AI visibility audit: built from your data

What does an AI visibility audit measure?

Not where you rank. Whether you are named.

When somebody asks an AI assistant which company to use for something, it does not hand back a list of links. It writes an answer. You are either in that answer or you are not, and an audit measures the “or you are not” — across a fixed set of questions, across the systems your buyers use, repeated often enough that the result is a rate rather than a lucky screenshot.

• Mention rate — how often you are named at all.
• Citation rate — how often your site is linked as a source. It is always lower than the mention rate, and the gap between the two tells you something on its own.
• Share of voice — your mentions as a share of everyone mentioned, so you can see who is being recommended in your place.

Each is reported per system, because they do not behave the same way. A company can look healthy in Google’s AI Overviews and be absent from ChatGPT. In our experience that is the usual starting point, not the exception.

Where do the questions come from?

This is where our audit parts company with most of them, and it is the part that decides whether the number means anything.

The common approach is a prompt pool: a bank of generic questions for a category — “best accounting software Greece”, “top AI agency Athens” — reused from client to client with the name swapped in. It is quick, and it produces a figure. What it cannot tell you is whether those are the questions your buyers ask.

We build the set from your data. The searches that already bring people to your site. The terms your ads match against. The questions your sales team hears on calls and your support team sees in tickets. Where that data exists, nothing is guessed: the questions are the ones real people asked, in their own words.

Which has a consequence we would rather say now than later. If the data is not there, we do not run the audit yet. A company with no Search Console history, no ad account and no record of customer questions cannot be measured against its buyers’ language, because nobody has captured it. The first step is to start collecting, wait until there is enough, and audit then. We would rather delay a measurement than pull one from a pool and call it yours.

Why is the baseline taken before anything changes?

Because afterwards it cannot be.

The moment a site changes, the clean picture of what the engines said before is gone. Every company that has invested in AI visibility without a prior baseline reaches the same dead end a year later: something shifted, and nobody can say what shifted it, or whether the money did anything at all.

So the baseline comes first. Before any recommendation, before any edit. Every answer is stored exactly as the engine wrote it — the text, not a summary — with the date, the system and the session conditions. When a number moves later, that archive is how you find out why.

Each question is asked more than once, in clean sessions with nothing personalised, because these systems vary between runs and a single answer is not a measurement. One thing worth knowing about Google in particular: its own documentation says a page only needs to be indexed and eligible for a snippet to appear in AI Overviews, with no additional technical requirement. That is why the Google column of an audit usually looks fine while the others do not, and why treating “AI visibility” as one thing produces the wrong plan.

What happens between the two measurements?

The work. This page will not detail it — the detail is what a client pays for — but the shape is not a secret.

The baseline findings sort into three groups. Things the engines cannot reach: content rendered in the browser that a crawler retrieves empty, a crawler blocked by a stray rule, structured data that says one thing on the homepage and another on the services page. Things about how the company resolves as an entity: whether one organisation, one address and one description exist consistently across sources you do not control. And things about the writing itself: whether a paragraph lifted out of the page still makes sense on its own, which is how these systems actually read.

The first group is fixed in weeks and is entirely in your hands. The second and third take longer, because part of the work happens on directories, review platforms and publications that are not yours. We say that in the plan instead of promising a date.

Why measure a second time?

Because a baseline on its own is a diagnosis, not a result.

Most audits end at the report. Ours includes a second measurement after the work — the same questions, the same conditions — so the comparison is like for like instead of estimated. That is not an extra. It is what makes it an audit. Without it you have a list of findings and no way of knowing which ones mattered.

Here is what that looks like. A Greek B2B professional services firm, years of proper SEO behind it. The same 172 buyer questions, before and after:

Google barely moved because there was almost nothing left to win there. The others moved because the work was aimed at what actually decides those answers. You cannot reach either conclusion from a single measurement.

The full case is published separately, method included.

And afterwards?

A second measurement tells you what three months of work did. It does not tell you what happens in month four.

For companies that want to keep watching, we run an AI visibility dashboard as a separate, ongoing service: the same prompt set, run every week across the same systems, every answer stored verbatim, and a login where you see mention rate, citation rate and share of voice per system, the full history, and which competitors are turning up in the answers you want. It is not part of the audit and it is not in the prices below. It is there for when a baseline needs to become a trend.

There is a live demo with real historical data. It is worth opening before you decide on the audit, because it shows what “after” looks like when somebody keeps measuring.

What does it cost?

The prices are published, because the cost question is one of the highest-intent things a buyer searches for and almost nobody in this category answers it.

The visibility audit is €550 — one-off, about a week. Standard, at €700, adds the second measurement a month later and a review meeting. Regional and multi-market packages have their own published prices. Everything excludes VAT.

Anything beyond that — several brands, several markets, a large prompt set, an enterprise scope — is quoted, because one number would mislead everybody at both ends. Bring the scope to a thirty-minute call and we will tell you which band you are in before anyone writes a proposal.

You may have read that anything under €1,500 is a tool export with commentary on top. Ours is not. The prompts are built from your data, every question is run repeatedly in clean sessions, every answer is archived verbatim, and we present the findings to you in person. The price is possible because we built the software that does the running, so the labour goes into the analysis rather than the copying.

What you get

• The prompt set, documented and dated, built from your data and yours to keep
• The baseline: mention rate, citation rate and share of voice per system, every answer archived verbatim
• The competitor picture: who the engines name instead of you, question by question
• The findings, grouped by what you control and what you do not, in the order to work through them
• With Standard, the second measurement on the same set, so what moved is measured and not guessed

All of it is yours whether or not you continue with us, and the findings are written so another team could act on them.