While most consumer AI chatbots used for mental health support were never designed with clinical rigor in mind, a project out of Dartmouth’s Geisel School of Medicine is taking the opposite approach. Therabot, a generative AI mental health chatbot developed by Dartmouth researchers, has been the subject of a published clinical trial, and coverage of the project resurfaced in August 2026 as public interest in AI therapy tools continues to grow.
Who Built Therabot and Why
Therabot was developed by a Dartmouth team that includes Michael Heinz, an assistant professor of psychiatry, and Nicholas Jacobson, an associate professor of biomedical data science and psychiatry at the Geisel School of Medicine. Their goal was to test whether a generative AI system, built and evaluated with the same scientific scrutiny applied to a drug or a therapy protocol, could safely and effectively support people dealing with common mental health conditions such as depression and anxiety.
A Different Model Than Consumer Chatbots
That distinction matters because the vast majority of AI tools people currently use for emotional support, from general-purpose assistants like ChatGPT to companion apps like Replika, were not built or tested as mental health interventions. They were adapted for that purpose by users themselves. Therabot instead grew out of a research program aimed at generating the kind of controlled, peer-reviewed evidence that clinicians and regulators typically expect before a tool is treated as a legitimate part of a care pathway. Jacobson, describing the broader motivation behind the project, said, “Digital technologies are one of the ways that folks have tried to enable better access to healthcare,” pointing to the persistent shortage of licensed therapists and long waitlists for care as the practical problem the tool is meant to address.
The Access Problem Driving the Research
The shortage Jacobson referenced is not abstract. Large swaths of the United States lack sufficient licensed mental health providers, and even where providers exist, cost and wait times keep many people from getting timely care. That gap has fueled demand for any tool, AI included, that can offer support between sessions or while someone waits for an appointment. Dartmouth’s research effort has positioned Therabot as an attempt to meet that demand responsibly rather than leaving the space entirely to unregulated consumer apps.
Skepticism Remains Even for Clinically-Tested Tools
Even a rigorously studied tool like Therabot operates in a field where skepticism runs deep. Mental health professionals broadly agree that AI chatbots, however well designed, should not be handed responsibilities like crisis intervention or clinical diagnosis. Reporting this year on AI chatbot harms, including an analysis of 185 real-world reports of mental health harms linked to chatbot use, has underscored that even well-intentioned tools can go wrong when a vulnerable user relies on them in place of professional care. Dartmouth’s own framing of Therabot as one tool among several approaches to closing the access gap, rather than a wholesale replacement for therapists, reflects that caution.
Where Therabot Fits in a Crowded Field
Therabot’s academic pedigree sets it apart from most of the AI mental health landscape in 2026, which researchers now describe as falling into roughly four camps: consumer chatbot startups such as Slingshot AI, enterprise and clinician-facing platforms such as Limbic, Lyra Health, and Spring Health, general-purpose AI labs including OpenAI, and university research groups like the Dartmouth team behind Therabot. Each camp is pursuing a different theory of how AI should intersect with mental health care, from fully automated support to tools meant to make human clinicians more efficient.
What’s Next for Clinically-Validated AI Therapy Tools
The continued attention on Therabot points to a broader question the mental health field is now grappling with: whether AI tools built and tested through traditional clinical research pipelines can scale fast enough to compete with the consumer apps that already have millions of users and no such evidence base. Dartmouth’s approach suggests one possible path forward — treating a chatbot less like a piece of consumer software and more like a medical intervention that must earn its place in care through published, peer-reviewed results. Whether that model can be replicated broadly, or whether regulators will eventually require it, is likely to shape how AI mental health tools are built and marketed in the years ahead.
