How Is AI Visibility Measured?

Starting With The Right Questions

Questions that contain your brand, such as “Is Firm X any good?”, do not measure visibility; anyone who asks that already knows you. The right question is one a customer who does not know you would ask. It names a product or service, a feature, a place or an expectation.

The list depends on the business. For an online shop, product and comparison questions come first; for a hairdresser, questions about place and reviews; for a software company, “which one suits us better” questions. The list has to be set once and kept fixed, or months cannot be compared.

Several Assistants

Google’s AI, ChatGPT and Claude can answer the same question with different brands. Each leans on different sources and summarises differently. So the result has to be kept separately for each assistant.

It is hard to know which assistant your customers use. Measuring one and assuming the others means seeing half the picture.

Several Tries

The same assistant can also answer differently from day to day. Even a small difference in wording can change the list. Deciding from a single try is like tossing a coin.

So each question is put to each assistant several times, and the result is calculated from the sum of the repeats. That separates a name that turns up now and then from one that turns up regularly.

Which Numbers Matter?

The visibility rate is the share of all answers that name you. Position is where your name sits in the list. Competitor share is who is recommended in your place, and in how many answers. Sources are the pages the assistant relied on in its answer.

Read together, these four show what a single score would hide: on which question, in which assistant and against whom you are behind. A single “AI score” melts those details into an average.

Recording The Sources

When an assistant shows sources in its answer, they are among the most valuable parts of the measurement. They show which pages are cited regularly and through which sources your competitors come out ahead.

Sources vary from assistant to assistant, and some show none at all. In that case only the answer itself is recorded. Stating that difference plainly in the report keeps the reader from drawing the wrong conclusion.

Reading It Alongside Google

The AI measurement is separate from the Google measurement, but not independent of it. For the searches drawn from the same question list, your Google position and the clicks you get show the other half of the picture.

Put the two side by side and you usually meet one of three situations: you are doing well in both, well in one and weak in the other, or behind in both. Each calls for a different priority.

If You Want To Measure It Yourself

It is possible to do this measurement by hand. A ten-question list, three assistants and a few tries for each question add up to a few hours of work a month. The answers have to be recorded in a table and repeated in the same way every month.

The difficulty is not in the numbers but in the consistency. The wording of the questions must not change, the tries must be made under similar conditions, and the sources must be recorded too. After a few months you have a record you can compare.

Where keeping that record feels like too much, a report that does the same job regularly takes over. What matters is the method itself; who applies it comes second.

Reading Over Time

One month’s result says little on its own. The value is in repeating the same measurement every month. A fall on one question, a rise on another, a place lost in one assistant: these only make sense side by side.

A good report shows the bad months too. A report that highlights only good news keeps you from understanding what worked.

The point of measuring is not to get a grade but to decide the next step. Once you know on which question you are behind and who comes up in your place, where to start often becomes clear by itself. Numbers are worth something only when they turn into a decision.