The Dartmouth AI Therapy Study: What a 51% Depression Reduction Actually Means

In March 2025, a team at Dartmouth's Geisel School of Medicine published the first randomized controlled trial of a generative AI therapy chatbot in NEJM AI — one of the most respected medical journals in the world. The results made headlines: participants with major depressive disorder saw an average 51% reduction in depression symptoms over 8 weeks.

That's comparable to outcomes from traditional outpatient therapy. And it raises an obvious question: if an AI chatbot can produce these results, what does that mean for how we approach mental health?

The answer doesn't fit a headline — and it matters for anyone evaluating tools like BridgeCalm.

What the study actually found

The Dartmouth trial, published in NEJM AI, enrolled 210 adults diagnosed with major depressive disorder, generalized anxiety disorder, or eating disorder risk. Participants were given access to Therabot, a generative AI chatbot trained on Cognitive Behavioral Therapy (CBT) principles.

Over 8 weeks (4 weeks of active intervention plus 4 weeks of follow-up), participants who used Therabot experienced measurable improvements:

• 51% average reduction in depression symptoms
• Significant reductions in anxiety
• Improvements in eating disorder risk markers
• Average usage of 6+ hours across the study — roughly equivalent to 8 traditional therapy sessions

These results were clinically significant and broadly comparable to traditional outpatient therapy outcomes.

What the study did NOT find

Be precise about what this study tells us — and what it doesn't.

Therabot was studied as a standalone AI therapy tool for people with clinical diagnoses. Participants were not simultaneously seeing a therapist. The chatbot was the intervention.

This is a different model from what BridgeCalm does. BridgeCalm is not trying to replace therapy. Jan, BridgeCalm's wellness companion, is designed to help you practice skills between therapy sessions — the daily check-ins, breathing exercises, thought challenges, and progress tracking that make your actual therapy sessions more productive.

The Dartmouth study proves an important point: AI-guided conversations based on evidence-based therapeutic techniques can produce real clinical improvement. But it studied a scenario where AI was the only support. BridgeCalm operates in a scenario where AI is the daily support, and a human therapist is the clinical support.

Why this study matters for between-session practice

The most relevant finding isn't the 51% number itself — it's how participants used the tool. They engaged in short, frequent interactions over weeks. They practiced CBT techniques through conversation. They built a consistent habit of checking in and working through challenges.

That pattern — brief, daily, structured practice — is exactly what decades of research on therapy homework has validated. A meta-analysis of 23 studies found that patients who practice therapeutic skills between sessions consistently have better outcomes. A 2024 review estimated the causal effect of between-session homework at a medium effect size (d = .53).

The Dartmouth study adds a new dimension to this: AI-guided practice can be an effective delivery mechanism for these between-session exercises. When participants had a conversational partner walking them through CBT techniques daily, they actually did the work. That's the part that most people struggle with on their own.

The homework compliance problem

Here's the uncomfortable context: between 20% and 50% of therapy patients don't complete their homework. Not because they don't want to — but because life gets in the way, the instructions are vague, and there's no one to help when they get stuck at 9 p.m. on a Tuesday.

Therapists have known about this problem for decades. The Beck Institute — founded by the pioneers of CBT — has written extensively about the gap between what patients are asked to practice and what they actually do.

What the Dartmouth study suggests is that the delivery mechanism matters. When a patient has a guided, conversational tool available whenever they need it — not just during a scheduled 50-minute session — engagement goes up. And when engagement goes up, outcomes improve.

How BridgeCalm builds on this research

BridgeCalm takes the insight behind the Dartmouth study and applies it within an existing therapeutic relationship rather than as a replacement for one.

Jan walks you through research-backed exercises daily. But unlike a standalone AI therapy tool, Jan also:

• Sends your therapist a summary before each session, so they know what you practiced and what was hard
• Tracks your mood over weeks and months, not just individual conversations
• Follows the specific therapeutic frameworks your therapist works within (CBT, DBT, ACT, and more)
• Routes you to crisis resources (988 Suicide & Crisis Lifeline, Crisis Text Line) rather than ever attempting to handle a crisis

The Dartmouth study showed that AI-guided therapeutic practice produces real results. BridgeCalm is built on the belief that those results get even better when a human therapist is in the loop.

What the Dartmouth study proves

The Dartmouth Therabot trial is the first large-scale RCT of its kind. It provides the strongest evidence to date that AI-guided, CBT-based conversations can meaningfully reduce depression and anxiety symptoms.

For BridgeCalm, the takeaway is clear: the underlying approach — daily, guided, research-backed practice delivered through AI — is clinically validated. The question isn't whether this kind of tool works. The question is how to build it responsibly, within appropriate clinical boundaries, and in partnership with the therapists who know their patients best.

That's what we're building.

Sources

• Dartmouth College. (2025). "First therapy chatbot trial yields mental health benefits." Dartmouth News
• NEJM AI. (2025). Therabot Randomized Controlled Trial. ai.nejm.org
• Mausbach, B.T., et al. (2010). "The Relationship Between Homework Compliance and Therapy Outcomes: An Updated Meta-Analysis." Cognitive Therapy and Research. PMC2939342
• Ryum, T. & Kazantzis, N. (2024). "Between-Session Homework in Clinical Training and Practice." Clinical Psychology in Europe. PMC11303922
• Tang, W. & Kreindler, D. (2017). "Supporting Homework Compliance in CBT: Essential Features of Mobile Apps." JMIR Mental Health. PMC5481663
• Beck Institute. "What is the Status of 'Homework' in CBT, 50 Years On?" beckinstitute.org