A randomized trial suggests purpose-built AI may help spot and support emotional strain.
Listen to this article:
Modern life is full of signals. A watch can tell us our heart rate, an app can tell us how many hours we slept, and our fridges can even alert us if we’re running low on eggs. But when emotional strain begins to accumulate, many people get no signal at all. The new paper, “AI-Powered Early Intervention for Depression and Anxiety: A Randomized Clinical Trial,” co-written by HBS AI Institute associate Julian De Freitas, explores whether generative AI can fill that gap. Depression and anxiety are widespread, disabling, and often undertreated, especially among young adults. Can an AI system designed specifically for mental health offer support to students before symptoms escalate, and help those already experiencing these feelings improve?
Key Insight: AI That’s Built to Help
“General-purpose AI tools offer personalization and engagement but lack the structure to deliver either safely and effectively.” [1]
The researchers understand that young people are already using AI for emotional support, but they draw a sharp distinction between general-purpose chatbots (like ChatGPT) and purpose-built mental health tools. General-purpose tools are widely available and can feel responsive, but they have not specifically been designed for the purpose of detecting risk, guiding users through evidence-based techniques, or avoiding harmful conversation patterns. Flourish (https://www.myflourish.ai/), a conversational AI mental health platform built around an AI coach named “Sunnie,” represents a different model. Rather than trying to replace a therapist, it’s designed to be the first stop, a coach that helps someone notice what they’re feeling, understand why, and practice concrete techniques pulled from cognitive and dialectical behavioral therapy, acceptance and commitment therapy, positive psychology, motivational interviewing, and positive and social psychology.
Key Insight: Testing AI Intervention at Scale
“All randomized participants were analyzed under an intention-to-treat approach to preserve the benefits of randomization.” [2]
To evaluate the platform’s efficacy, the researchers ran a parallel-group randomized clinical trial across two academic terms in 2025, involving 1,137 university undergraduates. Participants were enrolled without prior symptom screening and randomized to either the intervention group, which received access to the Flourish app and asked to use it at minimum twice per week for 6 weeks, or a care-as-usual control group that retained standard campus resources. The study tracked depressive and anxiety symptoms biweekly across four survey waves using validated screening tools. Participant retention remained remarkably strong, with 74.6% completing the final 6-week assessment. The participants were analyzed in the groups they were originally randomized into, regardless of whether they fully followed the intervention (the intention-to-treat approach). This preserves randomization because excluding or regrouping participants after assignment could make the groups systematically different.
Key Insight: Dual-Action Results
“Across both severity groups, treatment yielded more beneficial trajectories.” [3]
For students who began with elevated depressive symptoms, use of Flourish substantially increased the rate of meaningful clinical improvement compared with care-as-usual, or 52.3% versus 31.7%. For anxiety, the improvement was smaller but still statistically significant, at 50.6% versus 38.1%. For students who started with low symptoms, the effect looked different but was arguably just as important: they got worse more slowly than the control group. Depressive symptoms crept up over the semester in both groups (a familiar pattern during a demanding academic term) but climbed less among students using Flourish. No adverse events attributable to the intervention were identified across the entire study.
Why This Matters
This study showcases the potential for purpose-built AI: what happens when you build an AI tool for one job, and only one job, instead of asking a general-purpose model to stretch across every possible use? Flourish’s advantage wasn’t a smarter underlying language model; it ran on the same frontier models many companies already use. Its advantage was unique, customized architecture. For executives and business leaders evaluating how to make the most of their AI investments, this is a pattern worth studying.
Bonus
AI in mental health is most promising when it is purpose-built, evidence-based, and designed as a scalable layer of support rather than a substitute for professional care. This research shows how Flourish was associated with meaningful improvement for students with elevated depression and anxiety. Earlier work on Flourish first showed its strength as a proactive well-being tool, helping students build positive affect, resilience, belonging, and social connection. To see how this research began with the science of proactive well-being, check out The Science of Digital Flourishing.
References
[1] Cachia, Julie Y. A., et al., “AI-Powered Early Intervention for Depression and Anxiety: A Randomized Clinical Trial,” PsyArXiv (July 7, 2026): 5. osf.io/preprints/psyarxiv/nwz9q_v5
[2] Cachia et al., “AI-Powered Early Intervention for Depression and Anxiety,” 7.
[3] Cachia et al., “AI-Powered Early Intervention for Depression and Anxiety,” 12.
Meet the Authors
Julian De Freitas is an Assistant Professor of Business Administration in the Marketing Unit and Director of the Ethical Intelligence Lab at Harvard Business School, and Associate HBS AI Institute. His work sits at the nexus of AI, consumer psychology, and ethics.
Additional Authors: Julie Y.A. Cachia, Xuan Zhao, Wenxi Pu, Chi Liao, Merel Keijsers, Rebecca G. Fortgang