A visibility score is a choice, not a measurement

Here is a way to raise your AI visibility score by twenty points this afternoon, without changing a word on your website.

Add branded prompts to your tracked set. “What does Acme do?” “Is Acme any good?” “Acme pricing.” Your brand appears in nearly all of those answers, because your brand is in the question. Your visibility percentage climbs. Nothing about how AI treats you has changed.

Nobody in this category writes about this, for the obvious reason that it makes the headline number look soft. It is worth writing about anyway, because if you are about to spend money on one of these tools, it is the first thing you should understand.

What a visibility score actually is

Strip away the branding and the metric is simple: of the prompts we track, in what share does your brand appear?

Both halves are chosen. Someone picked the prompts. Someone defined “appear.” The result is a ratio whose denominator is an editorial decision.

The evidence that prompt choice moves the number

This is not just a theoretical worry. Three measured findings show how much the question set determines the result.

Funnel stage changes engine behaviour. Peec AI analysed roughly 500,000 prompts and found AI Overviews appeared in 86% of them overall, rising to 88.5% for decision-stage prompts. They also note the dataset skews toward commercial buying intent and excludes navigational queries.¹ That disclosure is the honest part, and it makes the point: the sample determines the finding.

Visibility is topic-level, not brand-level. Semrush tracked more than 50,000 brands across 1,094 US categories from January to June 2026 and found that only 21% of the most-cited domains in a category were also the most-mentioned brand, with a slightly negative correlation of −0.229 between the two.² Strength in one topic does not transfer. So a prompt set weighted toward your strong topics produces a strong score, honestly measured and thoroughly misleading.

Activation varies by category. The Washington University measurement study collected 55,393 trending queries across 19 categories and found AI Overview activation differing across them.³ Which categories your questions fall into changes how often there is even an AI answer to appear in.

Four ways a prompt set quietly flatters you

1. Branded prompts Your name is in the question, so you are in the answer. These are worth tracking — reputation and accuracy questions matter — but they belong in their own bucket, never blended into a headline visibility number.
2. Topics you already own Because visibility is topic-level, a set weighted toward your strengths reports strength. The prompts that matter commercially are usually the ones you are currently losing.
3. Phrasings that suit you "Best enterprise-grade tool with SOC 2" and "cheapest tool for a small team" are the same buying decision with different qualifiers, and they will return different brands. Whoever writes the qualifier picks the winner.
4. Silent revisions The most common version of this is not deliberate. Someone prunes prompts that "aren't relevant any more," the number jumps, and the jump gets reported as progress. If the prompt set changed, the series broke.

Keeping it honest

None of this makes the metric useless. Every measurement discipline has a sampling frame, and choosing one is not cheating. What matters is that the frame is deliberate, documented and stable.

1. Write the prompt set down, with reasons Each prompt should map to a decision a buyer actually makes. If nobody can say why a prompt is in the set, it is padding.
2. Segment before you blend Branded and unbranded reported separately. Funnel stage separately. Otherwise a mix shift reads as a performance change.
3. Freeze the core set Keep a stable core you never edit, so you have one continuous series. Run experimental prompts alongside it, not inside it.
4. Version it like code Date every change and annotate the chart where it happened. When the number jumps, the first question should be answerable immediately: did we change, or did the list?
5. Include the questions you lose A prompt set with no losses in it is not a measurement instrument. The uncomfortable prompts are the ones with information in them.

Questions we get asked

The short of it

Ask any vendor — us included — to show you the prompt list before the dashboard. Ask how branded prompts are handled. Ask what happens to the series when the list is edited.

A tool that answers those three questions well is measuring something. One that changes the subject is selling you a number.