What an AI visibility audit actually finds
Vorne Team ·
An AI visibility audit finds two things: the questions where assistants already name you, and the questions where they name somebody else. For Bloomistry, shared with their permission, 12 of 12 branded prompts named the brand and 0 of 10 discovery prompts did. The blended figure that hid that gap was the least useful number in the report.
The gap a blended score hides
We ran two prompt sets against the same store. Branded prompts named Bloomistry and asked about it directly. Discovery prompts asked the question a buyer actually types, with no brand in it.
The branded set came back 12 out of 12: asked about Bloomistry, the assistants described Bloomistry. That is a reading-comprehension result, not a visibility result. The discovery set came back 0 out of 10. Averaged across all 22 observations those produce a mid-range score that reads like a site doing moderately well, when in fact it was invisible at exactly the moment a buyer forms a shortlist.
Sample sizes are small, and we quote them for that reason. Twenty-two observations is a signal worth acting on, not a precise measurement, and a report that gave you one decimal place off 22 runs would be lying about its own precision.
This is the single most common way an AI visibility number misleads. If you take one thing from this piece, take this: never average branded and unbranded prompts into one figure.
What the content work produced
We drafted five articles against the questions the discovery set had exposed, each built only from facts the merchant had already approved. Scored with the same open check we run on customer pages, they averaged 91 out of 100, ranging from 298 to 458 words.
The check also flagged all five for the same thing: no attribution. For a florist describing its own delivery windows that finding is arguably fair to ignore, since the merchant is the primary source for its own opening hours. We are reporting it rather than suppressing it, because a scorer you only listen to when it agrees with you is not a scorer.
What we are not claiming
The articles are drafted and staged. They are not published, because publishing is the merchant's decision and the approval gate is the point of the product.
So there is no traffic result here, no ranking movement, and no revenue figure. A case study at this stage would have to invent all three.
- What is measured: the branded and discovery baselines, and the draft scores.
- What is staged: five articles awaiting a publish decision.
- What does not exist yet: any performance outcome, because nothing has been published.
Why publish this at all
Because the diagnosis is the part most teams skip, and it is the part that changes what you do next. A store that is invisible on discovery questions does not need more content about itself; it needs content answering the questions buyers actually ask.
The audit that produces those two numbers takes minutes and costs nothing: run a free AI visibility audit on your own site. What you do afterwards is the work.