What an AI Well-Being App Did and Didn't Do for 486 College Students
Nearly every patient I see has an app graveyard on their phone. Sleep trackers, meditation timers, mood loggers, symptom diaries. The story is usually the same: two good weeks, and then the notifications become one more thing to swipe away. So when a well-being app turns up in NEJM AI with a randomized controlled trial attached, I want to know what it changed.
What the Flourish trial found
A team from Flourish Science, Stanford, Chapman University, the University of Washington, Foothill College, and Harvard Business School ran a six-week randomized controlled trial across three US campuses in the fall of 2024 [1]. They enrolled 486 undergraduates and randomly assigned each one either to use an app called Flourish at least twice a week, or to carry on with whatever campus resources already existed. Nobody was screened for mental health symptoms first. That was deliberate. The app was designed to support well-being early, before distress escalates, rather than to treat it after the fact.
By week six, students with the app reported more resilience, a stronger sense of belonging on campus, and less loneliness than the control group.
A second result is more interesting. On several measures the app group didn't improve at all. The control group declined, and the app group held steady. Positive mood, mindfulness, and overall flourishing all fell over the semester in students without the app, and stayed flat in the students who had it. Staying steady through a hard stretch is a different kind of benefit than improving, and that difference changes what you should expect the app to do for you.
Depression, anxiety, and stress scores showed no measurable difference between the groups, and the statistics weren't close. Negative mood did not change either. The effects that did appear were small, in the range of d = 0.2 to 0.3. That's small enough that an individual student probably wouldn't notice the change in themselves, and consistent enough across 486 people to register in the numbers.
Two limitations affect how much you can conclude from this result. The comparison group got no app at all, so some of the benefit could come from simply having something to open rather than from anything Flourish does specifically. Several of the authors also work for the company that built the app, which the paper discloses openly.
What other chatbot trials have found
This trial fits a pattern that's been building for a few years. A 2025 systematic review pooled 31 randomized trials covering 29,637 adolescents and young adults and found AI chatbots produced a small-to-moderate reduction in mental distress overall, meaning the average chatbot user ended up modestly better off than the average person without one [2]. The most consistent results came from simpler retrieval-based systems, while the generative ones were rated inconclusive. A separate meta-analysis of 10 trials that same year found chatbots helped with depression and did nothing measurable for anxiety [3]. The Flourish trial also found no effect on anxiety.
A 2026 trial complicates this further. Researchers randomized 185 adults to a purpose-built AI therapy app, to plain ChatGPT, or to nothing at all [4]. Both AI groups beat the do-nothing group on depression scores by a similar margin. The purpose-built therapy app did not outperform general ChatGPT on any outcome measured.
One more data point on the social side: a 2025 trial of 100 Chinese university students using a CBT-based chatbot for seven days found moderate improvements in both depression and loneliness, and again no change in anxiety [5]. Across all five studies, the depression and loneliness signal is fairly steady. The anxiety signal is not.
Why belonging and loneliness improved
What caught my attention is which outcomes changed. Belonging went up and loneliness came down, and resilience improved alongside them. Those are measures of a person's social circumstances and nervous-system state rather than measures of disease.
I work mostly with complex chronic illness: mast cell activation syndrome (MCAS), gut motility problems, post-viral syndromes. Patients tell me the same thing over and over. Symptoms escalate during a bad stretch at work, or after a conflict at home, or after a long isolated winter. I can't run a controlled trial on that observation, so I treat it as pattern recognition rather than evidence. It still affects how I interpret a study like this one.
The activities inside Flourish are not new. Journaling, gratitude practice, cognitive reappraisal, mindfulness, and social-connection exercises have been in the literature for decades. Students averaged about three and a half active days a week and roughly three and a half activities a week beyond the chat feature. What the AI changed was the timing and the personalization of the nudge. The intervention underneath is old.
That's what I'd take from it. The technology is a delivery mechanism for practices that already had evidence behind them. If those practices help you, which app you pick is less important than whether something in your week already directs you toward other people.
Should you use a well-being app?
Treat a well-being app as support for human connection rather than a replacement for it. Every activity in this particular app directed the user toward someone else, and connection is the factor that keeps appearing across these trials.
Set your expectation at "buffers a hard stretch" rather than "changes how I feel." That's what the data supports, and it's a reasonable thing to want.
If you're dealing with diagnosed depression or anxiety, this class of tool has shown little for anxiety across several trials now. That's a conversation for a clinician who knows your history.
One safety note. A 2025 evaluation tested 29 AI mental health chatbots against standardized prompts simulating escalating suicide risk [6]. The researchers found that none of the 29 met their criteria for an adequate response, and that roughly half were inadequate outright, most often for failing to provide emergency contact information. If you or someone close to you is in crisis, reach for a person or a crisis line rather than a chatbot.
References
Cachia JYA, Zhao X, Hunter J, Wu D, Lin E, De Freitas J. AI for Proactive Mental Health: A Multi-Institutional, Longitudinal Randomized Controlled Trial. NEJM AI. 2026. DOI: 10.1056/AIoa2501293. Original article: https://ai.nejm.org/doi/full/10.1056/AIoa2501293
Feng X, Tian L, Ho GWK, Yorke J, Hui V. The Effectiveness of AI Chatbots in Alleviating Mental Distress and Promoting Health Behaviors Among Adolescents and Young Adults: Systematic Review and Meta-Analysis. J Med Internet Res. 2025;27:e79850. PMID: 41313175. DOI: 10.2196/79850
Chen TH, Chu G, Pan RH, Ma WF. Effectiveness of mental health chatbots in depression and anxiety for adolescents and young adults: a meta-analysis of randomized controlled trials. Expert Rev Med Devices. 2025;22(3):233-241. PMID: 39935147. DOI: 10.1080/17434440.2025.2466742
Kuta B, Novak L, Zidkova R, Furstova J, Malinakova K, De Winter A, Husek V. Effectiveness of a Fully Automated Mobile Therapeutic Versus a General Chatbot in Reducing Depression and Anxiety and Improving Well-Being: Feasibility Randomized Controlled Trial. JMIR Ment Health. 2026;13:e82642. PMID: 42018980. DOI: 10.2196/82642
Wang Y, Li X, Zhang Q, Yeung D, Wu Y. Effect of a Cognitive Behavioral Therapy-Based AI Chatbot on Depression and Loneliness in Chinese University Students: Randomized Controlled Trial With Financial Stress Moderation. JMIR Mhealth Uhealth. 2025;13:e63806. PMID: 40882177. DOI: 10.2196/63806
Pichowicz W, Kotas M, Piotrowski P. Performance of mental health chatbot agents in detecting and managing suicidal ideation. Sci Rep. 2025;15:31652. PMID: 40866537. DOI: 10.1038/s41598-025-17242-4
Supporting references retrieved from PubMed.
This article is educational and is not medical advice. It doesn't replace working with a clinician who knows your full history. If you're struggling with your mental health, please talk to a licensed provider or a crisis line. I'd rather you have a person than an app.
If this resonates with what you're experiencing and you'd like to explore a naturopathic approach, book a consultation with our clinic.
Comments