AI visibility is how often an assistant names your brand when someone asks it a buying question, and which sources it reads before deciding. It is not a ranking, it is not traffic, and it does not follow from either.
The two came apart, and there is data on how far.
Ahrefs looked at 15,000 queries and found that roughly 12% of the URLs cited in AI answers also appear in Google's top ten. Across engines the overlap is no better: only about 11% of cited domains are cited by both ChatGPT and Perplexity. An answer engine is not reading the results page. It is assembling from its own set of sources.
I measure one content asset that shows the same split at close range. It has produced 461 citations across 82 distinct pages in AI platforms. Over the same period it held four ranking keywords in Google and about three organic visits a month, on a domain with an authority score of nine. Almost no rankings, no link building, and citations regardless.
That is the practical point. A site can be invisible in search and present in answers, or the reverse, and neither result predicts the other.
Most audits check four: ChatGPT, Perplexity, Gemini and Google AI Overviews.
On the asset above, the distribution was ChatGPT 211 citations, Copilot 159, Google AI Mode 39, Perplexity 23, Gemini 19, AI Overviews 9. Copilot was the second largest source and it is the one routinely left out, partly because it gets written about least. If your audit skips it, you are missing roughly a third of your citations.
Distribution varies by category. The lesson is not that Copilot always ranks second, it is that assuming the four familiar names covers the field is how you end up optimising for the wrong reader.
Mostly not from you.
Of the thousand most-cited pages in ChatGPT answers, around 67% sit on domains no brand can realistically influence: Wikipedia, government and educational sites, and large news outlets. Reddit alone accounts for 46.7% of citations on Perplexity and about 21% on Google AI Overviews.
This is why AI visibility work is rarely a matter of rewriting your own pages. Your site is one input among many, and usually not the decisive one. The decisive question is which third-party sources your category's answers are built from, and whether you exist inside them.
A measurement worth the name has four parts.
A prompt basket. Twenty-five to thirty questions people actually ask before buying in your category, not the phrases you would like to rank for. Buying questions look like "which X should I choose for Y", not like keywords.
Multiple engines. Five, including Copilot. Each is run on the same basket so the results are comparable.
Competitors on the same basket. Your own numbers mean nothing without three or four others beside them. Absence is only interesting relative to somebody’s presence.
The cited sources. For every answer, the domains the engine read. This is the part most tools show last and it is the part that tells you what to do.
In May 2026 AMEC, the industry body for communications measurement, published a set of principles on AI visibility with a blunt warning: a visibility score reported without a link to an outcome is the new AVE — the advertising value equivalence metric the industry spent two decades getting rid of.
The warning is fair. A number that goes up while nothing else changes is decoration. A measurement is useful when it names which sources to enter, in what order, and what it should cost.
In rough order of effect for a small or mid-sized brand:
Presence in the third-party sources your category's answers are assembled from. Content structured so a machine can lift a determinate answer out of it, on a page a crawler can resolve without executing JavaScript. Consistent naming and description across the places that describe you. Freshness, because several engines visibly prefer recent material.
Link building barely appears on that list. The asset described above has an authority score of nine.
Measure before changing anything. Most categories turn out to have a settled answer that has been repeating for months, built from four or five sources, and until you know which ones, every fix is a guess.
A fixed-scope measurement of exactly what this page describes: a basket of real buying questions, five engines including Copilot, three competitors beside you, and the sources your category’s answers are built from.