Investigating the Intrinsic Mechanisms by which AI-driven Educational Games Enhance Student Cognition

Generative artificial intelligence is driving instructional games from static scripts toward intelligent forms that are generative, adaptive, and diagnostic. However, mechanistic evidence is still lacking on what kinds of AI-driven instructional games effectively promote learning and why. Grounded in a “design features - process mechanisms - learning outcomes” framework and targeting deep learning outcomes of understanding and transfer, this study collected 210 questionnaire responses from teachers and students who used AI instructional games during the past semester. Structural equation modeling (SEM) was employed to test pathways by which key design features - clarity of goals and rules, adaptive challenge, diagnostic feedback, and autonomy and control - affect learning outcomes via flow/immersion and cognitive engagement. Results indicate good scale reliability and validity and satisfactory model fit. Path analyses reveal a progressive, interlinked structure among AI instructional game design elements: clarity of goals and rules - diagnostic feedback - adaptive challenge - autonomy and control. These elements in turn enhance cognitive engagement by increasing flow/immersion, and ultimately significantly promote understanding and transfer. Based on these findings, the study offers classroom-oriented design implications: Prioritize transparent goals and rules, provide actionable diagnostic feedback, implement ability-matched adaptive support, and preserve learner autonomy and control to effectively convert immersive experiences into deep cognitive processing and transfer performance.