Illustration, fictional business. The firm, questions and figures below are invented to show the method; they are not results from a real audit.
The firm and the goal
Animated illustration: the Services page of a fictional firm, Quillwhistle Accountants. One plain sentence on the page is highlighted, and an AI answer quotes it and cites the page.
Quillwhistle Accountants is an imaginary four-person practice in Bristol. Most of its best work is accounts and independent examinations for small charities. The partners want to know one thing: when someone asks an AI assistant for an accountant for a small charity, are they named, and if not, why not?
An audit answers that in four parts: a fixed set of questions, a record of what assistants say, a check of the pages that should answer, and a fix list in priority order. Then the same questions are asked again later.
Step 1: the question set
The questions are written the way customers ask, not the way the firm describes itself. They mix recommendation, price and problem questions. The question set builder drafts a list like this in a minute.
- Which accountant in Bristol is good for a small charity?
- Who can do an independent examination for a charity near Bristol?
- How much does a charity independent examination cost?
- Best accountant for a sole trader builder in the Bristol area
- Do I need an accountant for a small charity?
- Accountant in Bristol with fixed fees
- Who prepares charity accounts in the East Midlands?
- What does an accountant charge for a small charity in Bristol?
Step 2: the record sheet
Each question is asked five times in a clean session. For each ask, the sheet records whether the firm is named, whether its site is linked and who is named instead. The method is the one in how to check if ChatGPT recommends your business.
| Question | Named | Site linked | Named instead (fictional) |
|---|---|---|---|
| Accountant in Bristol for a small charity | 1 of 5 asks | 0 of 5 | Two other fictional firms |
| Independent examination near Bristol | 2 of 5 asks | 1 of 5 | A fictional regional firm |
| Cost of a charity independent examination | 0 of 5 asks | 0 of 5 | General guidance sites |
| Sole trader builder accountant, the Bristol area | 0 of 5 asks | 0 of 5 | Online accountancy services |
The pattern matters more than any single cell. In this illustration, the firm is sometimes named for charity questions, never for price questions, and never linked as a source.
Step 3: checking the pages
Next, the pages that should answer these questions are scored across seven areas. In the example, the findings for the charity services page read like this:
- Answerability: weak. The page never says plainly that the firm does independent examinations.
- Verifiability: weak. No named accountant, no professional body, no date.
- Structure: fair. Headings exist but describe the firm, not customer questions.
- Freshness: weak. Fees mention a tax year that has passed.
- Authority: fair. A few directory listings, but with an old address.
- Crawlability: good. The page loads as text and crawlers are allowed.
- Safety: good. Nothing that would make an assistant avoid quoting it.
Step 4: the fix list
- Rewrite the opening of the charity page to answer the question in the first two sentences.
- Add a price section explaining how fees are set, since price questions drew no mention at all.
- Name the accountant who leads charity work, their qualification and the page date.
- Correct the old address on the directories that still show it.
- Add an FAQ with schema markup using the customer questions above.
Step 5: the re-check plan
The same eight questions are asked the same way once a month for three months, and the results go into a short report:
- named in X of Y asks;
- linked in X of Y;
- what was fixed and what is still open.
Our guide to reporting AI visibility without overpromising has the template.
Honest limits
Everything above is an illustration. In real audits, GEO Agency's tools run on Google Gemini, so page scores and answer simulations are estimates, not a record of what any assistant told a customer.
We do not promise rankings or a place in AI answers. If you would like us to run an audit like this for you, see the done-for-you AI visibility audit.
Common questions
Is this a real audit?
No. Quillwhistle Accountants is a fictional firm, and the questions, counts and findings are invented to show the method. They are not results from a real business.
How many times should each question be asked?
At least three to five times each, in separate clean sessions, and ideally on more than one day. Answers vary, so a single ask proves little.
Can I run this kind of audit myself?
Yes. The record sheet needs only a browser and a spreadsheet. The tools speed up the page scoring and site audit; the done-for-you service runs the whole thing for you.
How long until the re-check shows a change?
There is no fixed time. Assistants that search the web may pick up changes once pages are recrawled; some answers may not change at all. That is why the plan includes several monthly re-checks rather than one.