Who benefits from LLMs? Prompt engineering competence and equity in higher education

Abstract Large language models (LLMs) are rapidly becoming embedded in higher education, yet their benefits may be unevenly distributed. This study conceptualizes prompt engineering competence (PEC) as an equity-relevant interaction competence and examines whether differences in PEC are associated with perceived academic gains and perceived educational inequity. We adopt a two-stage quantitative design combining a cross-sectional survey with a within-subject prompt-quality experiment. Stage 1 reports a survey of Chinese undergraduates who have used LLMs ( N = 380), measuring PEC, perceived academic performance gains, perceived educational inequity, family digital capital, gaming experience, and a planning/independent-work tendency. Regression and mediation analyses show that PEC was strongly associated with perceived academic gains, while its association with perceived inequity was statistically significant but small. Background resources and learning-related experiences and dispositions explained substantial variance in PEC, especially gaming experience and planning/independent-work tendency, and the association between family digital capital and perceived academic gains was largely indirect through PEC. Stage 2 conducts a within-subject prompt-quality experiment across four higher-education task types and four popular models, comparing basic prompts with a student-oriented five-component refined prompt framework (Role–Goal–Input–Process–Output). Analysis of 400 matched prompt comparisons, comprising 800 model outputs, indicates that the refined prompt scaffold consistently outperforms basic prompts. Overall, the findings identify PEC as a potential competence pathway through which students may differ in their ability to convert LLM access into learning-related benefits, while structured prompt scaffolds may provide a scalable approach to reducing prompt-quality barriers in LLM-supported higher education.