Mapping the Content and Consistency of Suicidal Thoughts with Large Language Models: A Two-Study Longitudinal Investigation Among Adolescents and Young Adults
Year: 2026
Type: preprint
Study design: Observational Study
Abstract: Suicide is a leading cause of death. Measurement of suicidal thoughts and behaviors (STBs) relies on self-report (e.g., “How strong is your urge to kill yourself right now?”), yet little is known about how individuals translate internal experiences into numerical ratings. We used large language models (LLMs) to examine the content and consistency of suicide urges across severity ratings, between people, and over time. Participants with past-year STBs (Study 1: N=158 adolescents, Study 2: N=202 young adults) completed open-ended questions at baseline and follow-ups. We developed a two-stage topic-modeling pipeline and identified 13 (Study 1) and 14 (Study 2) topics reflecting suicidal thought content (e.g., Family/Friends Reactions, Depressed & Exhausted). Within- and between-person consistency were U-shaped, with highest agreement at endpoints of the scale. These findings highlight the potential of LLMs for advancing measurement of suicide risk and our understanding of how people interpret and respond to self-report rating scales.
Language: English
Subfield: Clinical Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7204846227