Cited Layer › Services › Category source map
Service
AI visibility audit: the category source map
An AI visibility audit measures who AI answer engines name when your buyers ask the questions that lead to a purchase — and, more usefully, which third-party sites those answers are built from.
I run ten of your buyers' real questions across ChatGPT and Perplexity, three times each, in one fixed mode: sixty recorded answers. In most categories two to four sites decide what the answers say. You get that list, whether you are on each site, what each one requires, the share of answers that name you, the competitors named in your place, and the questions and tactics that do not work in your category — with the reason. $600, five working days.
It is delivered in writing. No call is required at any point.
Who does AI visibility audits this way: an independent consultant, not an agency and not an automated score. The whole method is published on this site, including the measurement where my own site scored zero.
Why repeated runs
One run of a question is a coin flip. Here is the measurement.
On 21 September 2026 I asked two ChatGPT accounts the same question in the same minute, in the same mode, on the same free tier: AI search optimization agency.
Account one
Returned a shortlist of five agencies, with links.
Account two
Returned no shortlist at all. It explained what the service is and how to choose a supplier, and named nobody.
Ten days earlier, the same question asked twice fifty minutes apart returned five vendor names each time, of which two matched.
This is not an argument against measurement. It is the reason a single pass cannot be one. Any audit priced on one screenshot per question is selling a coin flip, and the buyer has no way to see it. Three runs per question, in a mode that never changes between the baseline and the re-test, is the smallest protocol that produces a number you can act on.
Protocol
What is actually done
-
Your questions, not mine
Ten questions are agreed with you first, in your buyers' wording. If you do not have them, they come out of your support inbox and your search console — not out of a keyword tool.
-
Sixty recorded answers
Ten questions × three runs × two engines, in one fixed mode: same browser, same account tier, temporary chat, runs separated in time. Each answer is saved in full, with its sources, and screenshotted.
-
A control question every session
A question whose answer is known to return a vendor list. If the control fails, the session degraded and its results are discarded rather than reported.
-
Source analysis
Every cited domain is recorded and classified: the vendors' own sites, directories and rankings, review platforms, media, communities. This is what tells you where the engines in your category actually get names — and it differs sharply by category.
-
The intersection
The sources both engines cite across your questions are intersected. What remains is the short list of sites that decide your category — usually two to four, not twenty. Their selection criteria are often visible in the answers themselves, so what each one requires can be read before anyone applies. Server logs are a separate product: the AI traffic check.
-
The verdict, including the negative half
Which of your ten questions can be won, which cannot, and why. In ChatGPT, six of ten realistic buyer questions return no vendor list at all — there is no position to enter, and no amount of work creates one.
Deliverable
What you receive
- Presence rate — the share of the sixty answers that name you, per question and per engine.
- The incumbent set — who is named in your place, and how consistently across runs.
- The deciding sites — the two to four domains both engines build answers from, whether you are listed on each, and what each requires.
- The full citation layer — every domain the engines quote, ranked by how often it appears.
- The fact gaps — the checkable claims your competitors publish and you do not.
- What does not work here — tactics that earn nothing in your category, measured rather than assumed. In one market forums were cited zero times in seventeen measurement runs.
- The unwinnable list — questions with no vendor list to enter, stated before you spend anything on them.
- The raw record — all sixty answers verbatim with sources, so you can repeat the measurement yourself.
Written report. Turnaround five working days. Delivered in English or Ukrainian.
Price
Start with the free check
| What | Scope | Price |
|---|---|---|
| Category check | Four of your buyers' questions in ChatGPT and Perplexity. The answers verbatim, every source listed under its answer. No scoring. Two working days. | Free |
| AI traffic check | Fourteen days of your server logs: which pages ChatGPT and Perplexity actually read, whether a ChatGPT drop is lost tracking or lost visits, and what your structured data claims that your pages do not. No prompts. Three working days. | $400 |
| Category source map | Ten buyer questions, three runs each, two engines. The two to four third-party sites that decide answers in your category, whether you are on them, what each requires, and which popular tactics do not work in your category. Five working days. | $600 |
| Placement program | Getting onto those sites, fixing what the logs showed, monthly re-measurement and ChatGPT fetcher monitoring. Guaranteed: the work and the measurement. Not a citation. Three months minimum. | from $1,500/mo |
The free category check exists because roughly half the categories I test cannot be won on the questions the client cares about. Thirty minutes of measurement answers that before either of us commits, and it is the reason I will sometimes tell you not to buy anything.
Limits
What I do not promise
No rankings, and no "top 3 in ChatGPT"
Positions do not exist in these systems. An answer names three to eight brands and the order between them is not stable — the measurement above shows the same question returning a shortlist to one account and no shortlist at all to another, in the same minute. Anyone selling you a position in an AI answer is selling something that has no definition.
What can be measured, and therefore what I sell, is the share of answers that name you, on a fixed question set, measured the same way before and after.
No single opaque "AI visibility score"
A score computed once, from one pass, with an undisclosed method, cannot be reproduced by you and cannot be compared to anything later. Every number in my reports comes with the question that produced it, the date, the mode, and the full answer text.
No guarantee that your category is winnable
Some are not, and I will say so in the free check rather than after the invoice. Where a category has no vendor list in the answers, there is nothing to optimise into.
Questions
Frequently asked
What is an AI visibility audit?
A measurement of how often AI answer engines name your brand when buyers ask the questions that precede a purchase, and of who is named in your place. A useful one records the exact questions, the exact answers, the sources cited, and repeats each question enough times to separate a real result from noise.
Who does AI visibility audits?
Specialist GEO and AEO agencies, a small number of independent consultants, and a growing set of automated tools. The difference that matters is not the label but the method: how many questions, how many runs per question, whether the mode is held fixed between measurements, and whether you are shown the raw answers or only a score.
How much does an AI visibility audit cost?
Automated tools start in single dollars per scan. Agency audits run from a few hundred to several thousand dollars, with large agency engagements above that. Mine is $600 for sixty recorded answers reduced to the sites that decide your category, and the category check that precedes it is free.
Which engines are covered?
ChatGPT and Perplexity, because those are the two where the buyer sees a shortlist of vendors. They behave very differently: Perplexity names service providers on almost any phrasing, while ChatGPT returns a vendor list only on particular constructions and answers everything else with advice. Adding more engines to a report inflates its page count more than its usefulness.
How long does it take?
Two working days for the free category check. Five working days for the full audit.
Do I need a call?
No. Everything is delivered in writing. Calls are available in Ukrainian and Russian if you want one.
What happens after the audit?
You can implement the findings yourself — the report is written to be handed to a developer — or ask me to do the work in the placement program. Nothing is required after the audit.
Why should I believe the numbers?
Because the method is published, the raw answers come with the report, and you can repeat any measurement yourself. The measurement of this site's own visibility is on it too, and at baseline it scored zero out of ten.
Start with the free category check
Send your site and three or four questions your buyers actually ask. You get back the eight runs, the names that appeared instead of yours, and the domains the engines cited — in two working days, in writing, with no call attached.
alex@citedlayer.com