Human-guided AI therapy bot can improve mental health care

Dartmouth researchers have completed the first clinical trial of a fully generative AI therapy system, finding that their AI therapy bot or “Therabot” significantly reduced symptoms in people with depression, anxiety, and at risk of eating disorders. Users reported forming relationships with Therabot comparable to those with human therapists, but the study’s authors caution that such tools remain far from ready for unsupervised deployment.

Though there are many generative AI therapy bots on the market right now, this trial offers the first scientific evidence that AI therapy might be effective while also highlighting critical safety challenges that remain.

Real results from virtual therapy

The Dartmouth team enrolled 210 adults nationwide in its trial, all with clinically significant symptoms of major depressive disorder, generalized anxiety disorder, or high risk for eating disorders. Half of these participants received unlimited access to Therabot for four weeks, while a control group received no intervention.

Among Therabot users, depression symptoms dropped 51%, anxiety symptoms fell by 31%, and eating disorder risk decreased by 19%. These changes all significantly outpaced the control group.

“Our results are comparable to what we would see for people with access to gold-standard cognitive therapy with outpatient providers,” said Nicholas Jacobson, associate professor at Dartmouth’s Geisel School of Medicine and senior author of the study, in a statement. “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.”

Users averaged six hours of interaction over the trial period—equivalent to about eight therapy sessions, or one 45-minute session per week. Patients rated their connection with the AI therapy system using standard therapeutic alliance measures, and their scores matched levels typically achieved with human therapists.

“We did not expect that people would almost treat the software like a friend. It says to me that they were actually forming relationships with Therabot, ” Jacobson noted. “My sense is that people also felt comfortable talking to a bot because it won’t judge them.”

Affordable mental healthcare 24/7

Therabot has the bonus of being available at any hour and place and for however long the user would like—the AI tool operates through a smartphone app where users can respond to prompts about their mental state and initiate conversations as needed. The researchers were surprised to find that many users initiated conversations with Therabot on their own and believe it might point to a relationship between availability and patient engagement.

Currently, therapy phone apps don’t require insurance, the lack of which is one of the many barriers to care in the U.S. The scale of this access crisis is staggering: 47% of Americans live in designated mental health workforce shortage areas, with more than half of U.S. counties lacking even a single psychiatrist. Over 56% of adults with mental illness receive no treatment whatsoever, which translates to 27 million people going without needed care.

Those who do have access to therapy face another potential hurdle, as many may not be able to afford it. Nearly half (42%) of people point to high costs and inadequate insurance as the biggest barriers keeping them from getting help. Even those with insurance struggle, as only about 30% of commercially insured patients with mental health diagnoses successfully connect with specialists, and 28% of people receiving therapy must go out-of-network, incurring substantially higher costs than they would for other medical care.

For a system already failing to meet demand, AI solutions that bypass both provider shortages and insurance complications represent a potentially transformative approach.

Autonomous AI therapy chatbots are still a risky proposition

Despite the encouraging results of the Therabot trial, lead author Michael Heinz, assistant professor of psychiatry at Dartmouth, emphasizes that significant risks remain with AI therapy tools.

“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,” Heinz said. His warning extends to even Therabot, which has been in development since 2019 with continuous input from clinical psychologists and psychiatrists and was trained on a specialized dataset based on cognitive behavioral therapy and evidence-based psychotherapy techniques.

The trial highlighted the critical challenges facing this field. The research team maintained constant oversight, ready to intervene if the AI responded inappropriately or if users expressed acute safety concerns. Such safeguards aren’t built into most commercial applications currently on the market.

“This is one of those cases where diligent oversight is needed,” said Jacobson. “There are a lot of folks rushing into this space since the release of ChatGPT… but the safety and efficacy is not well established.”

AI must work alongside therapists

The researchers involved with the Therabot trial envision a future where AI and human therapists work in tandem, not in competition. “There is no replacement for in-person care,” Jacobson emphasized, “but we would like to see generative AI help provide mental health support to the huge number of people outside the in-person care system.”

As AI mental health applications proliferate in a regulatory environment still catching up to the technology, this first clinical evidence offers both promise and caution, suggesting these new tools have enormous potential to help those in need of care, but must be deployed under close clinical supervision to be truly effective.