Linking AI Literacy, Self-Efficacy, Attitudes, and Achievement: A Mixed-Methods Study on the Moderating Role of AI Usage and Study Year
This study addresses a key gap in the AI in education literature by examining whether AI literacy, attitudes towards AI, and AI self-efficacy are related to academic achievement and whether their effects vary by duration of AI use and study year among undergraduates at an institution in Ho Chi Minh City, Vietnam.
The rapid growth of AI in higher education presents both opportunities and risks for student learning; however, empirical evidence on AI competencies and achievement remains limited, particularly in emerging contexts such as Vietnam. Drawing on Control-Value Theory, this study focuses on how students’ perceived control (skills, self-efficacy) and value (attitudes) in using AI tools relate to their academic outcomes.
This study employed a mixed-methods design that combined a cross-sectional survey of 376 undergraduates in Ho Chi Minh City, Vietnam, with qualitative analysis of open-ended responses from eight students. Quantitative data were analyzed using Spearman correlations, ordinal logistic regression, and moderation tests, while qualitative data were examined through template analysis to explain how and why students use generative AI tools in their learning.
The study integrates Control-Value Theory with AI-use competencies and tests two under-examined moderators (duration of AI use, years of study). It complements statistical results with qualitative insights that explain when and why AI competencies translate into achievement.
AI self-efficacy was the only significant predictor of a higher GPA category. AI literacy showed small positive correlations with GPA, whereas attitudes toward AI were not directly related to achievement. Moderation analyses indicated diminishing returns, as the associations between AI literacy and AI self-efficacy were stronger for lighter users and first-year students, and weaker with heavier use and in later years. Qualitative themes highlighted AI as a scaffold for summarizing, idea generation, and drafting; key frictions were accuracy, Vietnamese expression quality, and prompting skills.
Prioritize building AI self-efficacy with low-stakes practice, explicit error-spotting, claim-checking, and revision routines. Use a phased approach: offer light support for beginners, and in advanced years, require records of prompts and sources, comparative prompting, and clear citation and verification to reduce automation bias.
Pursue longitudinal or experimental designs to test causal effects of self-efficacy and verification training and compare domain-specific contexts where task complexity and error costs differ.
Practical guidance on the competent and critical use of AI can support the equitable, ethical, and effective integration of AI in higher education, helping students turn access to AI into real learning gains.
Future studies can expand the scope of this research by extending it to diverse institutions, majors, and incorporating interviews to deepen insight into when, how, and why AI use builds or undermines self-efficacy, as well as how this connects to students’ attitudes and perceived learning.
Back