Commercial

How we measure AI visibility

AI visibility is whether a model names your business when somebody asks it the question your buyers actually ask. It is measurable. It is not a ranking. And the method matters far more than the number, so the method is published here in full rather than described as proprietary.

What is being measured, precisely

A search engine returns a list. A model returns an answer. Those are different objects and they fail in different ways, so a rank tracker cannot tell you how you are doing inside an answer. What we measure is narrower and more useful: given a prompt a real buyer would type, does the answer name you, and where the surface shows its sources, is one of them yours.

We will measure where you show up on anybody’s website. We will only change a website we built.

Measuring where you stand is not the same as promising to move it, and keeping those two apart is why this method can be run honestly against anything.

The method, step by step

• Agree the prompt set with you, in writing. The prompts are the questions your buyers ask, phrased the way they ask them, not keyword strings. They are recorded and they do not change between runs, because a prompt set that drifts produces a trend line that means nothing.
• Run the same set across the answer surfaces that matter in your market. In practice that means the large general assistants and the AI answers that sit above the organic results.
• Record the raw answer, not a summary of it. Date, prompt, surface, the full response text, and the citation list wherever the surface exposes one.
• Classify every result the same way every time: named, cited, both, or absent. A human does the classification and the classification rules are written down before the first run.
• Repeat on a fixed monthly date. Same prompts, same surfaces, same rules, including the months where nothing moved.
• Report the movement with the raw answers attached, so you can read what the model actually said instead of trusting our count of it.

Named, cited, and the difference that matters

Being named is being in the sentence. Being cited is being in the source list. The two come apart constantly. A model will describe your service accurately and then link three directories instead of you, and it will also link your page while recommending somebody else in the paragraph above it.

We score them separately because they need different work. Being named depends on whether the facts about you exist in clean, quotable form somewhere a crawler can reach. Being cited depends on whether one of your pages is the most direct answer to the exact question asked. Collapsing both into a single score would hide which of those two levers to pull, which is the only decision the measurement exists to inform.

What this method cannot tell you

It is a sample, not a census. Models are non deterministic, personalised and versioned, so the same prompt can return different answers to two people on the same day. Anybody selling you an AI visibility rank is selling you a number with no denominator underneath it.

So we do not report a position, a share of voice or a percentage. We report what the answers said, on which dates, across a fixed prompt set, and whether the direction is improving. Where the sample is too small to call a change, the report says the sample is too small to call a change.

We also do not claim causation we cannot see. We can show that your name entered answers it used to be absent from. We cannot show you what inside the model put it there, and an agency that claims it can is guessing with confidence.

What the result changes

The measurement drives the work in one direction only. Absent from a prompt set your buyers use means the facts a model needs are missing, contradictory or buried, which is a content and structure problem on pages we control. Named but never cited usually means the answer does exist on your site, three paragraphs into a page about something else.

Both failures have the same fix shape: a page that answers one question completely, in the words the question was asked in, with the facts stated plainly enough to be quoted without being rewritten. That is the whole discipline. It is also why this method gates the work rather than decorating the report.

Why the method is published at all

Because how we measure AI visibility is the only part of this you can interrogate before you hire anyone. The numbers in a report are ours to produce. The method is yours to pick apart, and an agency unwilling to show you the method is telling you that the number is the product.