Uses WildChat
Prevalence of Mental Health Discussions in Publicly Available Generative AI Conversations
JAMA Network Open · Published
Studies how the definition of a mental-health conversation changes prevalence estimates in WildChat-4.8M. A classifier scores topicality, help-seeking intent, clinical language, and affective risk, with validation against independently coded conversations.
Data used
620,699 conversations from WildChat-4.8M.
Findings
The estimated share of mental-health conversations is 0.21% when the definition requires explicit help-seeking for an identifiable person, and 4.90% when any mental-health topicality qualifies.
Methods: corpus, four classification domains, and 370-conversation validation sample. Results: comparison of definitions. Read the source.
Dates and versions
Journal publication date. Publication record.
Published research letter.
Cite this paper
Ryan K. McBain, Li Ang Zhang, Alyssa Burnett, Jonathan H. Cantor, Hao Yu. 2026. Prevalence of Mental Health Discussions in Publicly Available Generative AI Conversations. JAMA Network Open.
Show BibTeX
@article{mcbain2026mentalhealth,
title = {{Prevalence of Mental Health Discussions in Publicly Available Generative AI Conversations}},
author = {McBain, Ryan K. and Zhang, Li Ang and Burnett, Alyssa and Cantor, Jonathan H. and Yu, Hao},
year = {2026},
journal = {JAMA Network Open},
publisher = {American Medical Association},
volume = {9},
number = {8},
pages = {e2630635},
doi = {10.1001/jamanetworkopen.2026.30635},
url = {https://doi.org/10.1001/jamanetworkopen.2026.30635}
}