Abstract:

Generative artificial intelligence is now widely used in higher education, yet the conditions under which its use relates to learning outcomes remain insufficiently explained. This study developed and tested an integrative model linking contextual conditions, experiential and regulatory antecedents, technology perceptions, artificial intelligence use, learning motivation, learning satisfaction, and learning outcomes. Artificial intelligence use is conceptualized as students’ behavioral engagement with generative AI tools (e.g., ChatGPT and similar large language model platforms) in academic contexts. In this study, AI use refers specifically to students’ application of generative AI for academic activities such as research support, problem solving, content generation, and routine study tasks, reflecting how AI is embedded in everyday learning practices. A quantitative cross-sectional survey was conducted with 380 higher education students in Kosovo, where generative AI adoption is predominantly student-driven and primarily centered on conversational AI platforms, and the data were analyzed using covariance-based structural equation modeling. AI use was operationalized as a multidimensional behavioral construct capturing frequency of use, time spent interacting with AI systems, and the extent to which these tools are integrated into task-based academic activities.

The findings showed that interaction quality and cognitive absorption were strong predictors of ease of use, while self-regulated learning and instructor support strongly predicted usefulness and also contributed to artificial intelligence use. Ease of use increased usefulness, and both perceptions predicted use. Artificial intelligence use emerged as the central behavioral mechanism, showing a strong association with learning satisfaction and a direct positive effect on learning outcomes. Learning motivation and learning satisfaction also contributed to learning outcomes, and the results supported partial mediation rather than full mediation. The model demonstrated substantial explanatory power for learning outcomes, indicating that the proposed framework captures a meaningful share of the variance in student learning. Two hypothesized links were not supported: usefulness did not directly predict learning satisfaction, and learning motivation did not directly predict learning satisfaction once actual use was included. The study shows that the educational value of generative artificial intelligence depends less on adoption alone than on the contextual, experiential, regulatory, and evaluative conditions that convert use into learning.

Keywords:

generative artificial intelligence, higher education, technology acceptance, self-regulated learning, cognitive absorption, perceived interaction quality, perceived instructor support, learning motivation, learning satisfaction, learning outcomes.