Large language model applications for real-time clinical mental health assessment: Current potential and future directions.

Abstract

Large language models (LLMs) have shown increasing promise in the mental health field. LLMs are especially well suited to play a role in the labor-intensive, costly process of clinical assessment, as they can interact with a patient or participant directly to conduct a mental health assessment. We conducted a preregistered scoping review to (a) describe the unique capabilities of LLMs for clinical assessment, (b) determine the current state of the field in applying LLMs to directly assess patient/participant mental health (including screening, diagnosis, and monitoring of symptoms), and (c) highlight future research to facilitate the application of LLMs. We included work published in both Chinese and English. Only 10 studies met criteria for direct LLM-based mental health assessment. The evidence base was recent and heterogeneous: Four studies focused primarily on diagnostic interviewing or classification, five on symptom or severity assessment, and one on task-based multimodal depression assessment. Studies varied across text, voice, and multimodal formats, and depression was the dominant target. Across studies, stronger performance tended to be reported in tools that used structured interviewing logic, domain-specific adaptation, and clinically anchored reference standards. However, the evidence base remains small, with many studies employing limited validation procedures, and heavily weighted toward early-stage or nonjournal publications. The limited pace of academic validation means that, at present, LLMs are best understood as emerging assessment-support tools rather than replacements for clinical evaluation. (PsycInfo Database Record (c) 2026 APA, all rights reserved).