Background:Chatbots powered by generative AI (Gen-AI) hold promise for building highly personalized, effective mental health treatments at scale, while also addressing existing user engagement and retention issues common among digital therapeutics. We present the first RCT testing an expert-fine-tuned Gen-AI-powered chatbot, Therabot, for mental health treatment.Methods:Participants (N=210) were randomized to a four-week Therabot intervention (N=106) or waitlist control (WLC; N=104). Subjects were clinically stratified into major depressive disorder (MDD), generalized anxiety disorder (GAD), or clinically high risk feeding and eating disorder (CHR-FED) groups using baseline symptom severity. Primary outcomes included disorder-specific symptom changes from baseline to four and eight weeks. Secondary outcomes included user engagement, acceptability, and therapeutic alliance. Cumulative link mixed models examined differential changes pre- to post-intervention and from pre-intervention to follow-up, between the Therabot and WLC groups.Results:The Therabot group showed large and significantly greater reductions in MDD (d = 0.845-0.903), GAD (d = 0.794-0.840), and CHR-FED (d = 0.627-0.819) symptoms relative to controls at post-intervention and follow-up. Therabot was well received and well-utilized (average use >6 hours), and participants rated the therapeutic alliance comparable to human therapists.Conclusions:This is the first RCT demonstrating the effectiveness of a fully Gen-AI therapy chatbot for treating mental health disorders. Results are promising for MDD, GAD, and CHR-FED symptom reduction. Participants were engaged with Therabot, reported exceptional therapeutic alliance, and rated the intervention highly. Fine-tuned Gen-AI chatbots are a feasible method for creating scalable, personalized interventions in mental health.
Recent advances in large language models (LLMs) create opportunities to expand access to mental healthcare. An emerging body of work focuses on LLM-based cognitive restructuring (CR), yet existing systems are often limited to generating alternative beliefs or engaging in short, scripted exchanges, offering little of the sustained interactivity required for effective restructuring. In this work, we present CRBot , an LLM-powered chatbot designed with mental health professionals to deliver CR. We conducted a two-week user study with 19 participants, complemented by expert review of conversation transcripts. Results show that CRBot can scaffold the core steps of CR, support self-reflection through Socratic questioning, and provide effective psychoeducation. At the same time, challenges arose, including rigidity, oversimplified questioning, pathologization of appropriate emotions, and misinterpretation of nuanced context. We conclude with design implications for building more interactive, flexible, and ethically robust LLM-enabled CR systems.