Seeking Late Night Life Lines: Experiences of Conversational AI Use in Mental Health Crisis

Key Findings

• — 67.9% (36 of 53) of respondents used ChatGPT in their crisis moment; 88.6% (47) said conversational AI was always or often available, whereas 52.8% had never tried a helpline such as 988 or Crisis Text Line.
• — Reasons for turning to AI: no access to a professional or peer at the time, fear of human judgement (33% of respondents) and fear of being a burden (26%); uses were mainly venting, advice about the specific situation and information seeking.
• — 79.2% (42) reported being extremely or somewhat satisfied with the AI interaction (mean 4.11 of 5, SD 0.95).
• — The sample skewed young (39.6% aged 18 to 25) and lower income (46% of those disclosing income were below the US median); 6 of the 16 expert interviewees were practising clinicians.
• — Experts held that human-to-human connection is an essential positive action in a crisis; the authors recommend designing conversational agents as bridges toward human connection rather than ends in themselves.

Methodology Notes

Survey (n = 53) of US adults who reported using a conversational AI agent during a self-defined mental-health crisis, recruited through Mental Health America's website and newsletter in May and June 2025; 16 expert interviews; qualitative coding with the stages-of-change model as interpretive lens. Limitations: self-selected sample, small n, retrospective self-report, no transcripts. Published in the FAccT 2026 proceedings (25 June 2026, DOI 10.1145/3805689.3812256); dl.acm.org returns HTTP 403 to this workspace, so the record was verified from the Crossref metadata (title, authors, date, venue) and the authors' arXiv preprint 2512.23859 (29 December 2025; PDF read).

Sources