Your AI Visibility Score Is Probably an Artifact
A 1,790-query audit finds most AI-visibility scores are artifacts, exposing why blending branded/unbranded prompts and single-engine scores mislead teams.
A six-week audit spanning 1,790 measurements across five AI engines found that most 'AI visibility' dashboards report statistical artifacts rather than genuine discovery signals. One audit covered a single company's own prompts; another covered 26 web-scraping vendors validated against an independent 10,000-mention dataset.
The biggest distortion comes from blending branded prompts ('is X any good') with unbranded ones ('best tool for Y') into a single average. Stripping branded prompts from one dashboard dropped a reported 35% visibility score and #1 ranking to a flat zero. Recall and discovery are different problems, and blended scores hide the one that actually predicts new customers.
A second audit of the web-scraping category showed one vendor ranked #1 on Perplexity, #3 on Google AI Overviews, and tied for third on ChatGPT search — same vendor, same month, three different 'truths' depending on which engine's citation supply chain you sample. A single blended visibility score averages across engines that structurally disagree, producing a number nobody can act on.
On-site GEO tactics like llms.txt showed zero measurable effect, since citations still trace back to third-party sources (Reddit, GitHub, listicles) rather than vendor blogs or metadata files. The author proposes four numbers to track instead: unbranded organic citation share per engine, branded recall tracked separately, a competitor citation-gap list weighted by CPC rather than search volume, and the actual domains AI engines cite on gap queries.