AI-Driven Adaptivity in Immersive Learning Environments: Towards a Unified Framework for Personalized Virtual Education

Virtual, Augmented, and Mixed Reality (VR/AR/MR) have matured into powerful learning technologies, yet most implementations remain static—offering identical content to all learners. Meanwhile, Artificial Intelligence (AI), particularly large language models (LLMs), enables dynamic adaptation and personalized tutoring. This paper presents a structured thematic synthesis of 28 recent open-access studies published during the early post LLM integration phase of immersive education research. The analysis identifies converging trends across cognitive, affective, and conversational dimensions of adaptivity. Across recent studies, adaptive mechanisms are frequently associated with reported increases in engagement, perceived presence, and short-term learning gains. Building on these findings, we propose the Tri-Layer Adaptive XR Framework (T-AXR) a unified model combining 1) cognitive adaptivity through performance analytics, 2) affective adaptivity through emotion and motivation sensing, and 3) conversational adaptivity via LLM driven tutoring. Integrating AI with XR requires design strategies that carefully balance automation with learner agency, transparency, and ethical considerations. The T-AXR framework offers conceptual guidance for the design of adaptive XR systems that aim to balance real-time personalization with immersive coherence.