More engagement, more improvement: a dose-response study of 100,000 Ash users
A 2026 study of 102,684 Ash users found five engagement patterns and a dose-response link between engagement and improvement in depression, anxiety, and social support.
Across 102,684 Ash users, engagement fell into five distinct patterns, and greater engagement tracked with larger improvements in depression — a dose-response gradient — alongside significant gains in anxiety and social support.
Study type
Observational study (engagement phenotyping)
Published in
arXiv
Partner
New York University, University of Virginia
Citation
Wolfe, E. C., Su, T., Tieleman, O., Hull, T. D., Malgaroli, M., & Stamatis, C. A. (2026). Engagement Phenotypes for a Sample of 102,684 AI Mental Health Chatbot Users and Dose-Response Associations with Clinical Outcomes [Preprint]. arXiv. https://arxiv.org/abs/2605.00275
A 2026 arXiv preprint characterized engagement among 102,684 users of Ash and linked it to clinical outcomes. Five engagement phenotypes emerged, and a dose-response gradient was observed for depression improvement and replicated in a larger model-predicted sample. Significant pre-post improvements appeared for depression (d=-0.51), anxiety (d=-0.57), and social support (d=0.22), and a stronger working alliance predicted greater improvement.
Two-thirds of users (66.9%) had at least one overnight session, and the authors caution against treating raw session counts as the primary engagement metric — different outcomes responded to different dimensions of use.
The pre-post design is observational rather than controlled. The work is offered as naturalistic evidence for the clinical value of purpose-built conversational AI.