What this does

They test the change you specify. We say which change to make — and then tell you whether it worked, including when it did not.

The established tools are experimentation platforms: you bring a hypothesis, they render variants and measure the effect, and they need roughly a thousand similar pages to do it. Neither tells you what to change. In their model that is your job. Here it is the product.

We name the change

Not “improve your title”. The words the page ranks for and never says, ordered by the demand behind them, with replacement copy already written and checked against the length Google will actually show.

We work at one page

A split test needs volume in both arms. This finds a page ranking on the first results page and never clicked, and acts on that page alone. The sites they cannot serve are the ones this is for.

We say when we cannot tell you

Before the work, not after. A page too quiet to measure is labelled as such, because a verdict of “no change” on a page without the traffic to show one means we could not see it — not it did not work.

We record what did not work

Every recommendation keeps its verdict, including “measured, and it did not help”. That is deliberately unsoftened: the value of a verdict is that it can be unflattering.

What we are not claiming

• That measuring our own advice is unusual. The established tools do causal measurement, and better. We claimed otherwise earlier in this project and withdrew it when we checked.
• Any percentage lift. As of the date above, no recommendation has completed its 28-day review window, so there is no honest number to quote — and inventing one would contradict everything else on this page.
• That we apply changes for you. Today a recommendation arrives as instructions and copy that a person applies.

Terms, including what we learn from your results, and Privacy.