One brand, one set of numbers. Each figure in the product is produced by a single shared calculation, so the AI visibility on your Overview is the same figure on your AI platforms page and in Brand Performance, and a report you export keeps the figure of the day it ran. This page is the rule book.
One percentage: of every AI answer captured in the window, the share that named your brand. It is always printed with both counts, for example 34 of 126 answers, so anyone can check it.
The figure divides the answers that named you by the answers read. One prompt asked on four AI platforms is four answers, and an answer counts once however many times it names you.
Only the AI platforms your plan reads are in it: 4 AI platforms on Spotlight and Brand, 5 on every agency plan. The count of platforms that named you is shown beside it as a count, never folded into the figure.
Before any answer is read the figure says Not measured yet. Once answers are read, a brand none of them named is shown as 0% with its counts, because that is the reading you most need to see.
Bands: 60% and above is Leading, 30% to 59% is Competing, 10% to 29% is Starting to show up, below 10% is Not showing up. The band word waits until the reading is thick enough to call; the percentage and its counts do not.
A mention is one AI answer that names your brand, counted once per answer per assistant. A citation is a unique page on your own domain that an answer linked to as its source. Both are counted over the last 30 days unless a panel states a different window, and both are scoped to the country pill you have selected.
Search volume, AI search volume, traffic and opportunity sizes are not read from your AI answers. They are estimates, modelled from search, link or mentions data, and every one of them carries an Estimate label, the kind of data it rests on, the market and the date of the reading. They are rounded to what the method can carry, and we print no plus or minus range, because none of these methods produces one.
Monthly Google searches, averaged over 12 months, for the market shown.
Estimate, search data. Printed to two significant figures, so 34,874 reads as 35,000.
How often people ask AI about this topic in one month, in the market shown. One figure for AI search as a whole, not split by platform. A long prompt may carry its core phrase's demand.
Estimate, search data. Printed to two significant figures, so 34,874 reads as 35,000.
How many people search each month for the need behind a prompt, in its market: Google searches for the short phrase shown, or AI searches where Google has no figure. The bars compare a prompt with the busiest one: green and bold is high demand, amber medium, red low.
Estimate, search data. Printed to two significant figures, so 34,874 reads as 35,000.
Search volume times the share of clicks usually taken at that position. Search features and AI Overviews take clicks this does not see.
Estimate, search data. Printed to two significant figures, so 34,874 reads as 35,000.
Visits from Google modelled from every ranking on record for the domain. For comparing domains, not your analytics.
Estimate, search data. Printed to two significant figures, so 34,874 reads as 35,000.
Summed over the keywords SKYA sampled from stored search results for this domain: each keyword's volume times a click share for its position, or the traffic figure on record for that keyword where there is one. It covers only those keywords, not the whole domain.
Estimate, SKYA model. Printed to two significant figures, so 34,874 reads as 35,000.
Search volume times the click rate at the best position a new page can realistically reach, discounted for difficulty. A size, not a forecast.
Estimate, SKYA model. Printed to one significant figure and written "around", so 380 reads as around 400.
Average cost per click advertisers pay in this market, in the market's currency.
Estimate, search data. Printed to two decimals under 10, otherwise to two significant figures.
A 0 to 100 score of how hard the top ten results are to enter.
Estimate, search data. Printed as a whole number.
A 0 to 1000 rank of the domain's link graph: how many sites link to it and how strong they are. It moves slowly; a jump of 50 in a month is large.
Estimate, link data. Printed as a whole number.
Headline mentions and citations are added up from the same per-assistant rows shown underneath. If we hold a wider reading that those rows do not account for, we say the breakdown is incomplete instead of quietly inflating the headline.
Ratios such as the share of answers that link to you are only printed when the numerator and denominator come from the same window and the same assistant set. When they do not, we withhold the figure rather than print an impossible one.
With only your brand in the tracked set there is no share to report, so we show "not measured" and prompt you to add competitors. We never report 100% share or a rank of number one out of one.
A metric sitting at zero is shown as flat, whatever the underlying delta says. Movement is only claimed when there is something to have moved.
Medians and averages drawn from fewer than eight dated observations are marked as a sample. You still see the number, but never framed as a settled finding.
Until answers have been read, AI visibility reads "Not measured yet" rather than zero. A percentage built on a handful of answers swings wildly week to week, which is why the band word waits for a thick enough reading.
Where a leading domain is named, it is checked against your own domain first. If they match, the comparison is dropped.
At least 14 days of readings across three or more assistants.
At least five days of readings, or three assistants with a solid volume of answers.
Early readings. Treat every number as a first sketch.
Tell us the project and the screen and we will trace it back to the readings that produced it. Every figure carries its source, its window and its sample size.