We put our own brand on the meter first and published every reading.
Before selling this to anyone, we ran the full method on our own site and published every record it produced, including the part that does not flatter us. The machine layer is built and verified. The citation panel is live, dated, and sampled weekly. This is the same report format a client receives, run on our own brand first, which is why you can check the method before you buy it.
Before: thin machine-readable hygiene
The site looked finished to a human. Crawlers could reach it; that part worked. What was missing was the machine-readable hygiene layer: no structured data describing the business, no brief written for the crawlers, and no stated policy on who was allowed in. That is a hygiene gap, not a proof that AI answers were citing us elsewhere or that fixing it would change whether they do; the citation number is tracked separately, below.
| Machine layer | Before | After |
|---|---|---|
| Machine-readable business identity | No structured data describing who we are | Complete, checked per engine |
| Machine brief for AI engines | Missing | Published, written to the audit |
| AI crawler access policy | Undefined | Explicit, every engine covered |
| Site discoverability layer | Missing | Published and kept fresh |
| Canonical and social metadata | None | Complete |
| AI crawler user-agents reaching the site | Untested, undocumented | 7 of 7 returned HTTP 200, in writing |
Had that first version shipped as it was, it would have scored about 30 out of 100 on our own scanner: crawlers could reach it (a separate, passing check), but the structured-data and hygiene checks that also feed the score were largely missing. Readability is a hygiene signal, not evidence of citation; it does not by itself show whether AI answers named us before or after.
The method, applied to ourselves
Hadal mapped exactly what AI crawlers could and could not see: access, structure, entities, gaps.
A machine layer written to the audit, not from a template. It states to the engines what the site never quite did.
Every one of the 7 crawler user-agents checked and documented, with the HTTP status in writing. A user-agent access check you can reproduce, not proof the real crawler reached the site from its own IP.
The whole before and after captured in the same report format a client receives, with the exact curl command behind each check included.
Part 1 done: the readability layer is built and checked
Today deepoceanstudio.com carries a complete machine layer. All 7 AI crawler user-agents we test return HTTP 200: a user-agent access check you can reproduce, not proof the real crawler reached the site from its own IP. The readability score is 94 out of 100 on our own scanner, a hygiene signal measured separately from whether any engine names us. It read 100 out of 100 until 2026-07-18, when we cut the homepage FAQ block and its FAQPage schema went with it.
Readability alone did not produce a citation. In the week the score read 100, sampled 2026-07-08, our citation score on a fixed panel of buyer questions was 0 of 8. A perfect machine layer with zero citations is why we treat readability as a hygiene signal that does not by itself produce citations, and why this engagement has a part 2.
Part 2: the number that counts, measured on the public ledger
The questions a buyer would ask an AI engine about our own category, sampled in ChatGPT via a published protocol on an intended weekly cadence with missed runs shown, every named domain recorded, every intervention annotated on the record as it happens. Entry 1, sampled 2026-07-08 on panel v2: 0 of 8. Entry 2, sampled 2026-07-16 on panel v3 and sampled twice the same day: 0 of 11. Entry 3, sampled 2026-07-23 with three control questions alongside it: 0 of 11. The same rows record who the answers named instead, and which sources those answers drew on. Our own number is published before anyone else's. Watch what happens on the public meter.
This is my own brand, on purpose. Engagement #0 means I do to myself exactly what I do for a client, and I publish it, including the zero. If the number moves, you will see when it moved and what class of work preceded it. If it does not move, you will see that too. Every client engagement runs this same protocol and publishes to the same ledger, on the same terms.
Nicolò Brignoni
Founder, Deep Ocean · follow the work
Prefer to read the raw method first? The AI Readability Reports run the same scan across 54 real businesses.