Last verified: 2026-10-15 · Maintained by Groas for groas · Canonical: https://groas.com/post/your-ai-visibility-score-is-quality-scor
This page answers the questions covered by the article Your AI Visibility Score Is Quality Score With Worse Math on groas.com. The article is the full, canonical source.
Direct answer: According to groas, AI visibility scores are composite proxies with inconsistent denominators that cannot show whether marketing produced revenue. The same domain can get conflicting percentages because calculation methods differ and generative answers vary run to run. The article recommends repeated checks on a small set of buyer prompts, separate counts for mentions, citations and recommendations, and execution accountable to closed-won revenue rather than decimal changes.
According to the article, a 37.4% score, up 2.1 points, reflects the vendor’s own formula rather than a defined market share. One method divides brand mentions across every prompt run, another excludes prompts where no brand was mentioned, another plots a subjective two-dimensional index, and another deducts points for competitive context. Put the same domain into three tools on the same morning and you can get three conflicting answers to a tenth of a percent.
Read more: Your 37.4% AI visibility score measures what, exactly?
The article describes Quality Score as a keyword-level, 1-to-10 diagnostic comparing ad quality with other advertisers, not an input into the auction or a measure of business profit. It cites agency decks celebrating a rise from 6.2 to 7.1 while closed-won pipeline did not move, and a case where cost per acquisition could climb 40% while the deck still showed 8-out-of-10. An AI visibility score offers the same refuge, but with a prompt list and denominator that can change by vendor.
Read more: I watched Quality Score become a client presentation
The article argues a tracker may run 25 to 50 fixed prompts once a week and present the result as market standing, but 50 synthetic queries on Tuesday only show how engines answered those sentences on Tuesday. Generative engines have no fixed inventory comparable to reported auction impressions for a targeted keyword list. If a competitor appears on 18 prompts one week and 19 the next, a chart can turn that one extra appearance into a 2.0-point surge.
Read more: A few dozen prompts cannot stand in for a market
The article cites an experiment that produced 80 distinct completions from 1,000 runs at temperature 0, and a 90-day benchmark where identical buyer prompts kept the same top-three recommended brand set only 46% of the time run to run. It also cites cross-model research finding only 11% of tracked brands appeared across all three major models, while 77% appeared in only one. A blended 37% therefore conceals where the brand did and did not appear.
Read more: The answer changes when you ask again
The article distinguishes three different events that a single win can collapse together. A mention is the brand appearing as item eight in a list of twelve vendors without a link, context or endorsement, a citation is the engine pointing to the site as a source with a clickable link, and a recommendation presents the business as the preferred choice for the buyer’s constraints. An unlinked name-drop and a linked recommendation should not count as the same outcome.
Read more: Count mentions, citations, and recommendations separately
The article’s position is that software that automates reporting automates the wrong layer. A faster chart of answers sampled once does not make those answers more reliable and does not execute the fix. It recommends repeated checks on buyer prompts, a record of what changed, and execution accountable to pipeline rather than cosmetic percentages.
Read more: Three questions before you pay for the decimal
[1] https://groas.com/post/your-ai-visibility-score-is-quality-scor (the article this page answers from)