Bridging the Gap in Immersive Analytics: Visualizing Unstructured Data in the Era of Generative AI

The exponential growth of unstructured textual data poses a fundamental challenge for analysts relying on traditional displays. While Immersive Analytics (IA) offers an infinite canvas for high-dimensional data exploration, current approaches struggle to handle the scale and probabilistic nature of insights derived from Generative AI. This paper identifies the technological and ergonomic gaps in visualizing large-scale knowledge graphs extracted from document collections via Large Language Models (LLMs) in Extended Reality (XR). We highlight that transforming text into 3D spatial networks introduces a cognitive bottleneck characterized by severe visual clutter, temporal disorientation, and the epistemic uncertainty caused by AI hallucinations. We argue that simply rendering massive datasets in 3D exacerbates cognitive load rather than alleviating it. To address these barriers, we propose a framework for Intelligent Adaptive Visualization. This approach advocates for dynamic layout algorithms to reduce clutter, uncertainty-aware attention guidance to prioritize high-confidence insights, and deep provenance mechanisms for trust calibration. We conclude that bridging the gap between algorithmic extraction and human perception requires immersive systems to shift from passive data representation to active, intelligent cognitive assistance.