How we audit AI visibility, step by step: the International Academy of Football
An AI visibility audit answers one question: when a buyer asks an answer engine who to choose, are you named, and if not, why not? For the International Academy of Football the buyer is a family choosing a football and study programme abroad. Here is the audit in five steps.
The International Academy of Football is a London programme, run with the West Ham United Foundation, that combines high-performance football training with academic study for international players. It is one of the four clients that have paid CrawlerPal for this work. Its programme figures are not published, so this page shows the audit itself, with the academy's market as the example. Two steps carry real figures from a first pass we ran on 17 November 2026: 12 buyer questions in one answer engine for step 2, and eight public pages for step 3.
What does the audit cover?
| Step | When | What you get |
|---|---|---|
| 1. Scope and prompt list | Days 1 to 2 | About 20 buyer questions, approved by you |
| 2. Sampling | Days 2 to 4 | Around 240 logged answers from four engines |
| 3. Your site and your presence elsewhere | Days 2 to 4 | What an engine can read, and where you are listed |
| 4. Competitors and sources | Days 2 to 4 | Who is named instead, and on the strength of which pages |
| 5. Diagnosis and ranked plan | Day 5 | A 90-day plan ranked by effort and impact |
That is five working days for one market. Each further market or language gets its own prompt set, because the engines answer differently in each.
Step 1: who is asking, and what do they ask?
The audit starts with the buyer. The website comes later. For an academy the decision is slow and expensive, it is often made from another country, and two people make it: the player and a parent. So the questions come in more than one voice and more than one language.
We write about 20 questions across four moments, without the brand name in any of them.
- Category discovery. "Best football academies in England for international players."
- Comparison. "Football academy in the UK or in Spain, which is better for a 17 year old?"
- Recommendation. "Which football academy in London combines full-time training with A levels or BTEC?"
- Problem. "How can my son get seen by English scouts while finishing school?"
You approve the list before any sampling starts. How to find the questions your buyers ask is covered here.
Step 2: what do the engines answer?
Each question runs on four engines, ChatGPT, Perplexity, Claude and Google AI Overviews, three times each, in a clean session so that earlier chats do not bend the result. For every answer we log whether the brand is named, whether one of its pages is cited, who is recommended instead, and which pages the answer leaned on.
Named and cited go in separate columns. They are different events, and the gap between them is where much of the plan comes from.
We ran a first pass of this step for this page. On 17 November 2026 we put 12 unbranded questions to Claude, with web search on, once each: three per moment, including the four above.
| Type of question | Questions | Academy named | Its site among the pages retrieved |
|---|---|---|---|
| Category discovery | 3 | 1 | 1 |
| Comparison | 3 | 0 | 0 |
| Recommendation | 3 | 2 | 1 |
| Problem | 3 | 1 | 1 |
| All questions | 12 | 4 | 3 |
Source: CrawlerPal first pass, Claude Opus 5.5 through the API with web search on, one run per question, 17 November 2026.
The academy was named in 4 of 12 answers: for academies in London, for training combined with A levels or BTEC, for a gap year, and in an answer about cost. In those four answers it appeared under four different names, which is the naming question step 3 looks at. It was absent from all three comparison questions. The name that came up most, in 4 answers, was another private academy.
Read it with its limits: one engine, one run per question, no location set, and one answer given without a search. Of the three answers that retrieved a page, two used the academy's own site and one the site of the company that runs it.
Step 3: what can an engine read about you?
On your own site we check four things: whether the page shows its content without JavaScript, whether AI crawlers are let in, whether structured data says who you are and what you offer, and whether pages answer a question in their first lines.
Away from your site we check where you are listed. For an academy that means programme directories, agents who resell places, review sites, and the forums where parents compare notes. We also set your inbound links against those of the academies that were named.
Here is that check on the academy's own site: the home page, the academic year, the gap year programme, the women's performance academy, about us, student life, student visas and the coaches page.
| Check | Result on 8 pages |
|---|---|
| Main text readable without JavaScript | 8 of 8 |
| AI crawlers allowed in robots.txt | Yes, none blocked |
| Some structured data | 8 of 8 |
| Course, offer and school markup | 3 of 8 |
| FAQ markup | 0 of 8 |
| Headings phrased as a question | 0 of 8 |
| Full name in the page text | 3 of 8 |
Source: CrawlerPal check of eight public pages on internationalacademyfootball.com, page source only, 17 November 2026.
The site is readable and open to crawlers, so access is not what holds it back. The gaps are in what the pages tell a machine. Five of eight do not say they describe a course. None answers a question under a question heading. And the full name appears on three pages, while the other five say The Academy of Football, so an engine has to work out that the two are the same place.
Two things do not count as findings. An llms.txt file is cheap hygiene and almost nothing reads it, and a high agent-readiness score does not mean you get cited.
Step 4: who is named instead, and why?
For each group of questions we list the brands the engines recommended and the pages that carried them. Often a short list of comparison pages and directories decides a whole group of answers, and being absent from those pages explains more than anything on your own site.
This step also gives a quick read of the market: who the competitors are in the answers, which may differ from the ones you watch, and which questions nobody answers well yet.
Step 5: what is in the plan?
The audit ends in a ranked plan, read line by line by an analyst before it reaches you. It holds six lists.
- The questions to work on first, by group.
- The changes to make page by page: answers, FAQ blocks, structured data.
- The new pages to write, as an editorial plan.
- The technical fixes.
- The outside pages where you need to be listed, and what to put there.
- The link and citation openings we found.
It also fixes the baseline. Every later measurement uses the same questions, engines and conditions, so the before and after can be compared.
How do we measure success for a purchase like this?
An academy has few buyers and each one matters, so traffic is the wrong yardstick. The test is whether the same questions, run again after the fixes, name the academy more often in the groups closest to an application. After that we look at enquiries.
To size it, take what one enrolment is worth and divide the fee by it. That is the number of extra enrolments the work has to win to pay for itself. We do not promise a number. If the questions we worked on have not moved after 90 days, we re-test at no extra cost and rework the fixes that did not land.
Three client examples. This page covers the audit. The Berlitz Canada example covers the whole sequence with published figures, and the Berlitz UK example goes through the monthly work job by job.
What this means for you
You can run a small version today. Write five questions a buyer would ask before choosing a programme like yours, leave your name out, and run them in a clean session. The one-hour method is here. If you are named in none, the audit is how you find out why.
To scope one for your market, book a consultation. Prices are in the pricing section.
Frequently asked questions
How long does the audit take?
Five working days for one market: two for scope and the prompt list, two for sampling and the site checks, one for the diagnosis and the ranked plan. You approve the prompt list before any sampling starts, and an analyst reads the plan before it reaches you.
Is this the same as the free AI Visibility Score?
No. The free score is an estimate from five answers you give in a chat, and it takes minutes. Nothing is sampled. The audit measures: about 20 questions on four engines, three runs each, with the dates and conditions written down. It is the first phase of the Implementation Sprint.
Why four engines?
Because they do not cite the same pages. Roughly 86% of cited domains are unique to one engine, so a brand can be named in ChatGPT and absent in Perplexity for the same question. Measuring one engine tells you about that engine only.
Do you need access to our website?
Not for the sampling, which reads public answers, or for the site checks, which read public pages. Read access to Search Console and analytics helps with the plan and is needed later to track results. Nothing is changed on your site during the audit.
See how you show up today.
Five questions, answered in minutes: an estimated visibility score, the competitor most likely taking your citations, and the fix categories that matter. No card, no commitment. Prefer to talk it through first? Book a consultation with a strategist.