An AI visibility audit: what the machines say about you.
Your buyers are asking ChatGPT and Gemini which vendors to shortlist, and getting an answer. Almost nobody knows what that answer currently is. This is the three-week engagement that tells you, and it is the usual first step into our online marketing services.
● Start here
Tell us about the business.
It takes about a minute. It reaches the people who would do the work rather than a sales inbox, and if we are not the right fit for what you need we will say so.
Goes to hello@katama.io. No sequences, no newsletter, no list.
● What is an AI visibility audit
An AI visibility audit measures how often a company is named in the answers ChatGPT, Google AI Overviews, Perplexity and Gemini give to its buyers’ questions.
It produces a fixed prompt set, a share of answer score for each engine, the list of your pages that get quoted, the competitors recommended in your place, and a ranked list of fixes. The baseline is yours to keep and re-run, whether or not you hire the agency that built it.
You cannot manage a number you have never measured.
Your rankings report stopped answering the question.
For twenty years, position in Google was a usable proxy for whether a buyer would find you. That proxy has come apart. The share of Google AI Overview citations drawn from the top ten organic results has fallen to roughly 38%, from about 76% in late 2024.
So a company can hold every ranking it wanted and still be absent from the answer its buyer was actually shown. The reverse happens too, which is the opportunity rather than the threat.
Citation overlap
38%Of AI Overview citations now come from the top ten organic results, down from roughly 76% in late 2024.
The honest position is that almost no B2B company has a baseline. Not a screenshot of one ChatGPT session, but a fixed set of buyer questions, run across every engine that matters, scored the same way each time so the number means something next quarter.
Without it, every decision about AI search is argued from anecdote, and every agency proposal you receive is unfalsifiable. That is the actual reason this engagement exists, and it is why we sell it separately rather than folding it into a retainer.
Yours to fix
88.3%Of citations at the B2B consideration stage point back at pages the brand itself owns.
Citation overlap
38%
Of AI Overview citations came from top ten organic results in 2026, down from roughly 76% in late 2024. A rankings report no longer predicts this.
Brand-controlled pages
88.3%
Of citations at the B2B consideration stage point at pages the brand itself owns, which is what makes an audit actionable rather than interesting.
AI visitor value
4.4x
The reported conversion value of a visitor arriving from AI search against a traditional organic visitor.
Never fetched
97%
Of llms.txt files across 137,000 sites were never requested by any AI crawler. The audit also tells you what not to buy.
What you get.
The prompt set is the part that decides whether any of this is worth reading. We build it from your recorded sales calls, your search data and the language in your own pipeline, not from a keyword tool, because the questions buyers ask a chatbot are longer and more specific than the ones they type into Google.
Then every prompt runs across five engines, several times, because the answers vary between runs and a single pass tells you almost nothing.
The structural read comes last. We go through the pages that should be getting cited and mark them against the tactics the peer-reviewed research actually measured, rather than against a checklist someone wrote on LinkedIn.
● Included
- A prompt set of 40 to 60 buyer questions, built from your data
- Every prompt run across ChatGPT, Google AI Overviews and AI Mode, Perplexity and Gemini
- Share of answer scored per engine, with a method you can re-run
- The competitors recommended in your place, named and counted
- Every URL of yours that got quoted, with the passage that was lifted
- A structural read of your key pages against the measured GEO tactics
- Entity and schema audit: what an engine can resolve about you today
- robots.txt read for AI crawler access, and what any block is costing you
- A ranked list of fixes, sized by effort, with the four to do first
- A ninety-minute session walking your team through all of it
How the work runs.
Build the prompt set
Week one. Call recordings, search data and your own pipeline language turned into the questions your buyers actually ask. You approve the list before anything is measured, because the prompt set is the audit.
Run the baseline
Weeks one and two. Every prompt across five engines, repeated, scored per engine. This is where most companies find out that a competitor they do not rate is the default recommendation in their category.
Read the pages
Week two. The pages that should be cited, marked against answer structure, cited statistics, attributed quotation and entity markup. Plus the crawler access check, which is occasionally the whole finding.
Deliver and walk through
Week three. The document, then ninety minutes with the people who have to act on it. You leave with four things to do first and a number to re-measure against.
What gets scored.
Four numbers, each defined in the document so anyone can reproduce them. If a metric cannot be re-run by your team without us, it does not go in.
Share of answer
The proportion of the prompt set where you are named at all, per engine. This is the headline number and it is usually lower than anyone expects.
Cited pages
Which of your URLs get quoted and which passage was lifted, which tells you what to build more of far more precisely than a traffic report.
Competitive displacement
Who is recommended instead of you, on which prompts, and whether it is the same two names every time.
Crawler access
Whether GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended can read you at all. A blocked crawler makes every other finding academic.
Who this suits.
- Your buyers research vendors before they contact anyone
- You want a number before you commit to a program
- You have real service or product pages for an engine to read
- Someone internally can approve page changes afterwards
- You want a guaranteed position, which nobody can sell honestly
- You want the document to be the end of it, with nothing changing after
- You need it inside two weeks, which we would rather decline than rush
- Your buyers genuinely do not use these tools, which is worth checking rather than assuming
Fair questions.
What do we actually get?
A document and a working session. The document is the prompt set, the baseline scored per engine, the list of your URLs that got quoted with the passage that was lifted, the competitors appearing instead of you, and a ranked list of fixes with the four we would do first. The session is ninety minutes walking through it with whoever has to act on it.
How long does it take?
About three weeks from the kickoff call. Most of that is building a prompt set worth trusting, because a baseline measured against the wrong questions is worse than no baseline.
Do we have to hire you afterwards?
No. It is priced and delivered standalone, and the prompt set is yours to re-run with anyone. We would rather you had the number than the engagement, because roughly half the fixes in a typical audit are things a competent in-house team can ship without us.
Can we not just check ChatGPT ourselves?
You can, and you should, before you pay anyone for this. What is hard to do casually is a fixed prompt set run across five engines and scored the same way every month, which is the difference between an anecdote and a baseline you can manage against.
Is this the same as an SEO audit?
No, and it does not replace one. An SEO audit tells you why you do or do not rank. This tells you whether you are being quoted, which is now a substantially different question: the overlap between the pages that rank and the pages that get cited has fallen sharply since 2024.
Works well with.
Answer engine optimization
The program the audit feeds. The baseline tells you what to fix, this is the team that fixes it and keeps measuring.
Read →SEO
The other half of visibility. Ranking still raises citation probability substantially, and both workstreams read the same pages.
Read →Website conversion
A citation sends a pre-qualified visitor to a page that then has to earn the enquiry unattended.
Read →