Researchers at Dartmouth College built Therabot, a chatbot trained and supervised specifically for therapy use, and ran what they describe as the first randomized controlled trial of a generative AI therapy tool, enrolling 210 adults with depression, anxiety or risk of an eating disorder.
Participants were randomly assigned to four weeks of Therabot or a waitlist. Among those with depression, scores on a standard symptom scale fell by an average of 6.13 points with Therabot versus 2.63 points on the waitlist, and by eight weeks Therabot users were down about 51 percent from baseline, with smaller gains for anxiety and eating concerns.
Users rated their working relationship with Therabot close to how patients typically rate a human therapist, and the improvements held up at a follow-up check weeks later. The trial was run by the team that built the tool, was short at four weeks, and compared Therabot only to a waitlist rather than to an active human-delivered therapy.

Read the original study
Four weeks of texting, a clear gap over the waitlist.
Source
- Heinz et al., NEJM AI, 2025. https://doi.org/10.1056/AIoa2400802