed this article?Feel free to share it and join the conversation about the future of AI and mental health!
#AITherapist #MentalHealth #TechForGood
