Beyond the hype: Decoding how generative AI shapes academic achievement in higher education

Abstract

The study investigates how university students’ adoption and usage of ChatGPT, a generative AI large language model (LLM), influences their academic achievement. The study further explores how these relationships differ by academic discipline. Using a cross-sectional online survey, data were collected from 461 students at a Ghanaian public research university. Measures included ChatGPT literacy, self-efficacy, behavioural intention to use a mostly used generative AI LLM, that is ChatGPT, and self-reported academic achievement. Structural equation modelling (SEM) and multi-group analysis (STEM vs. non-STEM) were employed to test hypothesized relationships and examine potential disciplinary differences. Accordingly, four of the hypothesized paths were supported, showing that higher ChatGPT literacy and self-efficacy enhance students’ behavioural intentions to use ChatGPT, which in turn positively relates to academic achievement. Notably, the effect of behavioural intention on academic achievement varied significantly by academic discipline. Students from STEM fields exhibited stronger effects, indicating the disciplinary context shapes how ChatGPT usage translates into academic gains.

The study investigated how university students’ adoption and usage of ChatGPT influences their academic achievement. The study further explored how these relationships differ by gender and field of study. Using an online cross-sectional survey, data were collected from 461 students at a Ghanaian public university. Measures consisted of ChatGPT literacy, self-efficacy, behavioural intention to use ChatGPT and academic achievement. Structural equation modelling and multi-group analysis were employed to test hypothesised relationships and examine potential gender and discipline differences. Findings show that higher ChatGPT literacy and self-efficacy significantly boost students’ behavioural intention to use ChatGPT, which positively impacts academic achievement. Notably, the effect of behavioural intention on achievement was stronger among science, technology, engineering and mathematics (STEM) students, highlighting the role of field of study in shaping ChatGPT’s educational benefits. The study presents a predictive model linking ChatGPT adoption factors to academic success and emphasises the need for targeted support, especially for non-STEM students, to foster equitable artificial intelligence (AI) integration. It offers practical guidance for educators and policymakers to develop discipline-sensitive strategies that enhance ChatGPT literacy and self-efficacy, ensuring more inclusive and effective use of generative artificial intelligence (GenAI) tools in higher education.

Implications for practice or policy:

• University administrators should mandate AI literacy training during orientation to ensure equitable student engagement.
• Non-STEM programme leaders can improve outcomes by embedding discipline-specific AI tasks into core curricula.
• Equity and inclusion officers must monitor demographic disparities in adoption to provide targeted support.
• Institutional policymakers should establish integrity frameworks that balance AI-assisted learning with rigorous assessment.