Current approaches to AI visibility measurement — variously branded as GEO (Generative Engine Optimization), AEO (Answer Engine Optimization), or LLMO (Large Language Model Optimization) — converge on a single architecture: audit a domain, aggregate signals, return a score between 0 and 100. This paper argues that score-based AI visibility measurement is a category error. Large language models do not represent entities uniformly; they exhibit entity-specific bias patterns that vary by recognition level, geographic context, and training data density. Two structurally distinct failure modes emerge: citation invisibility in low-recognition entities, and a counterintuitive Brand Hallucination Paradox in high-recognition entities, where elevated citation rates coincide with elevated factual error rates. We introduce the Per-Entity Bias Map as the correct unit of analysis, the Progressive Audit-Chain as a three-stage measurement protocol, and the Dynamic Dimension Architecture as a goal-sensitive scoring framework. We further identify CEE AI Visibility Coordination Failure — a structural regional problem that individual optimization cannot resolve. Six open empirical gaps are identified. This paper claims these as named research contributions and proposes a validation protocol for each.
Zoltan Varga (Tue,) studied this question.