What an AI visibility tool actually measures

Open any AI visibility dashboard and you will meet a wall of numbers: visibility, share of voice, position, sentiment, citations. They look like five different measurements. Most of them come from one source, the saved answers, read in different ways.

Every one of these tools does the same basic job. It sends a set of questions (the industry calls them prompts) to AI assistants such as ChatGPT, Gemini or Perplexity, saves the answers, and then reads each answer looking for your brand. Everything on the dashboard comes from what it finds in those saved answers.

So the most useful thing you can do before comparing tools is learn to see the answer underneath the number.

The anatomy of one answer

Here is what a tool is looking at when it scores a single response. The question and brands are made up; the parts are the ones every tool measures.

Count mentions across many answers and you get visibility. Compare your mentions with rivals’ and you get share of voice. Average the order and you get position. Classify the descriptions and you get sentiment. Count the pages and you get citations. Each tool then wraps these in its own names and formulas.

Same idea, different sums

The words are shared across the category. The arithmetic is not. Here is how five tools define “visibility” or its closest equivalent in their own help pages:

Notice the units. Some tools count every answer, some count each prompt on each platform per day, and a weighted measure like Ahrefs’ impressions deliberately gives popular questions more say. None of these is wrong. They answer different questions, which is why a “35%” in one tool and a “35%” in another are not the same thing.

Five questions to ask about any number

1. Which answers?Which platforms, how many runs, which country and language.
2. Counted how?Per prompt, per response, or weighted.
3. Against whom?Share of voice depends entirely on the rival list you set.
4. How many samples?Small samples swing; good tools show the count.
5. Can I see the answer?The saved text is the evidence behind the score.

The last question matters most. Scrunch’s guide, for example, shows the sample behind a rate, such as “43% (37 of 86 responses)”. That kind of detail turns a number you have to trust into one you can check.

What none of these numbers tell you

An AI visibility score measures what an assistant said in the answers a tool collected. It does not tell you how many people asked that question, whether anyone clicked, or whether the answer changed a purchase. Some tools add search volume or website traffic data to get closer to that, and they label it. Treat everything else as a well-organised sample of AI answers. It is still very useful, as long as you read it that way.

Our guide to reading AI visibility results covers the evidence behind the numbers. If you want to go one level deeper, our pieces on citations, share of voice and reading week-to-week change take each measure apart in turn.