The AIVI Framework: A Measurement Model for Brand Visibility in Large Language Models

The emergence of Large Language Models (LLMs) has fundamentally changed how users discover information. Instead of retrieving ranked hyperlinks, modern AI systems synthesize responses by integrating information from multiple sources and recommending entities directly within generated text. Consequently, traditional search visibility metrics—including keyword rankings, backlink profiles, and click-through rates—no longer provide sufficient insight into whether a brand is discoverable inside AI-generated answers. This paper introduces the Artificial Intelligence Visibility Index (AIVI), a measurement framework designed to quantify brand visibility across LLM-based retrieval environments. Rather than evaluating websites alone, AIVI measures how consistently an entity is recognized, cited, recommended, and semantically represented by generative AI systems. The proposed framework decomposes AI visibility into six measurable dimensions: Entity Recognition Rate (ERR), Visibility Recall (VR), Citation Consistency (CC), Recommendation Confidence (RC), Prompt Robustness (PR), and Cross-Model Stability (CMS). These dimensions are combined into a normalized composite score that enables quantitative comparison across different language models and prompt variations. Unlike conventional SEO metrics, AIVI focuses on entity-centric visibility within generative retrieval systems, providing a practical methodology for evaluating brand discoverability in conversational AI. The paper defines the conceptual architecture of the framework, presents a scoring methodology, discusses implementation considerations, and outlines validation strategies for future empirical research. The framework is introduced as a measurement model and is accompanied by a small feasibility pilot across three language models and six entities. The pilot demonstrates that all dimensions except Citation Consistency can be computed from collected responses, and surfaces two boundary conditions that any implementation must handle. The paper aims to establish a common vocabulary for evaluating visibility in the emerging field of Generative Engine Optimization (GEO) and to support future standardization efforts in AI-native search measurement.