AI visibility: how the scan works

What we measure

AI visibility measures whether AI assistants mention your brand when someone asks a question in your category. We put those questions to several assistants and count how often you are mentioned.

Two kinds of question, on purpose

We ask two kinds of question:

• Brand questions name your business ("What does {brand} offer?"). Useful to see what the AI knows about you, but they flatter: by naming you, you force the AI to mention you.
• Neutral questions only name your category, never your brand ("Who are the best-known providers in this category?"). This is the honest baseline: if you are mentioned here, it happens on your own merit.

That is why we calculate your share of voice from the neutral questions. It is your share of the names the AI comes up with by itself.

Why we ask each question twelve times

One answer is not a measurement. AI assistants do not give the same answer every time, so we ask each question twelve times and look at the pattern.

That number is not arbitrary. Research on repeated measurement shows a margin of error of about 27 percentage points at five repetitions, about twelve at twelve, and little gain after that. Twelve is where the curve flattens.

That is also why a measurement takes a while. You get that time back as a result you can base a decision on.

A percentage with a margin

Every measurement in your dashboard shows how certain we are. That label is not about your brand but about our measurement: how many observations sit underneath it, and how wide the margin around it is.

• High certainty means the result lies within a narrow band.
• Medium or low means the real number may be a fair way off the one shown.

Seeing "medium" more often than you used to? That is because we have been stricter since late September. The result has not got worse; the label has got more honest.

And just as important the other way round: "not seen" does not mean "never mentioned". With few observations, a brand you did not see may in reality be mentioned quite regularly. That is why we show the margin and not just the number.

The free scan is an indication, not a measurement

The free scan on our site deliberately does far less work: three questions, twice per AI source. That is enough for a first impression and it costs you nothing.

For such a scan we do not attach a certainty label, but state what the result rests on: how many checks were done, and that it is a first indication. That is the same truth, in words you cannot mistake for a verdict on your brand.

There used to be a label, and with so few observations it almost always said "low certainty". That was understandably read as "VEO does not know" or "my brand is doing badly", when it was about the size of our own sample.

The measurement in your dashboard asks each question twelve times instead of two, and does carry a certainty label.

Which assistants we ask

We ask Perplexity, Google Gemini and ChatGPT, each with live web search switched on. If a source gives no usable answer, we leave it out rather than pretend.

If a source drops out completely for a longer period, we are now notified automatically. Measurements from before and after such an outage are not directly comparable: fewer sources were behind them.

How often we measure

On Growth and Agency we measure automatically every week. On Free and on a beta account there is no automatic measurement; you start one yourself.

You can also start a measurement yourself. How many per month is shown on your plan page.

Competitors

The competitors you see are not made up by us. They come from the answers themselves: if the AI mentions other brands next to or instead of you, we count them. A name that comes up only once we drop as noise.

Why AI mentions your competitor and not you

There is usually an identifiable reason. Read Why does AI mention my competitor and not me? on our site.