The application of large language models (LLMs) in psychological support for university students: A scoping review

Abstract

Background

University students globally face heightened risks for mental health challenges, yet significant barriers to traditional care persist. Large language models (LLMs) have emerged as a promising technology for providing scalable and accessible psychological support; however, the evidence base remains scattered and requires synthesis.

Objective

This scoping review aimed to systematically map the existing literature on the application of LLMs in providing psychological support to university students, describing the range of applications, summarizing reported outcomes, and identifying key evidence gaps.

Methods

A scoping review was conducted following the Joanna Briggs Institute (JBI) manual and reported per the PRISMA-ScR guidelines. A systematic search of PubMed/MEDLINE, PsycINFO, Scopus, Web of Science, IEEE Xplore, and ACM Digital Library was performed for peer-reviewed literature published between January 2020 and December 2025. Studies involving LLM-powered interfaces (e.g., chatbots) for student psychological support were included. Data were extracted and synthesized narratively.

Results

Of 18 included studies, most were randomized controlled trials (38.9%) published in 2025 (68.4%), with a marked geographical concentration in East Asia (China, 38.9%; South Korea, 16.7%; Japan, 5.6%). This geographic concentration limits the generalizability of findings to other cultural and educational contexts. Studies were categorized into six thematic groups based on primary purpose: direct psychological support interventions (n = 8), assessment tools (n = 2), companionship chatbots (n = 2), educational AI tools with mental health outcomes (n = 2), feasibility studies (n = 3), and hybrid models (n = 1). Interventions were most frequently delivered via Cognitive Behavioral Therapy (CBT, 44.4%) using mobile apps, with GPT-based models and minimal human supervision. Preliminary efficacy was most robust for reducing symptoms of depression (effect sizes Cohen's d 0.30-0.83) and loneliness (d ∼0.60), with mixed results for anxiety. Usability and acceptability were generally high, with key benefits being 24/7 accessibility, anonymity, and emotional support. Significant challenges included technical limitations, lack of personalization, linguistic/cultural barriers, and unresolved ethical concerns regarding crisis management and algorithmic bias. Major evidence gaps were identified concerning long-term effectiveness, hybrid human-AI models, and research with underrepresented populations (e.g., international students).

Conclusions

LLM-powered tools, particularly CBT-based chatbots, show promising feasibility and preliminary efficacy for addressing student depression and loneliness, though methodological limitations in the underlying evidence warrant interpretive caution and unresolved questions of data governance and institutional responsibility remain. However, these findings derive largely from studies conducted in East Asia (notably China), which limits their generalizability to other cultural and educational contexts. The field is nascent and characterized by technological optimism counterbalanced by significant practical and ethical limitations. Future development must prioritize complementary, equitable, and ethically rigorous integration into holistic student support ecosystems across diverse global settings. Future development must prioritize complementary, equitable, and ethically rigorous integration into holistic student support ecosystems rather than aiming to replace human care.