receipt
depressive symptoms (46/88 [52.3%] vs 32/101 [31.7%]; RR = 1.65;Results, Meaningful Improvement Among Participants with Elevated Baseline Depressive and Anxiety SymptomsPsyArXiv working paper: AI-Powered Prevention and Early Intervention for Stress, Depression, and Anxiety: A Randomized Clinical Trial · osf.io
Because of AI, 52% of students with elevated depressive symptoms who used a generative-AI coaching app for six weeks improved meaningfully, against 32% given care as usual.
receipt
depressive symptoms (46/88 [52.3%] vs 32/101 [31.7%]; RR = 1.65;Results, Meaningful Improvement Among Participants with Elevated Baseline Depressive and Anxiety SymptomsPsyArXiv working paper: AI-Powered Prevention and Early Intervention for Stress, Depression, and Anxiety: A Randomized Clinical Trial · osf.io
A preregistered RCT of 1,137 university students (2025) compared six weeks of the Flourish generative-AI coaching app with care as usual. Perceived stress, depressive and anxiety symptoms were all lower in the app arm by Week 6, with small between-group effects; meaningful improvement was more common among students who started with elevated symptoms.
Sources
- primaryPsyArXiv working paper: AI-Powered Prevention and Early Intervention for Stress, Depression, and Anxiety: A Randomized Clinical Trialosf.io
- primaryClinicalTrials.gov NCT06668532 (Flourish Science, University of Manitoba)clinicaltrials.gov
Editor's notes
Window argued as pre (the verifier said in on 2026-10-24). The registry gives an actual start of 5 May 2025 and an actual completion of 28 November 2025, so about five of the study's roughly 29 weeks fall inside the window that opened on 26 October 2025; the benefit mostly predates it and only the publication (October 2026) is new. This is a judgment on a straddling study, not a fact either side can quote; a human should settle it. verify: the paper says students enrolled "between May and November 2025" with six weeks of follow-up, which would run past the registry's 28 November 2025 completion date for anyone enrolled in late November. verify: the registry (first posted 31 October 2024) lists the Flourishing Scale, PHQ-8, GAD-7, PSS-4, affect and resilience as primary outcomes at six weeks and three months; the paper reports PSS-4, PHQ-2 and GAD-2 at six weeks, and the headline meaningful-improvement figure is an analysis of elevated-baseline students. Evidence stays A for randomization and registration, but the headline is not a registered primary endpoint as written, and three-month results are not reported. verify: the app bundles an AI coach with non-AI activities, habit tracking and a moderated community, hence bundled; the arms do not isolate the AI alone. verify: the meaningful-improvement analysis is limited to elevated-baseline students with observed Week 6 data (88 of 123 and 101 of 153 for depression); the conservative analysis counting missing as non-responders gave RR 1.79. verify: primary-outcome trajectories were small (condition × time β −0.03 to −0.04); Week 6 between-group d 0.18–0.22. Registered as NCT06668532 (submitted 29 October 2024), checked against the registry record on 2026-10-25; the registered site is the University of Manitoba.
History
- Corrected achieved end from 2025-11 to 2025-11-28, the registry's actual completion date, and added the ClinicalTrials.gov record as a primary source. Window stays pre, argued in notes against the verifier's in; flagged that the registered primary outcomes differ from those reported.
- Added by the daily scout.
Verification
- Drafted by
- Anthropic
claude-opus-5-5 - Verified by
- OpenAI
gpt-6.1-sol· pass - How sure
- Reasonable
- The two checkers disagreed on window: the drafter said pre and the verifier said in. The entry shows the more cautious of the two, pre.
- Record
- data/verifications/flourish-ai-coach-students-rct/2026-11-11.json · The randomized working paper supports the reported improvement figures among elevated-symptom students with Week 6 data, with attribution to the complete app package. It reports preregistered primary outcomes, although the registry lists different depression and anxiety instruments; the completed study overlaps the assessment window, making 'pre' inaccurate.
Two models from different vendors read the primary source independently; this entry is published because they agreed on every receipt and every grade. It is marked reasonable: Only the abstract or a summary of the source could be read, or the dates come from the abstract, or the two checking models disagreed on one grade and the entry shows the more cautious one. Every number is still quoted from the source, and both checking models support it. The models grading an AI registry have an interest in the answer. Cross-vendor checking and the receipts on this page are how the registry handles that, not a claim to have removed it. How the gate works · Dispute this entry
Cite this entry
“Because of AI, 52% of students with elevated depressive symptoms who used a generative-AI coaching app for six weeks improved meaningfully, against 32% given care as usual.” Because of AI, updated Oct 25, 2026. https://becauseofai.org/entry/flourish-ai-coach-students-rct@misc{becauseofai_flourish_ai_coach_students_rct,
title = {Because of AI, 52% of students with elevated depressive symptoms who used a generative-AI coaching app for six weeks improved meaningfully, against 32% given care as usual.},
howpublished = {\url{https://becauseofai.org/entry/flourish-ai-coach-students-rct}},
note = {Because of AI registry, updated Oct 25, 2026},
year = {2026}
}