What we counted
We took 60 answers and counted every brand mention twice: once by where it appeared in the sentences, and once by where it appeared in the list of sources the model cited.
Between 35% and 48% of all brand mentions came from the citation list alone - the brand appeared as a URL the model had read, not as a name the model had said.
In Gemini, links accounted for 62% of all characters in the answer.
Why the distinction is not academic
Most AI visibility tools count both as a mention. That produces a number that can rise while the thing you care about does not move at all: your page is being read, and a competitor is being recommended in the sentence above the link to it.
It also explains a result that otherwise looks like noise - a brand whose visibility climbs steadily while nothing changes in the pipeline.
What it changes in how you write
Being cited is a retrieval problem: the page has to be reachable, parseable and specific enough to be worth pulling. Being recommended is an argument problem: somewhere, in language a model can lift, it has to say why you are the right choice for a described situation.
The same page can do both, but not by accident. A thorough guide earns citations easily and recommendations rarely, because it explains a topic without ever taking a position on who should be chosen.
How we report it
Citations and named mentions are separate lines in our reporting, never a single 'visibility' figure. When the two diverge - and they often do - the gap itself is the finding, and it usually points at content that informs without ever recommending.