An AI visibility audit needs a reproducible question set, saved answers and separate measures for mentions and links. This worked example uses invented data to explain the method. It is not a client result or a live test of Pixel Pulse Media.
Define the test
Imagine a commercial photography business in one city. The plan uses ten unbranded buying questions across five chosen AI search experiences: 50 planned checks. Each record includes the full prompt, platform, date, market, available model or mode, answer, sources and test notes.
Keep questions that explicitly name the business in a separate brand-accuracy test. Including them in the main discovery rate would make it easier to receive a mention without demonstrating that a buyer could discover the business.
Calculate the sample results
| Sample observation | Count or calculation |
|---|---|
| Planned checks | 50 |
| Checks with a usable generated answer | 40 |
| No generated answer or unavailable result | 10, recorded separately |
| Answers naming the business | 8 |
| Answers linking to the business website | 4 |
| Mention rate among usable answers | 8 / 40 = 20% |
| Website citation rate among usable answers | 4 / 40 = 10% |
| Coverage of the planned checks | 40 / 50 = 80% |
The denominator matters. Dividing the mentions by all 50 planned checks gives 16%, which answers a different question. Report which denominator you used and keep it stable. Do not silently classify an unavailable answer as a competitor beating the business.
Turn observations into investigations
Suppose several answers link to an industry directory and the business has no relevant profile there. That is a discovery worth investigating, not proof that buying a listing will earn a citation. Check the directory’s relevance, accuracy and editorial standards before doing anything.
If an answer describes a discontinued service, inspect the cited page and the business’s current service information. Correct obsolete facts where you control them. If a page is inaccessible, record the actual status or directive and test the repair. An absent mention alone cannot diagnose a crawler problem.
Write an action record that can be checked
A useful finding includes the affected URL, evidence, proposed change, owner, effort, priority and verification step. “Improve authority” is too vague. “Replace the discontinued service description on the linked page, confirm the live copy, and repeat the same question set” can be acted on.
Check crawlable content and conventional SEO first. Google’s AI optimisation guide does not require a special AI schema or a separate machine-readable file to appear in its generative search features.
Repeat without claiming causation
Repeat the same 50 checks in a separate run. Two complete runs produce 100 responses, not 50. Compare results by platform as well as overall, and note product or mode changes. An improvement after an edit is useful evidence to investigate; it does not isolate the edit from changing answers, competitors or search systems.
The full testing guide explains the access and measurement checks. The Fix Kit supplies the workbook, while the audit service supplies the review and written priorities.