Docs › Concepts
How your visibility is scored
Your AI Visibility score is eight separate measurements, blended. Here is what each one asks, and why a citation you earned is worth more than one you were handed.
In 10 seconds
- Your score is eight separate measurements, each asking a different kind of question, blended into one headline.
- Two of those measurements are about being known; five are about being suggested. They are weighted 30/70 in favour of being suggested.
- A mention only counts as evidence if the question did not already say your name. Most brands' raw numbers are flattered by their own echo.
The short version
A single "visibility percentage" is a tempting number and a misleading one. "Do you know Yaseminkahve?" and "What is the best Turkish coffee brand?" are not the same question, they do not have the same kind of answer, and a system that scores both by looking for your name will tell you the first one is a win.
So EOMA scores eight kinds of question separately, using the instrument each one deserves.
Three of them name you. Recognition ("what do you know about Brand X?"), market position ("how does Brand X compare to its competitors?"), and sentiment ("what do people say about Brand X?"). These measure whether the engine knows you, and they are scored on an absolute basis — the engine either has real information about you or it does not.
Five of them do not name you. Share of voice, share of voice in a niche, product recommendation, buying guide, and category discovery. These are the questions a buyer actually asks before they have heard of you, and they measure whether the engine suggests you. They are scored against the best-performing brand in your market rather than against a perfect 100.
The headline blends the two: 30% knows-you, 70% suggests-you. The weighting is deliberate. Being known is table stakes; being recommended to a stranger is the thing you are buying. EOMA never collapses the pair into one figure on screen, because the gap between them is usually the whole story. A brand that scores 84 on knows-you and 0 on suggests-you is famous and absent — and that diagnosis is invisible if you only read the blended number.
How it works
Each question is answered by the instrument that fits it. A question that says your name cannot be scored on rank, because rank is meaningless when the answer was handed the subject. A question asking for a shortlist is scored on share of the list, not raw position — being third of three is the bottom of the shelf, third of twenty is the top. A comparison where the engine named no competitor at all is excluded rather than counted as a win.
"I have no information about that brand" scores zero. The engine disclaiming you is not a mention. This sounds obvious and it is the single most common way visibility numbers get inflated across this whole category of tool.
Being absent scores zero. Not being measured yet scores nothing at all. Zero means "we asked, and you were not there". Blank means "we have no evidence". Sentiment in particular reads not measured rather than a neutral score for most brands, because reading tone from an answer is a separate, paid analysis that has only been run for some brands. A neutral face would claim people feel nothing about you, which nobody measured.
The five suggests-you axes are scored against your market's real ceiling. Nobody reaches 100. The best-performing brands EOMA has measured top out around 47% on share of voice; most markets peak in the low teens. Scoring against an absolute 100 would tell the category leader it was failing. So each of those axes is expressed as a fraction of the best rival's score on the same instrument, measured on the same set of questions.
The score covers 28 days, not one run. It pools every answer from the last four weeks — typically several runs and well over a hundred answers. If a brand has had no runs in 28 days, the score reads not measured. The window never stretches to reach older data, so you never see a stale number wearing a current date.
What this means for you
Your citation count is probably flattered by your own echo, and the fix changes the number a lot. When a question says "tell me about villa sepeti", any answer citing villasepeti.com is the engine quoting you back to yourself. That is not evidence anyone found you. EOMA separates the two, and ranks the sources board on earned citations only — the ones from questions that never mentioned you.
Measured across every brand with a website: villa sepeti showed 76 citations, 71 of them handed over, 5 earned. Yaseminkahve: 31 shown, 1 earned. Six of the ten brands measured had zero earned citations. If you are reading a raw citation count anywhere else, assume most of it is echo until you check.
A brand's own row survives at zero earned, on purpose. "Cited 30 times, never unprompted" is a finding worth acting on. Dropping the row would render it as the flatly different "not cited at all".
Low suggests-you with high knows-you is a content problem, not a brand problem. The engines know who you are; they are not reaching for you when somebody asks the buying question. That is the gap your weekly actions are aimed at.
The score moves slowly and that is correct. Publishing something today does not change what an engine says tomorrow. Judge the direction over weeks.
Common misunderstandings
| People assume | Actually |
|---|---|
| "One percentage describes my AI visibility." | It cannot. Eight categories are measured on four different instruments, and one figure that averages them hides which half you are failing. The knows-you / suggests-you pair is the smallest honest summary. |
| "100 is the target." | It is unreachable on the five suggests-you axes and always has been. The measured market leaders sit between 10% and 47%. Your number is a fraction of what the best brand in your market achieves. |
| "A citation is a citation." | Only if the question did not hand it to you. An answer to "tell me about my brand" citing your own site proves nothing about discovery. |
| "Sentiment is blank because nobody has opinions about us." | Blank means the tone analysis has not been run for your brand. It is a separate paid step, and EOMA leaves it empty rather than substituting a mention rate wearing the word "sentiment". |
| "My score dropped, so something got worse." | Check whether the window moved. The score pools 28 days; a run falling out of the window changes what is being averaged. Read several weeks, not two points. |
Related
- What a run is — where the answers being scored come from
- Competitors and share of voice — who sets the ceiling your score is measured against
- AI provider visibility — the screen where this score lives