Immersive Virtual Reality (VR) environments create new opportunities for interactive learning by integrating embodied control, real-time feedback, and spatial attention. To understand how learners engage with these environments, we use gaze-ray casting—a lightweight proxy for attention based on headset orientation—to examine how visual focus on avatar gestures shapes task behaviour and perceived system interactivity. Our findings show that gaze predicted system interaction ( \(\beta = 0.726, {\text {R}}^2 = 0.61\) ), but this effect only enhanced presence for learners with high technological efficiency (interaction \(\beta = 0.59\) ). Crucially, gaze only predicted task activity when the UI overlays did not compete with avatar gestures. When interface elements occluded gesture space, gaze engagement dropped, and presence declined. Nonlinear analyses also revealed a tipping point: when both gaze and efficacy were high, dizziness reports spiked, an overload pattern missed in linear models. These results suggest that gaze-based engagement fosters immersion only when learners can fluently interpret system cues. Without adaptive support, active users may encounter strain rather than flow. Gaze-ray data, thus, offer a scalable diagnostic for detecting alignment, or friction, between learner attention and system design. Designing for immersive learning requires synchronising visual demands with user confidence and capacity, ensuring that interaction remains a source of motivation, not disruption.