Therabot

Randomized Trial of a Generative AI Chatbot for Mental Health Treatment

Patient / Population Intervention / Exposition Comparaison Critère de jugement

In 210 adults with clinically significant depressive or generalised anxiety symptoms or at clinically high risk for feeding and eating disorders, four weeks of Therabot, an expert-fine-tuned generative-AI therapy chatbot reduced symptoms at 4 and 8 weeks compared with a waitlist control.

N
210
Schéma
National waitlist-controlled RCT (Dartmouth College)
Critère
Symptom change from baseline to 4 weeks (post-intervention) and 8 weeks (follow-up) for MDD, GAD and CHR-FED
Pertinence
2
RésultatMDD symptom change −6.13 vs −2.63 at 4 wk and −7.93 vs −4.22 at 8 wk (d=0.845–0.903); GAD −2.32 vs −0.13 and −3.18 vs −1.11 (d=0.794–0.840); CHR-FED −9.83 vs −1.66 and −10.23 vs −3.70 (d=0.627–0.819). Average use >6 h; alliance rated comparable to human therapists.
Heinz MV, et al. NEJM AI. 2025;2(4):AIoa2400802. 10.1056/AIoa2400802
Discussion et critique

The first RCT of a fully generative-AI therapy chatbot — large effect sizes on clinical-level symptoms, and a therapeutic alliance participants rated comparable to human therapists. A waitlist comparator inflates effects relative to an active or attention control, follow-up was only eight weeks, outcomes were self-reported, and the safety of unsupervised generative AI in people with mental illness is not addressed in the abstract.