Abstract
Randomized evidence for purpose-built generative AI mental health support remains limited. In this preregistered trial, 1,137 unscreened university students were assigned to 6 weeks of Flourish or care as usual. All three primary endpoints were met, with more favorable trajectories for perceived stress (β = -0.03, p = 0.010), depressive symptoms (β = -0.03, p = 0.006), and anxiety symptoms (β = -0.04, p = 0.001). Benefits were observed across baseline severity, with less deterioration at low severity and greater improvement at elevated severity. Among participants with elevated symptoms, improvements were larger with Flourish for depression (d = 1.03 vs 0.56) and anxiety (d = 0.91 vs 0.60), with meaningful improvements of 52.3% vs 31.7% and 50.6% vs 38.1%, respectively. Reductions in perceived stress emerged as one potential pathway to broader improvements in depressive and anxiety symptoms. These findings support scalable first-line AI support spanning prevention and early intervention.