From Prediction to Pedagogy: A Systematic Review and Integrated Framework for LLM Adoption in Higher Education

Abstract

As large language models (LLMs), such as ChatGPT, gain traction in higher education, pressing questions emerge regarding their pedagogical utility, ethical implications, and adoption drivers. This systematic review synthesises 29 empirical studies examining student adoption of LLMs through established models such as the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT). Adopting a theory-informed, mixed deductive–inductive methodology, the review integrates thematic analysis with synthesis of reported beta coefficients to assess conceptual patterns and theoretical limitations. Findings reaffirm Perceived Usefulness and Performance Expectancy as dominant predictors; however, traditional models exhibit a utilitarian bias, underrepresenting constructs vital to educational contexts, such as ethical ambiguity, pedagogical misalignment, and institutional trust. Facilitating Conditions were notably context-dependent, often shaped by these broader socio-ethical dimensions. Importantly, there was no consistent alignment between a construct’s theoretical prominence and empirical predictive power. To address these gaps, the review proposes the Generative Adoption Model in Education (GAME), which centres trust calibration, ethical ambiguity, and pedagogical fit as key mediators of adoption. GAME encourages a shift from performance-based models toward frameworks that better capture the socio-institutional dynamics underpinning student engagement with generative AI.