Can an app powered by generative AI improve mental health?
A study of nearly 500 college students by Harvard Business School Assistant Professor Julian De Freitas and colleagues suggests that an AI agent focused on strengthening well-being can proactively boost mental health, at least in the short term.
With demand for therapists and mental health support outstripping available resources, many providers are looking to use technology to supplement conventional treatments. Billions of dollars in venture capital have flowed to mental health AI startups, such as Wysa, Slingshot AI, and Woebot Health, in recent years, according to the health tech consultancy Galen Growth.
Undergraduate students from three US universities were recruited for a study on the Flourish app. About half received access; the others were waitlisted. Over six weeks, an AI-based well-being coach named “Sunnie” checked in with students through the app, prompting brief reflections on their emotions and positive behavior change.
De Freitas collaborated with Julie Y. A. Cachia and Xuan Zhao, cofounders of Flourish Science, the seed-stage startup behind the app; John Hunter of Chapman University; Delancey Wu of the University of Washington; and Eta Lin of Foothill College. De Freitas is a Flourish Science shareholder and adviser.
“AI for Proactive Mental Health: A Multi-Institutional, Longitudinal, Randomized Controlled Trial,” published in November 2025, found that the AI app improved students’ mental health by boosting emotional, social, and general well-being, rather than by lowering negative mental health symptoms like anxiety or depression. The app focused on prevention, mindfulness, psychoeducation, and social connections, not diagnosis or treatment.
According to the authors, the paper is the first to show that AI-powered apps can proactively prevent mental health problems, rather than merely react once problems have already emerged and become acute.
The app was designed by psychologists, and students received relevant scientific insights and tailored, evidence-based recommendations for coping, such as reframing stressful situations (cognitive reappraisal), nature walks, and deep breathing. They earned rewards for completing activities and received information on mental health resources.
One limitation of the study is that it measured improvements reported by students using the app compared with those in the waitlist group, so the researchers cannot rule out that students benefited simply because they were using an interactive app. While qualitative analyses suggest that users benefited from specific app features, the paper notes that future research should include an active control group.
AI companion guardrails
Critics of some general-purpose AI and companion apps, including De Freitas, have warned of the potential for damaging manipulation and hallucinations. App makers should consult with psychologists and mental health specialists and create a crisis protocol for escalating potentially serious situations, the authors note.
The findings suggest that an AI well-being coach that’s thoughtfully designed for a specific purpose “can help individuals and communities cultivate resilience, belonging, and collective flourishing in a manner that is early, scalable to the population level, and stigma-free,” the authors write.
AI for Proactive Mental Health: A Multi-Institutional, Longitudinal, Randomized Controlled Trial
Cachia, Julie Y.A., Xuan Zhao, John Hunter, Delancy Wu, Eta Lin, and Julian De Freitas. "AI for Proactive Mental Health: A Multi-Institutional, Longitudinal, Randomized Controlled Trial." Harvard Business School Working Paper, No. 26-030, November 2025.
