What AI search visibility is, and how you measure it

What is AI search visibility?

AI search visibility is whether a company is named, and how it is described, when someone asks a question and receives a composed answer rather than a list of links. The surfaces this covers are not one product: they include the AI features built into search results, and the standalone assistants people now open directly to ask what they would once have typed into a search box. What they have in common is the output. A search engine returns a ranked list and leaves the choosing to the reader; an assistant returns a paragraph that has already done some of the choosing.

The difference between a list and an answer is the whole reason a separate measurement exists. A ranking describes where a page sits in a list of options. An answer describes who is worth considering, in prose, often naming a handful of companies and omitting everyone else. You can rank respectably and be absent from the answer built on the same subject, because the two are produced by different processes reading different things.

Why does rank tracking not measure it?

Rank tracking answers one question — where does this page sit for this keyword — and answers it well. It cannot tell you whether you were named in a composed answer, because there is no position to record: either the answer mentioned you or it did not, and if it did, the words it used to describe you matter as much as the mention itself. A tool built to record a number cannot record a description.

There is a second reason, and it is the one most people meet first. Answers are not stable in the way a ranking is. The same question asked twice can produce different wording, a different set of names, and a different set of sources, because the systems composing them are probabilistic and are updated continuously by their providers. A single check therefore tells you almost nothing on its own. What tells you something is the same question set asked repeatedly, over weeks, with every result kept — which is a measurement design problem before it is a tooling one.

What is actually worth measuring?

Eight things, and only the first two have equivalents in ordinary search reporting. Whether you appear at all. Where you appear, when the surface has an order. How you are described, which is the one that decides whether a mention helps you. How often you are named relative to named competitors. Whether the description reads as positive, neutral or negative. Which sources the answer was built from. Who is being named instead of you. And which questions have no answer anywhere you own — the gap that explains most of the others.

The source list is usually the most useful line, because it is the only one that converts directly into work. A share-of-voice number tells you that you are losing; a citation list tells you what you are losing to. If the answers on a subject are consistently built from one publication, a directory and a competitor's own page, then those are three specific things to address, and two of them are not on your website.

Can you improve it, or is it out of your hands?

Parts of it are genuinely out of your hands. No one controls what a model outputs, no one should promise a placement, and any account of this work that offers a guaranteed position in an answer is describing something it cannot deliver. What is in your hands is the material the answer is assembled from: whether your own pages state plainly what you do and who you do it for, whether the entity behind your name is unambiguous, and whether the sources these systems lean on have anything accurate to say about you.

Measurement therefore comes first in any honest version of this work. You find out what is currently said, and which sources it comes from, before deciding what to build — because the alternative is producing content on a hunch and having no way to tell whether it changed anything.

Where does the detail live?

Three pieces go further than this one. On what a first measurement actually turns up, including our own: we ran the method on ourselves the day this site launched and published the result. On why a company can be well known and still absent from the answers. And on why a search engine and an assistant frequently disagree about the same company, which is less mysterious than it looks once you know what each is reading.