Depression Symptoms and Relational Patterns in 187k ChatGPT Histories
Analyses 187,093 ChatGPT conversations donated by 766 participants who completed the PHQ-8, comparing users below and at or above the moderate-symptom threshold on topics, timing, language and disclosure context.
Publisher
Companion Proceedings of CSCW 2026 (ACM); University of Pennsylvania
Published
10 Oct 2026
Added
1 week ago
Key Findings
- Participants at or above PHQ-8 10 used ChatGPT more for mental-health, interpersonal, loneliness, self-focused and support-seeking conversations, with pronounced late-night and recurring month-level patterns
- Their language contained more first-person singular pronouns and absolutist terms
- They more often engaged ChatGPT in high-disclosure contexts, but professional redirection was not more frequent
- Language-based prediction of symptom level was modest (AUROC 0.591) and insufficient for screening
Methodology Notes
Companion-proceedings paper (poster or extended-abstract format, length not confirmed). Verification: OpenAlex abstract and Crossref record (created 2026-10-09, published 2026-10-10); dl.acm.org walled. Full text not read.
Sources
Authors
Neil Kumar Reddi Sehgal, Dunigan Folk, Lyle Ungar, Sharath Chandra Guntuku
Tags
Cite This
APA
Neil Kumar Reddi Sehgal et al. (2026). Depression Symptoms and Relational Patterns in 187k ChatGPT Histories. Companion Proceedings of CSCW 2026 (ACM); University of Pennsylvania. https://doi.org/10.1145/3785651.3831553