This study investigates immersive learning in virtual reality (VR), focusing on emotional engagement, technology efficacy, and AI interaction. Using Structural Equation Modeling (SEM), we explore both self-reported factors (e.g., Technology Acceptance, Emotional Engagement, Presence) and system-derived variables (e.g., GPT Usage, System Behavior). Results show that higher emotional engagement significantly enhances both technology adoption and efficacy. The model demonstrated excellent fit (CFI = 0.988, RMSEA = 0.069), explaining 54.5% to 70.7% of the variance in key constructs. Perceived presence was a strong predictor of adoption and engagement, emphasizing the role of immersive features in supporting learner interaction. A surprising finding revealed that AI interaction, while enhancing emotional engagement, also reduced the sense of presence, suggesting a trade-off between cognitive involvement and immersion. These findings highlight the importance of emotional design and AI interaction in improving engagement and outcomes in VR-based education.