First AI Therapy Chatbot Proven in Clinical Trial Shows Big Results

A Dartmouth-developed AI chatbot, Therabot, has shown significant success in reducing symptoms of anxiety and depression in a new clinical trial, rivaling outcomes of gold-standard therapy with a human provider.

The chatbot, designed to simulate open-ended conversations using cognitive behavioral therapy (CBT) and other evidence-based mental health techniques, was created by researchers at Dartmouth, including associate professor Nicholas Jacobson and research psychiatrist Dr. Michael Heinz. Their goal: to offer an accessible, always-available tool for people experiencing mental health challenges.

How Therabot Works

Therabot engages users in natural, supportive text conversations, using a custom training set built around modern psychotherapeutic practices of for psychotherapy and cognitive behavioral therapy. For example, when a user mentions feeling anxious or overwhelmed, Therabot might respond with, “Let’s take a step back and ask why you feel that way.”

The app also includes safeguards. If Therabot detects high-risk language, such as suicidal ideation, it immediately prompts users to contact emergency services or a crisis hotline via an on-screen button.

Therabot’s Effectiveness in Treating Key Conditions

Participants used the Therabot smartphone app to type responses to prompts about how they were feeling or to start conversations when they needed support.

Results from the clinical trial showed significant symptom reduction across multiple conditions:

• Depression: Participants diagnosed with major depressive disorder experienced an average 51% reduction in symptoms, leading to clinically significant improvements in mood and overall well-being.

• Generalized Anxiety: Users reported a 31% average reduction in symptoms, with many shifting from moderate to mild anxiety, or from mild anxiety to levels below the clinical threshold for diagnosis.

• Eating Disorders: Among participants at risk for eating disorders—a population traditionally more difficult to treat—Therabot users saw a 19% average reduction in concerns about body image and weight, a result that significantly outpaced outcomes in the control group.

The researchers conclude that, while AI-powered therapy still requires careful oversight from clinicians, it holds strong potential to deliver real-time support to individuals who lack consistent or immediate access to mental health care.

“There is no replacement for in-person care, but there are nowhere near enough providers to go around,” Jacobson says. In the U.S., each available provider serves an average of 1,600 patients with depression or anxiety alone, highlighting the overwhelming demand for mental health care, the researchers note.

Michael Heinz, the study’s lead author and an assistant professor of psychiatry at the Center for Technology and Behavioral Health and the Geisel School of Medicine, says the trial results highlight the important work still needed to ensure generative AI can be used safely and effectively in mental health treatment.

“While these results are very promising, no generative AI agent is ready to operate fully autonomously in mental health where there is a very wide range of high-risk scenarios it might encounter,” says Heinz

Inside the Clinical Trial

The trial included 106 participants from across the U.S. diagnosed with major depressive disorder, generalized anxiety disorder, or an eating disorder. A separate control group of 104 individuals with the same diagnoses was monitored but did not have access to Therabot.

Participants in the Therabot group had unlimited access to the app over four weeks, with conversations tailored to individual concerns. Notably, about 75% of this group were not receiving any other therapeutic or pharmaceutical treatment during the study.

The app engaged users in personalized conversations, adapting its questions and responses based on each individual’s input. Researchers reviewed these interactions to confirm that Therabot’s replies consistently aligned with established therapeutic standards.

After four weeks, participants were evaluated using standardized clinical questionnaires.

During the second four-week phase of the study, users could continue initiating conversations with Therabot, but the app no longer sent prompts.

By the end of the full eight-week period, all participants who used Therabot showed statistically and clinically significant reductions in symptoms.

Real-World Impact

Jacobson emphasized that these improvements weren’t just numerical—they reflected real, daily-life benefits. On average, users spent six hours engaging with Therabot, equivalent to eight traditional therapy sessions.

“Our results are comparable to what we would see for people with access to gold-standard cognitive therapy with outpatient providers,” Jacobson said. “We’re talking about potentially giving people the equivalent of the best treatment you can get in the care system over shorter periods of time.”

Perhaps most notably, users reported a strong therapeutic alliance with Therabot—a psychological term denoting trust and collaboration between a patient and provider. Jacobson observed that people often initiated conversations themselves, sometimes in the middle of the night, and many responded to the app as if it were a friend.

“We did not expect that people would almost treat the software like a friend,” he said. “My sense is that people also felt comfortable talking to a bot because it won’t judge them.”

“Therabot is not limited to an office and can go anywhere a patient goes. It was available around the clock for challenges that arose in daily life and could walk users through strategies to handle them in real time,” Heinz says.

Safety, Oversight, and Ethical Considerations

The study emphasized that while AI tools like Therabot offer benefits, their development must be accompanied by strict safety standards. Dr. Heinz noted that the team was prepared to intervene in real time if any user showed signs of acute distress.

“The feature that allows AI to be so effective is also what confers its risk—patients can say anything to it, and it can say anything back,” he said.

“This trial brought into focus that the study team has to be equipped to intervene—possibly right away—if a patient expresses an acute safety concern such as suicidal ideation, or if the software responds in a way that is not in line with best practices,” Heinz says. “Thankfully, we did not see this often with Therabot, but that is always a risk with generative AI, and our study team was ready.”

Earlier evaluations of Therabot over the past two years showed that more than 90% of its responses adhered to best therapeutic practices, giving the team confidence in its clinical trial readiness. Still, Jacobson cautioned against the flood of AI tools emerging after ChatGPT's release, warning that flashy demos may lack the rigorous oversight and safety needed in real mental health settings.

“This is one of those cases where diligent oversight is needed, and providing that really sets us apart in this space,” Jacobson said.

What This Means

Therabot’s success in this clinical trial marks a significant milestone in the development of AI-powered mental health tools. It demonstrates that generative AI, when grounded in evidence-based practices and carefully monitored by clinicians, can meaningfully reduce symptoms of depression, anxiety, and eating disorders—conditions that affect millions and often go untreated due to systemic barriers.

With each U.S. provider facing an average of 1,600 patients seeking care for anxiety or depression, the mental health system is under immense strain. Tools like Therabot could help bridge that gap, offering timely, personalized support to people who might otherwise wait weeks or months for in-person therapy—or never receive it at all.

Importantly, the study also highlights the need for rigorous clinical oversight, ethical guardrails, and long-term evaluation. AI’s strength lies in its scalability and 24/7 availability, but those same qualities introduce risks if left unchecked. The trial’s success hinged on continuous review of Therabot’s responses and readiness to intervene when safety concerns arose.

As digital therapeutics evolve, Therabot offers a model for how human-led AI can extend mental health care in responsible and effective ways. This study moves the conversation beyond hype and into the realm of clinically validated impact, setting a new standard for future AI tools in health care.

Read the full clinical trial summary from Dartmouth’s Center for Technology and Behavioral Health here.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.