Beyond Rankings: Measuring Vendor Visibility in AI-Driven Discovery

Abstract

As large language models replace traditional search engines as primary discovery interfaces, businesses face a fundamental measurement problem: LLM outputs are stochastic. The same query submitted multiple times produces different vendor sets, different framings, and different recommendations — making ranking-based metrics structurally inapplicable.

We introduce a four-dimensional probabilistic visibility model — Inclusion, Stability, Influence, and Coverage (ISIC) — as the replacement measurement framework for AI-driven discovery. Across 27 queries spanning 9 problem areas and 81 GPT-4o responses, results indicate an average Jaccard overlap of 0.03, with 19 of 27 queries showing zero overlap between LLM and Google results — indicating that these two discovery channels direct buyers toward fundamentally different resources.