A generative AI app built to nudge students toward small well-being habits made a measurable difference over half a semester. In a randomized trial at three US colleges, students who got the app reported more positive emotion, more belonging and resilience, and less loneliness than students who did not.
The study was released in November 2025 as Harvard Business School Working Paper 26-030. Its authors are Julie Y. A. Cachia and Xuan Zhao of Flourish Science Inc., John Hunter of Chapman University, Delancey Wu of the University of Washington, Eta Lin of Foothill College and Julian De Freitas of Harvard Business School.
Cachia and Zhao work for and sit on the board of Flourish Science, the company behind the app, and De Freitas advises it and holds shares. The paper discloses this, and notes that the other authors led recruitment and data collection at the three sites.
A six-week trial on three campuses
The app, called Flourish, centres on an AI chatbot named Sunnie. It runs brief emotion check-ins and conversations, then suggests science-based activities such as journaling, listing three good things, giving a compliment or taking a nature walk.
In Fall 2024, 486 undergraduates at Chapman University, the University of Washington and Foothill College were randomly assigned either to get the app, with instructions to use it at least twice a week, or to join a waitlist with their usual campus support. Students filled in surveys at weeks 0, 2, 4 and 6.
Participants used the app on 3.49 days a week on average, more than required. The authors estimate typical use at about 5 to 20 minutes a week.
Boosting and buffering
By weeks 4 and 6, the app group reported higher positive affect than the control group, despite starting at similar levels. Within that, feelings of calm rose for app users, while a general sense of well-being dipped in the control group but held steady with the app.
Loneliness fell over the semester in both groups but dropped faster among app users. Belonging and resilience rose in the app group while staying flat in the control group.
Mindfulness declined among controls but held steady among app users, and flourishing, a broad measure covering meaning, engagement and relationships, showed the same protective pattern.
No change in clinical symptoms
The app did not significantly change negative emotions, depression, anxiety, stress or academic self-efficacy. The authors attribute this to a mostly non-clinical sample, a short trial and light use, and describe the app as a preventive tool rather than a replacement for treatment.
They also note that comparing the app with a waitlist, rather than an active alternative, means some of the benefit could come from general engagement or expectations. Effects did not differ significantly by demographics or baseline mental health.
After six weeks, 58% of app users who answered an open question spontaneously described improvements in their overall well-being.
Study Details:
- Title: AI for Proactive Mental Health: A Multi-Institutional, Longitudinal, Randomized Controlled Trial
- Authors: Julie Y. A. Cachia, Xuan Zhao, John Hunter, Delancey Wu, Eta Lin, Julian De Freitas
- Journal: Harvard Business School Working Paper No. 26-030
- Publication Date: November 2025
- Link: https://www.hbs.edu/ris/Publication%20Files/26-030_d076e0d0-3c7f-4728-a481-ef1fb9f9641b.pdf