Yulias Prihatmoko, Henry Praherdhiono, Nunung Nindigraha, Kevin Herdinata Cahyadi Firdaus, Deka Dyah Utami
Artificial intelligence (AI) integration in higher education has shown promising potential, yet empirical findings on its actual contribution to student learning performance remain inconsistent. Most existing studies focus on comparing AI-supported and non-AI-supported instruction, while little attention has been given to variations within the same learning context. This study aims to analyze the impact of AI integration on student performance using a within-class meta-analytic design, with particular attention to the moderating role of AI usage duration. Data were collected from 135 undergraduate students enrolled in an Educational Technology program who experienced uniform AI-based instructional interventions. Statistical analyses included Welch's ANOVA and linear regression to examine whether weekly AI usage hours predicted academic outcomes. Results indicated no significant differences in performance across AI usage groups (p > 0.05), and duration of use was not a significant predictor of final grades. However, descriptive findings revealed high engagement with AI tools such as ChatGPT, Gemini, and Canva, primarily for summarizing readings, retrieving references, and answering content-related questions. These findings suggest that the quality of AI integration into learning design is more critical than the duration of usage. The study contributes to theory by highlighting the role of AI as cognitive support rather than performance determinant and provides practical implications for educators to focus on meaningful AI-based instructional design. © 2025 IEEE.
Universitas Negeri Malang, Dept of. Educational Technology, Malang, Indonesia
