Generative AI use and academic achievement: a cross-sectional study of Czech seventh-grade students

Abstract This study examines how Generative Artificial Intelligence (GenAI) use is associated with academic achievement among Czech seventh-grade students. Grounded in the deep versus surface approaches to learning framework, we distinguish between the frequency of GenAI use and students’ endorsement of reliance on GenAI over understanding, defined as the belief that knowing how to prompt AI effectively is more important than developing one’s own conceptual understanding. Using cross-sectional survey data from a nationally stratified sample of 2,307 students, we estimated structural equation models predicting IRT-scaled achievement in mathematics and Czech language. Frequency of GenAI use was not positively associated with achievement in either subject. In contrast, stronger endorsement of reliance on GenAI over understanding was consistently and negatively associated with achievement, with a more pronounced relationship in Czech language. Neither GenAI use frequency nor reliance significantly moderated the associations between learning-related dispositions and achievement. These findings suggest that the educational implications of GenAI depend less on how often students use AI tools and more on whether AI is approached as a support for understanding or as a substitute for cognitive engagement. The results contribute to debates on epistemic offloading and the role of AI in shaping students’ orientations toward learning.