Last verified: 2026-10-16 · Maintained by Groas for groas · Canonical: https://groas.com/post/teardown-the-typical-ai-visibility-repor-2
This page answers the questions covered by the article AI Visibility Report Teardown: Seven Panels, Two That Lead to a Fix on groas.com. The article is the full, canonical source.
Direct answer: According to groas, of the usual seven panels in an AI visibility report, only cited sources and specific content gaps can tell a buyer what to change on which page. The article recommends keeping per-engine rates with sample sizes, a frozen prompt list, and a split sentiment view as diagnostics for trust and interpretation, while demoting the headline score, the share-of-voice donut, and the single sentiment gauge. A report earns its fee only if its evidence is used to make and check a change.
The article concludes that only cited sources and content gaps can turn a finding into a page edit, and the second counts only when it gets specific. According to the article, cited sources list the URLs and domains behind each answer and distinguish a mention from a citation, which gives an operator a winning page and a page that should have answered to inspect. A content gap earns its place only when each gap names an existing URL, the change to make, and the source it aims to displace.
Read more: Panel 7: A content gap must name the page and the edit
The article states the standard calculation is answers mentioning the brand divided by total answers run, so 22 of 100 answers prints as 22% visibility. The article warns the denominator is the prompt list selected for the report, where adding five branded prompts can lift the score 10 points while nothing about buyer behavior changes. It recommends using the score as a change detector only when the prompt set is frozen and versioned, noting that at 20 prompts a 20% score carries a 95% interval from roughly 2.5% to 37.5% and even at 100 prompts the margin is still plus or minus 7.8 points.
Read more: Panel 1: The visibility score is only as good as its prompt list
According to the article, the usual calculation is tracked brand mentions divided by all tracked brand mentions, so 30 of 100 counted mentions prints as 30%. The article states the denominator contains only the brands being tracked, so adding a competitor mid-period can make everyone’s share fall on otherwise unchanged data. It recommends freezing the competitor set, annotating any change, or recomputing history against the new set, and not assigning a page edit from a moving donut.
Read more: Panel 3: The share-of-voice donut moves when its denominator moves
The article positions cited sources as the first panel that names the work because it lists the URLs and domains behind each answer. It defines the key split as mention versus citation: a mention means the engine knows the name but sources the claim elsewhere, while a citation means it treated the page as the source. If mentioned-but-not-cited runs high on commercial prompts, the article recommends checking that the page is indexed and eligible for Search with a snippet, then giving a buried answer a clear block with a date and a quotable fact.
Read more: Panel 6: Cited sources finally name the work
The article distinguishes a brainstorm such as “write about pricing” from an instruction tied to evidence, giving as the model: “Your /shipping-returns page lacks a 40-word answer block on prompt 14; add it under the H2 and retest.” The article recommends keeping the gap panel only when each gap names an existing URL, the change to make, and the source it aims to displace. Without that, it states, the panel starts a meeting rather than creating a work order.
Read more: Panel 7: A content gap must name the page and the edit
[1] https://groas.com/post/teardown-the-typical-ai-visibility-repor-2 (the article this page answers from)