In short
Therabot is an experimental AI therapy chatbot built by researchers at Dartmouth College. In 2025 the team published the first randomized controlled trial of a generative-AI therapy chatbot in the journal NEJM AI. Across 106 US adults, Therabot users saw average symptom reductions of 51% for depression, 31% for anxiety, and 19% for eating-disorder concerns versus a control group. It matters because it is the first rigorous trial evidence for this kind of tool, but Therabot is a supervised research prototype, not a product. It is not available to the public, and its results come from controlled study conditions rather than everyday use.
What Therabot is
Therabot is a generative-AI therapy chatbot developed by researchers at Dartmouth College. Unlike many wellness apps that wrap a general chatbot in friendly branding, Therabot was purpose-built and fine-tuned by a clinical research team to deliver therapy-style conversation grounded in evidence-based techniques, and it was created specifically to be tested in a formal trial.
Therabot exists only inside Dartmouth's supervised trial. Its significance comes from how it was evaluated rather than from any commercial release. When people talk about the Dartmouth AI therapy chatbot, they are almost always referring to Therabot and the trial built around it.
If you are in crisis or thinking about suicide, do not rely on any app or chatbot. In the US, call or text 988 to reach the Suicide and Crisis Lifeline, available 24 hours a day. Tools like Therabot are not designed for emergencies and were never meant to serve as a crisis service.
Therabot is the one I point to when people say AI therapy is all hype. A real trial, real supervision, real results. It also proves the point: the version that works is the one built and watched by clinicians.
The first randomized controlled trial of a gen-AI therapy chatbot
In 2025, the Dartmouth team published results from a randomized controlled trial of Therabot in NEJM AI, a journal from the publishers of the New England Journal of Medicine. A randomized controlled trial, or RCT, randomly assigns participants to either the treatment being tested or a comparison group, which makes it the most rigorous common way to judge whether an intervention actually causes an effect.
This study is widely described as the first RCT of a generative-AI therapy chatbot. That distinction matters. Chatbots have played therapist since ELIZA in the 1960s, and earlier mental-health tools such as Woebot and Wysa were tested in trials, but they delivered more scripted, rule-based content. Therabot uses generative AI to hold a more open, free-flowing conversation, and the Dartmouth trial was the first to put that newer style of tool through a randomized test.
Participants were 106 US adults with clinically significant symptoms of major depression, generalized anxiety, or elevated risk for an eating disorder. Over the course of the study, people who used Therabot were compared against a control group that did not, allowing the researchers to estimate how much of any improvement could be attributed to the chatbot itself.
What the study found
The headline finding, reported by Heinz et al. in NEJM AI and in Dartmouth's announcement, was that participants who used Therabot saw an average 51% reduction in depression symptoms, a 31% reduction in anxiety symptoms, and a 19% reduction in eating-disorder concerns compared with the control group. The benefits spanned all three areas the study focused on, and in each case the direction was the same, with the Therabot group improving more than the comparison group.
The researchers also reported high levels of engagement, with participants using the tool meaningfully across the study period, and described a sense of connection between users and the chatbot that was notable for a piece of software. Engagement matters because many digital mental-health tools fail not because the content is wrong but because people stop using them within days.
Participants also rated their sense of alliance with Therabot, the trust and collaboration usually measured between patients and clinicians, at levels comparable to what people report in outpatient psychotherapy, a striking result for a piece of software. The honest takeaway is that the results were positive and statistically meaningful, not that they prove a chatbot can match a human therapist.
Why Therabot matters
Therabot matters because it supplies something the AI therapy field has mostly lacked: rigorous, randomized evidence for a generative-AI tool. For years, the conversation about AI therapy has run ahead of the data, with many consumer apps making confident claims while most AI therapy studies to date have been small, short, or uncontrolled. A well-designed RCT published in a respected medical journal is a meaningful step up in the quality of evidence.
The result also hints at a possible role for these tools in a stretched mental-health system. There are not enough clinicians to meet demand, waitlists are long, and cost is a barrier for many people. A supervised, evidence-based chatbot could, in principle, extend support to people who would otherwise get nothing while they wait, although that promise still has to be tested at scale.
Finally, Therabot raises the bar for the whole category and shows what best practices for AI chatbots in therapy look like in action. It proves it is possible to study a generative-AI therapy tool properly, which makes it harder for other products to lean on vague wellness language without evidence. The credible future of AI therapy looks more like Therabot's careful trial design and less like marketing.
The important caveats
The biggest caveat is that Therabot is not publicly available. You cannot download it, and the results do not describe what happens when an unsupervised person uses an off-the-shelf chatbot at home. The trial ran under controlled conditions, with a defined protocol and clinical oversight in the background, which is very different from the open app market.
The evidence is also early. A single trial, however well conducted, is only a starting point. Important questions remain about how long the benefits last, how the tool performs across more diverse populations, how it handles signs of crisis or risk, and whether the effects hold up when the study team is not monitoring use. Replication by independent groups will matter a great deal.
It is also a mistake to read Therabot's results as proof that AI can replace therapists. The study tested a specific, purpose-built, supervised tool against a control group for particular symptoms. It did not show that generative AI is equivalent to a licensed clinician, that it is safe for severe or complex conditions, or that any random chatbot will produce the same outcome. The careful conclusion is the right one: promising evidence for a specific research tool, with real limits.
Can you use Therabot, and what to do instead
For now, you cannot use Therabot as a member of the public, and there is no consumer version to sign up for. Any site claiming to offer the Dartmouth Therabot to ordinary users should be treated with caution, since that is not how the tool currently exists.
If you are working out how to use AI as a therapist substitute today, look instead at consumer apps that name a clear clinical approach such as CBT, point you to crisis resources, and are transparent about privacy, while remembering that most have far less evidence than a controlled trial provides. Used with realistic expectations, those tools can help with mild symptoms and skill-building between or before formal care.
If your symptoms are significant, persistent, or worsening, the better step is a human professional. No AI tool, including Therabot, diagnoses, treats, or cures mental illness, and none replaces a licensed clinician or a crisis line. In the US, call or text 988 if you are in danger or thinking about suicide. If you would rather work with a person, browse licensed therapists in our directory.
Key takeaways
- Therabot is a generative-AI therapy chatbot built by researchers at Dartmouth College, created specifically to be tested in a formal trial.
- In 2025 the team published the first randomized controlled trial of a generative-AI therapy chatbot in NEJM AI.
- Across 106 US adults, the study reported average symptom reductions of 51% for depression, 31% for anxiety, and 19% for eating-disorder concerns versus a control group, with high engagement.
- It matters because it provides rare, rigorous evidence for AI therapy, raising the bar for a field long short on controlled trials.
- Therabot is a supervised research prototype, not a public app, so the results come from controlled conditions and cannot be reproduced by downloading it.
- The evidence is early and limited, and it does not show that AI can replace a licensed clinician. In the US, call or text 988 in a crisis.
Want care available now?
Browse licensed therapists in our directory.
Frequently asked questions
What is Therabot AI therapy?
Therabot is an experimental AI therapy chatbot built by researchers at Dartmouth College. It uses generative AI to hold therapy-style conversations grounded in evidence-based techniques, and it was created to be tested in a formal clinical trial rather than sold as a consumer app. In 2025 the Dartmouth team published the first randomized controlled trial of a generative-AI therapy chatbot using Therabot.
What is the Dartmouth AI therapy study?
It is a randomized controlled trial of Therabot, a generative-AI therapy chatbot built at Dartmouth, published in the journal NEJM AI in 2025. It enrolled 106 US adults with significant symptoms of depression, anxiety, or eating-disorder risk, and those who used Therabot showed larger symptom reductions than a control group. It is widely described as the first RCT of a generative-AI therapy chatbot.
What were the Therabot study results?
Participants who used Therabot saw average symptom reductions of 51% for depression, 31% for anxiety, and 19% for eating-disorder concerns compared with a control group, per the Dartmouth team's NEJM AI paper and press release. Engagement was high, and participants rated their alliance with the tool at levels comparable to outpatient therapy. The results came from controlled study conditions rather than everyday use.
Is Therabot available to the public?
No. Therabot is a research prototype tied to Dartmouth's clinical trial, not a released consumer product. You cannot download it or sign up for it, and there is no public version. Any site claiming to offer the Dartmouth Therabot to ordinary users should be treated with caution, because that is not how the tool currently exists.
Does Therabot prove AI can replace therapists?
No. The trial tested a specific, purpose-built, supervised tool against a control group for particular symptoms. It provides promising early evidence that a carefully designed AI chatbot can help with some symptoms, but it does not show that generative AI equals a licensed clinician, that it is safe for severe or complex conditions, or that any chatbot would produce the same outcome. It is one early study, not a final verdict.
Is Therabot safe to use in a crisis?
No. Therabot was studied as a supervised research tool, not as an emergency service, and it is not publicly available in any case. No AI chatbot is built to handle a mental-health crisis. If you are in danger or thinking about suicide, contact a professional or, in the US, call or text 988 to reach the Suicide and Crisis Lifeline, available 24 hours a day.