Agent view — served from /case-studies.md.
Results
This practice launched in 2026. Rather than dress up someone else's numbers or publish a stock photo with an invented quote, here is exactly what we can prove today, what we will publish when we can, and what we are offering the first companies who let us.
What we can prove today
Including the checks we fail. It runs hourly, records its rubric version, and links the raw JSON so you can check the arithmetic yourself.
It is not a substitute for a client result and we are not pretending it is. But it does demonstrate the one thing a case study would: that the engine runs, produces evidence, and that we will publish a number we did not choose.
See the live audit→46 of 49 scored dimensions passing. Two failures and one partial, all published with the reason.
verified · not illustrative
The format
Three worked examples, in the shape the real ones will take. Same rubric version at both ends, evidence for every claim, and the things that did not improve reported alongside the things that did.
A single-page application where the product and pricing pages returned an empty shell to any crawler that did not execute JavaScript. AI assistants described the company using a three-year-old press release, because that was the only readable source.
Machine-readable prices and stock levels that had drifted away from what shoppers actually saw. Assistants comparing products quoted stale prices and flagged items as available when they were not.
A hosting default, never reviewed, that disallowed all AI user-agents. The firm had been invisible to AI assistants for eleven months and had attributed the traffic decline to market conditions.
These three are worked examples, not clients. The scenarios are drawn from failure patterns the audit is built to detect; the numbers show what the report looks like, not what anyone achieved.
Founding clients
We need real results to publish. You need this done at a lower price. That is a straightforward trade and we would rather make it openly.
Thirty per cent off the audit and the first three months of any retainer, in exchange for permission to publish the before and after — named if you are comfortable, anonymised by sector if you are not. You approve every word before it goes up, and you can withdraw permission at any point.
If the audit finds little worth fixing, we will say so and publish that too. A study showing a site was already in good shape is still an honest study.
Ask about founding client terms→