A bibliometric analysis of large language models in mental health research

Abstract

Background Mental health disorders constitute a critical global health challenge, accounting for nearly 30% of the worldwide non-fatal disease burden and resulting in annual productivity losses exceeding $1 trillion. Despite increased awareness and investment, significant barriers, such as limited access to care, societal stigma, and a shortage of trained professionals, continue to obstruct effective service delivery. Recently, LLMs have emerged as transformative tools in mental health research and practice, offering potential applications in diagnostics, therapeutic support, and patient engagement. Objectives This study aims to provide comprehensive bibliometric analysis of research on LLMs in mental health from 2020 to 2025. Specifically, it examines publication growth and temporal trends to capture the evolution of scholarly interest, assesses country- and institution-level contributions to highlight geographic and organizational patterns, and analyzes the distribution of publications across key academic sources. In addition, the study explores prevailing research themes and their evolution over time through keyword analysis and conceptual mapping. Methods To ensure comprehensive coverage of the literature, searches were conducted across seven major sources: Web of Science, Scopus, PubMed, Dimensions, OpenAlex, Lens, and the Cochrane Library. The combined datasets comprised 426 articles, which provided a foundation for subsequent analyses. A bibliometric approach was then employed to analyze the collected data, focusing on publication trends, collaboration networks, and thematic evolution within the field. Results The analysis revealed a rapid acceleration in scholarly output, with a compound annual growth rate of 140%, driven by advancements in models such as GPT-3 and GPT-4, alongside strategic funding and industry initiatives. The geographic distribution of research demonstrated a significant imbalance, with high-income countries, particularly the United States, dominating the field, while low- and middle-income countries were underrepresented. Thematic evolution showed a transition from foundational research (2020–2022) to applied studies with emerging topics including chatbots, multilingual models, and therapeutic applications. Conclusions The findings underscore the need for more inclusive and globally integrated research efforts to fully realize the potential of LLMs in mental health care. Future research must prioritize equity, strengthen pathways for clinical translation, and develop robust ethical and evaluation frameworks. Policymakers, funders, and clinicians must champion interdisciplinary, ethically guided research to ensure the safe, effective, and socially responsible integration of LLMs into mental health systems.