What these views show
The layout is similar across tools of this kind, and Wix built one straight into its dashboard in September 2026. From the business name, the address and the site description, the tool generates a set of questions of the sort someone might type into a chat window. Those questions run against ChatGPT, Gemini and Perplexity. For each one you then see whether your own site appeared in the answer, in which position, and how many sources the answer was built from. Above that sits a percentage per provider, below it a comparison with competitors and a list of the most frequently cited sources.
That is a lot of information, and most of it is useful. The percentage at the top still deserves its own paragraph, because it gets read first and carries the least.
What happened in one afternoon
We ran the test on a website we look after, three times within the space of an hour. Between the runs nobody touched the site: no new text and no changed setting. The twenty questions stayed the same as well.
| Run | ChatGPT | mentioned | Gemini | Perplexity |
|---|---|---|---|---|
| First run | 45% | 9 of 20 | 55% | 60% |
| Third run | 30% | 6 of 20 | 55% | 60% |
On ChatGPT three of twenty questions flipped, a third of the figure. Gemini and Perplexity showed the same numbers in both views.
Everything below the percentage moves too. The same question about a local specialist sat at position 1 of 13 sources in the first run and position 1 of 12 in the third. Another question came back as a no both times, once out of 12 sources and once out of 19. The answer is composed afresh on every run, and the scoring follows it.
One question, three different answers
An example from the same test shows why a single provider says little on its own. Asked about a particular treatment near the town, ChatGPT named the business first, Perplexity also named it first, and Gemini left it out entirely. Gemini instead wrote into its answer that the town lies only seven and a half kilometres from the nearest larger town, and then listed four providers from there.
That is no malfunction. Gemini read the question more broadly and widened the radius. For the business the picture still stands: first with two providers, absent with the third. An average across three such systems hides exactly that difference, and in practice the difference is the most interesting thing in the whole test.
Why the number sits higher than it looks
Two counting rules push the figure up, and both are stated openly in the vendor's own documentation.
Being used as a source counts as a mention. If the site is absent from the answer itself but was drawn on as a source, the vendor still counts it. The help page puts it plainly: "Your site was not mentioned in the answer but was used as a source." In practice those are two very different things. A reader either sees the business named in the middle of the text, or sees a footnote under a paragraph recommending somebody else.
Brand questions go into the same pot. Some of the questions contain the business name, along the lines of "What treatments does [business] offer?". The tool labels them as brand related and counts them towards the same percentage. A question naming a business is almost always answered with that business. It measures whether the machine can look up a known address, rather than whether it suggests the business unprompted.
Taken together: the share of questions where a stranger genuinely sees the name is lower than the figure on display.
The question list belongs to the business being scored
In the same tool every question can be deleted, your own can be added, and one button regenerates the whole set. For the work itself that is right: anyone who knows their customers writes better questions than a generator.
For the number it means something else. Delete the four questions where you are missing and the next run looks considerably better, while visibility itself has stayed exactly where it was. A score only carries as far as the list it stands on.
When a number appears at all
According to the help page the view needs at least fifty unique visitors in the last thirty days. Below that it stays empty.
For many small businesses that is everyday life. A tradesperson with thirty visitors a month will never see a figure here, even though a mention in an AI answer would do them the most good. How many people already search this way is something we wrote up separately.
What the view is good for anyway
All of this is an argument for opening it. Three parts of it are good, and two of those are more useful than the number above them.
- Most of it hangs on the source list. Under "top sources by references" you can see where the answers draw their knowledge from. In our case that was a trade directory, the municipality's own page, a professional association and a Wikipedia article, so mostly other people's sites. It is the most concrete to-do list such a tool produces: having a clean entry there does more than another paragraph on your own home page.
- Almost as useful are the individual questions: which question was answered without the business appearing, and who stood there instead? That is a competitor analysis for free, readable question by question.
- And the trend carries once several runs sit side by side. A single value on a single day mostly tells you about that day.
How we measure
We ask the same question repeatedly and record one of three levels for every measurement: clearly named, mentioned only in passing, absent. Being used as a source without appearing in the text lands in the middle level.
The questions come from what people actually type, and they stay the same over time so that two measurements can be compared. Questions containing the business name run separately, because they answer something else.
Every measurement is filed with its date and the full answer text. When a level jumps we look at what changed in between, rather than taking the new number as the truth. How to reproduce this yourself in five minutes is in How to check whether AI already recommends you. What the models base their recommendation on is in GEO: how to get recommended in ChatGPT and Perplexity.
Frequently asked
Is a score like this worthless?
As a direction it works; as a number to decide on, much less so. It reliably tells you whether a business is practically absent from AI answers or turns up regularly. The gap between 45 and 30 per cent, on the other hand, should not be read into, because it appears without anyone doing anything.
Why does the figure move when nothing on the site changes?
Because the answer is generated afresh on every query. The model searches again and picks different sources as it goes. In our test the number of sources drawn on for a single question went from 12 to 19 between two runs on the same day.
Does it count if my site appears as a source under the answer?
In the vendor's score it does, according to their own help page. For a person reading the answer, though, a footnote is a different thing from seeing your own name in the text. That is why we keep the two apart.
When does the view show data at all?
From fifty unique visitors in the last thirty days. Smaller sites see nothing there, even when they already appear in AI answers.