Visibility score

Visibility answers the first question in AI search: are you in the room?

Definition

Visibility(entity, day, engine) = answers mentioning the entity / total answers

Where total answers is every answer captured for your active visibility prompts on that engine that day — whether or not any tracked entity was named.

Presence is counted at answer level. An answer naming you five times counts once, so a single enthusiastic answer cannot distort the number.

Measured on unbranded prompts only

Visibility is computed over visibility-intent prompts, which by rule cannot contain your brand name.

This matters more than any other detail on this page. A prompt naming you guarantees an answer naming you; including branded questions in a visibility figure measures your own name back at yourself and inflates the number for free. Perception prompts — where naming you is the point — are measured separately and never enter this metric.

If you have seen a competitor's tool report implausibly high visibility, a branded prompt set is usually the reason.

Why the denominator includes answers with no brands

Because those are real losses. When an assistant answers a category question without naming any tracked entity — recommending an approach, an in-house method, or brands you do not track — you lost that answer as surely as if a competitor had won it.

This is the difference between Visibility and Share of Voice:

• Visibility — of all answers, how many name you? Am I in the room?
• Share of Voice — of answers naming any tracked entity, how many name you? When I am in the room, how much do I own?

Visibility rising while Share of Voice falls means the category is getting more coverage and you are capturing less of the new attention.

The confidence interval

Every visibility figure ships with a 95% interval, and the interface quotes the change noise floor next to the delta.

Read them together. "46%, ±2.1 points, 3 runs per prompt" is a claim you can take to a review. The same 46% over a handful of answers is not, and the interval is what tells you which one you are looking at.

Interval width narrows as evidence accumulates: more prompts on the same subject, more days in the window, more runs per prompt. It is why the 30-day window is the default and the 7-day view carries visible caveats.

Brand Rank and Product Rank

Your position among tracked entities ordered by visibility. Rank 1 is best, and the label follows the Overview's scope toggle — brands under brand scope, products under product scope.

Rank is a blunt instrument by design: it is the number a stakeholder understands instantly. It also moves in jumps, since a fraction of a point can cost two ranks in a crowded set. Use rank for reporting and visibility for analysis.

Reading it

By engine. Visibility differs sharply across ChatGPT, Google AI Overviews, Google AI Mode and Perplexity, because each weighs sources differently. A blended number hides your worst engine, which is usually the one worth fixing.

By group. Brand visibility and product visibility are different fights. See Products.

Over time. Day-to-day noise is expected. Look at a fortnight, and check significance.

Benchmarks

There is no universal "good" number. It depends on category concentration, how many brands engines typically name, and your prompt mix. The only benchmarks that mean anything are your own trend and your competitors measured on the same prompt set — which is exactly what the competitor table gives you, since the answer set is identical for every entity in it.

Treat any cross-industry visibility benchmark with suspicion; it was computed on a different prompt set from yours.

When it moves and you did nothing

• An engine changed its model routing or retrieval behavior
• A source that carried you was updated, deprecated or outranked
• A competitor published something that is now being cited
• Seasonal shifts in what gets published in your category

Before hunting internally, check Sources for a change in the domains the engines are reading. Most unexplained visibility movement is a source-selection change.