A thirty-minute call
Bring the surface you're worried about — what customers ask, what the answers touch. If a check isn't worth running, we say so on the call.
How we work
From first call to signed record.
What happens, what arrives, and how long it takes. No discovery theater — most of this page is the engagement itself.
The engagement
01
02
03
04
05
Five steps, most of them ours.
A sampled scan
Before an engagement is scoped, we run a sampled scan of one public AI surface and send you what we saw. It's a sample, not the engagement — enough to decide whether the full check matters.
Capture and corroborate
We ask the AI what your customers ask, preserve each answer with its prompt and timestamp, and check each claim against what you actually publish. Anything that doesn't reproduce across runs gets dropped, not reported.
The report and the fix spec
Two documents, written for an owner: what we found and what it costs you, then the exact fixes — what to change, where, and how you'll know each one worked. The receipts live in an appendix, not the body.
Re-scan and certificate
Your web person applies the fixes — we don't sell that work inside a verification engagement. Inside thirty days we run the same checks again, and when the surfaces hold up, the engagement closes with a signed accuracy certificate in the public registry.
What arrives
Documents an owner can act on.
- The findings report: what we checked, what we found, and the single worst finding quoted — first page tells the whole story.
- The fix specification: one block per fix — the change, the page it goes on, the exact copy to paste, who should do it, and the check that proves it's done.
- The evidence appendix: each capture with its prompt, surface, and timestamp, so any claim in the report can be reproduced.
- After the re-scan: the accuracy certificate, registered at its own permanent URL.
How we verify
The rules we hold ourselves to.
A finding has to reproduce. In our July sweep, one apparent scheduling discrepancy did not survive a second run ten minutes later. It was dropped and was not claimed. The fabricated free consultation that did reproduce, twice, is the one in the log.
A clean result is a real result. Six of the thirteen businesses in that sweep checked out exactly, and we said so. We'd rather tell you your surfaces are fine than invent a problem to bill against.
We do not promise placement or rankings in an AI answer — there is no ranking to buy, and anyone selling one is selling something unfalsifiable. What we claim, you can check.
Inside a verification engagement we don't sell the fix — the inspection and the repair stay separate, on purpose. And if we later build or host AI for you, that surface stops being ours to verify independently: it gets continuous QA and monitoring instead, labeled as our own work.
Start
Bring one surface.
The first call is thirty minutes with Cooper. If you'd rather start with the intake form, it lives on the engagements page.
If the work continues past verification — advisory, or building and operating AI for your business — that continuation is scoped the same way: after a call, in writing.