What four weeks covers, and what it does not

One question, one change, one report

We agree on a single buying question a shopper would plausibly ask before choosing between you and an alternative. Everything in the window hangs off that question.

From it comes one diagnosis, one proposed change, your decision, and — if you approve it — one verification and one report.

What we look at

Public information, and what you tell us about it. That is the whole input surface.

• The pages you publish about the product in question.
• What an AI answers when it meets the agreed question, and the reason it gives.
• Which sources that reason leans on, yours and otherwise.
• What you tell us is out of date, incomplete or wrong in your own material.

What we leave alone

A short list, and it does not get renegotiated mid-window. If a diagnosis points somewhere on it, the diagnosis stops there and says so.

• Your systems. We are never added as an administrator anywhere, and we never ask to be.
• Your customers. We do not want lists, exports, contacts or anything about the people who buy from you.
• Your pricing and your commercial terms.
• Anything you told us was not on the table when we agreed the question.

What a pilot can measure

Two things, both about the change itself, and both stated plainly.

Whether the approved action was actually performed, in full, on the asset it named. And whether the environment we care about had an observable chance to see it afterwards.

What a pilot cannot measure

Whether a shopper chose you because of the change. Whether an AI system weighs your material differently now. Whether anything about your revenue moved.

None of these become knowable by running the pilot longer or by reading the same evidence more generously. They are outside what this method can establish, and a report that implies otherwise is the thing this page exists to rule out.

What you need before starting

Not much, but all of it. A pilot missing any one of these produces a result that cannot be read.

• A public page about the product, that you control and can edit.
• A buying question you actually care about the answer to.
• One asset you are willing to change and able to change back.
• A named person who can approve a change without a committee.
• A window in which nothing else large is being changed on the same pages.

How it ends

With a written report at a stated grade, and a recommendation about whether a second pass is worth doing.

Often the honest report is that the change shipped, nothing observable followed, and the reason is unknown. That is a real result and we say it in those words. A pilot that can only end well is not a pilot.