See the audit before you ask for one
This is a real audit of a US home builder, with the names, cities and URLs removed and the full report cut down to six findings. Scroll through it. Each callout says what we did, what it means, and what you would do next.
The figures are from the original run. They show what an audit looks like, not what any engine says about your client today.
The setup: five questions become 264 probes
Audit inputs
on your lot home builder [City], [State]
cost to build a custom home on your own land [State]
affordable custom home builder [Region]
shell program home builder [State]
on your lot home builder [Second city], [State]
What we did
We turned five seed questions into 264 live probes across ChatGPT, Perplexity, Gemini and Google AI Mode, spread over four buyer personas. Local questions name the city outright, never "near me", because an engine can't tell where a simulated searcher is standing.
What it means
You get a sample of what buyers actually type, not one prompt. The report states a confidence level for the run, and says plainly that it describes this run, not how far the numbers would move on a repeat.
What you do
Check the questions against your client's real market before anything else. If a seed question is wrong, every finding after it is wrong, so this is the first thing we ask you to look at.
What the engines say: who gets named, and by which engine
Visibility snapshot
What we did
We captured every full answer and every source it cited, then recorded whether the business was named, whether it came first, and how it was described.
What it means
When nobody names the business in the question, engines name it about 3 times in 10. The four engines also sit 18 points apart, so a tracker that checks one engine would tell you a very different story depending on which one it checked.
What you do
Treat the unassisted rate as the number to beat, and report it per engine. Each engine needs its own look, and the follow-up audit measures the same thing.
Where it lost: the answer cited someone else
Evidence gap
cost to build a custom home on your own land [State]
Winning source: a national cost guide
States cost per square foot for builder-grade and custom builds, with a total range.The client's page
Describes affordability in words, with no figure an answer could quote.What we did
For each question the business lost, we compared its page with the source the engine chose, using embedding models and a reranking model. It is the same category of retrieval system these engines use to decide what to cite.
What it means
59% means the page is on the right topic but lacks the specifics an answer needs. That is a content gap, not a missing page, and the report says which. Under 30% usually means the page doesn't exist yet.
What you do
Add the missing fact to the page that already exists instead of commissioning a new one. It is a smaller job, and the report tells you exactly what the sentence has to contain.
The gap, named: the sentence that is missing
Findings
What we did
We read each answer for what it says about this business specifically, so a rival's phone number or reviews are never pinned on your client. We also logged the claims engines make that the site never states.
What it means
Not "lost citation, unclear why". Three specific, checkable facts, each with a confidence level.
What you do
Publish a price range, state the license number on the page, and answer the cost and timeline worries head-on, where a buyer will read them.
Fixes, ranked, with the evidence behind each
Prioritized fixes
1. Publish your own price ranges on the main builder pages
Predicted outcomeWinning pages state cost per square foot and a total range. This client's pages carry only qualitative language.
Does not establish: that adding prices will raise visibility, only that competing sources show explicit figures and this client's pages do not.
2. State licensing and credentials on the site
Directly observedEngines already credit the business with a license the site never states.
3. Answer the reservations engines raise unprompted
Directly observedCost surprises and delays appear in answers without anyone asking about them.
What we did
Every fix links back to a specific finding. Nothing here is a generic best-practice list you could paste into any builder's report.
What it means
Two badges keep us honest. Directly observed means we measured it. Predicted outcome means the gap is measured but the effect of fixing it is not, because no report can measure a change before it happens.
What you do
Hand the ranked list to your client as it is, or use it as the scope of work for your next retainer conversation.
Verify: close the loop
Verify after implementation
Re-run these questionsaverage cost per square foot custom home build [State]
[Client] [City] pricing and cost
What we did
We keep the original run as a baseline, along with the exact questions and engines, so the same test can be repeated later.
What it means
A follow-up asks the same questions the same way. Any change is measured against this run, not against anyone's memory of it.
What you do
When the fixes are live, ask for a follow-up audit. We compare it with this baseline and show what moved, what didn't, and what to try next.
In the full audit, too
Local checks
Map-pack results, and name, address and phone consistency across listings, including businesses with several locations.
A developer brief
Ready-to-paste structured data and plain-language tasks for whoever edits your client's site.
Your brand on it
White-labeled, in language written for your client to read, not a login you have to explain every month.
A person reads it first
Every audit is reviewed by a human before it reaches you.
Want one for a client of yours?
Send us one client site. We'll email to confirm it, then send your first audit free, white-labeled, within 48 hours.