What is included in an AI visibility audit: the deliverables, itemised

"Audit" is the most inflated word in this market. The same word carries published prices from $99 to $4,500 (we checked — the full price ladder is here), and nobody has agreed on what the work actually contains. So instead of another definition, here is the deliverable list — ours, item by item, each one countable. If you are shopping elsewhere, take this list with you and ask each vendor which items they skip.

1. Thirty real buyer questions, across five AI surfaces

Thirty questions per report: 25 unbranded buyer questions across four buying stages — working out what exists (3), comparing approaches (8), choosing who to use (9), about to book or buy (5) — plus 5 that name your business directly: identity, facts, pricing, competitor head-to-head, and reviews. The same question set runs across five AI surfaces: ChatGPT (searching the web), Gemini (searching the web), Claude (from memory), Google AI Overviews, and Google AI Mode. The weighting is deliberate: being named to someone who is comparing options is worth more than being named to someone still reading, and different engines give different answers — so the audit measures all of it, not one chat window.

2. All 176 answers, reproduced verbatim

Every answer the questions produce is recorded in full — 176 AI answers, each individually graded, every one reproduced verbatim in the report's appendix. This is the difference between a measurement and a verdict: any claim we make about your report can be traced back to the raw text. Our published sample prints all of it for exactly this reason — read a complete real report before you decide whether to pay for one.

3. The measurement underneath: 302 API calls

302 AI API calls per report — the calls that produce answers, plus the batched calls that grade them. On top of the main run: your 3 key questions are each asked 3 times to test stability (6 repeat answers), because assistants that search the web do not answer identically twice. Local businesses also get map-pack reach: measured from 5 points around your location — your own door plus four about 5 km out — for each of your 3 key questions, 15 map-pack probes in all.

4. A claim-by-claim fact-check of what AI says about you

Every checkable claim the five direct probes produce is extracted and checked against your live pages. In the published sample — a real client run, pseudonymised — that was 333 claims: 178 checked and correct, 14 wrong, and 141 that were not on the site at all — statements about the business that its own pages neither confirmed nor contradicted, leaving AI to fill them from elsewhere. The third category is the one most businesses never think about, and often the easiest to fix: the claim is not wrong, it is just unclaimed.

5. A price-claim audit

Prices get their own audit, because cost questions are where buyers decide. In the same sample: 53 price quotes. The 36 that matched the live website catalog exactly are collapsed in the report; the 17 with no price on the site to check them against are listed row by row, engines side by side, so you can see which assistant assembled which figure from which third party.

6. The four technical pillars

The diagnosis behind the score, in a single matrix: AI crawler access (are the crawlers blocked by directive), structured data (does the JSON-LD actually parse), service and price pages (can a machine read what you charge), and third-party citations (which sites the answers are built from, and whether you are on any of them). Each pillar gets a status and a pointer to the fix.

7. A fix roadmap sorted by who can do it

Not a list of suggestions — a sequence. Stage 1, stop the bleeding (owner or marketing, first): the entity and pricing claims that are actively wrong. Stage 2, technical fixes (whoever runs the site, this week): schema and rendering. Stage 3, win the sources (PR or growth, a monthly push): the citations and communities the answers are drawn from. Every item carries the evidence behind it and who can execute it, so the list can be handed straight to the people named on it.

8. Three paste-ready fix blocks

Generated from what the run measured about your business — no placeholders left to fill in: a disambiguation paragraph for the home page and your Google Business Profile; a complete llms.txt for the site root (why and how, in our llms.txt guide); and a JSON-LD block to insert just before the closing head tag of your home page. Hand all three to whoever maintains the site. The text is in English on purpose: it is for your site and for the AI systems reading it.

What it costs, and where to start

The audit above is $4.99 once, no subscription, with a 30-day money-back and delivery in about 20 minutes; the method behind it (METHOD v1.3) is published and frozen. The same measurement structure powers our published, measured examples: the pseudonymised med spa sample and a measured family dental practice.

But start smaller than any of this: the free check tells you whether AI systems can technically read your site, and shows you all 30 questions it would ask, in full, before you pay anything or even make an account. If the deliverables list above answers a question you were going to ask a vendor, it has done its job.