G-STEP: A human-centered framework for the design and deployment of responsible LLM-based mental health chatbots

Abstract

Large language model (LLM)–based chatbots are rapidly emerging as a promising vehicle for delivering companionship, psychoeducation, on-demand mental health support, and clinical workflow automation or augmentation. However, inconsistent or inadequate crisis handling, along with well-documented hallucinations, bias, and opaque accountability of LLMs, pose significant challenges for responsible deployment and clinical use of those chatbots. This paper introduces key streams of research on LLM-based Mental Health Chatbots (LLM-MHCs) and critically examines their benefits and risks. Then, we propose a five-dimensional humancentered framework for responsible design of LLM-MHCs that consists of governance, safety, transparency and explainability, empathy, and personalization (G-STEP). Finally, we outline several directions for future research, including conducting longitudinal randomized clinical trials with patients with mental health conditions to rigorously assess the effectiveness and user experience of LLM-MHCs; developing open benchmarks and comprehensive metrics for evaluating their efficacy, reliability, and equity; and supporting continuous human-in-the-loop and responsible deployment of LLM-MHCs.

Keywords

mental health / large language model (LLM) / chatbots / user safety / risk / governance

Rights & permissions

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.