Large language models for digital mental health: an HCI-centered scoping review

Abstract

Large language models (LLMs) are rapidly reshaping digital mental health, yet how these systems are designed, used, and evaluated remains poorly characterized. We conducted an HCI-centered scoping review of 84 studies examining mental health tasks, stakeholders, interaction paradigms, foundation models, evaluation methods, and outcome measurement. Counseling and support and self-help and well-being accounted for 73.8% of studies, while clinician support and crisis and risk support remained uncommon. Mixed-methods user studies predominated (42.9%), with only three randomized controlled trials and eight studies classified as longitudinal or field evaluations. Technical and response performance was assessed in 81.0% of studies, compared with 29.8% assessing mental health symptoms and clinical outcomes and 7.1% including longitudinal follow-up. Outcome profiles differed across intended tasks. Building on these findings and established evaluation guidance, we propose a task-oriented framework aligning evaluation domains, measurement approaches, and assessment timing with intended use.