How it works
Asked like a customer. Counted like an auditor.
An audit is only worth what its worst claim can survive. Every number in your report is counted from the stored AI answers of your own audit.
-
You tell us who you are
Shop name, location, and the services you want measured. That's the whole intake. We don't ask who your competitors are — that's the part you're paying us to find out.
-
We ask what drivers ask
Questions written the way drivers really ask: “best brake shop near me” and “my car's overheating, who do I call.” We write them, and every shop with the same services gets the same set — 34 to 97 questions, depending on the services it offers.
Every question is asked with web search off and on. ChatGPT, Claude and Gemini answer from memory and again after searching the web; Perplexity always searches, so it answers with search on only. Every question is asked 3 times in each mode.
-
Every answer gets a verdict
- Recommended
- The answer endorses the shop by name.
- Mentioned
- Named, not endorsed.
- Not found
- The customer never hears of the shop.
- Inconsistent
- The assistant's own repeat answers split with no majority. The result stays in the count, but it is not counted as recommended.
A shop counts as Recommended only when the answer endorses it by name, and the quote we keep must appear word for word in that answer. If we can't verify the quote, the answer counts as not finding you. If an answer couldn't be judged at all, it's left out of the count. Matching a business name inside free AI text is imperfect: it can miss a mention, and it can credit the wrong business.
-
The rivals get counted too
Every answer that names another shop is logged: who the AI named instead of you, and how often. You never tell us who they are — we find out, and every local shop is checked against Google, so a local rival in your report is a real business and not something an assistant invented.
-
A quality gate that can say no
Before anything is delivered, the whole audit passes a quality gate: internal checks on the evidence, the counts, and the claims. If the gate can't stand behind the audit — a provider misbehaved, the evidence is thin, anything looks off — the audit is held instead of delivered, and you're told.
The units
Answers and results
Each ask is stored as its own answer. One assistant's 3 answers to one question, in one mode, combine into one result. Recommendation figures count results; "named" figures count answers — and the report says which is which.
Sample report
The report: findings and the evidence behind them
How often you're recommended — from AI memory and when the AI searches the web — assistant by assistant. Which shops the AI names instead of you. The questions where you never come up. The sites the assistants cited when they searched.
What we deliberately don't do
No dollar estimates
We will not tell you "you're losing $X/month." Nobody can trace that number to evidence, so it doesn't appear — anywhere.
No borrowed statistics
No industry factoids we can't verify. The numbers in your report come from your own audit.
No guessed reasons
We won't claim to know why an AI chose a rival unless the evidence shows it.