Optimize

Everyone ranks the fixes. We grade them afterwards.

Any tool can hand you a prioritised list. The question that decides whether the list was worth following is the one almost nobody answers: three months later, did the number actually move? We write down what should move before you act, and we publish the verdict after, including when the answer is no.

Start with the free auditNo account, no card, full result.

Four steps, and the fourth is the rare one

We measure

Real answers from real engines, with the number of answers printed beside the result. Nothing modelled.

We recommend, and say what should move

Each action carries the metric we predict will move and how we will check it. That sentence is written before you act, not after.

You make the change

You, or your agency, or your developer. We do not write your pages or place links, which is what makes the next step worth anything.

We grade what we predicted

On the next measurement we compare the metric to its own baseline and publish the verdict, including the ones you would rather not read.

The two ways the loop is usually left open

Plan and never check. Tools that rank opportunities and attach a projected return, then never compare the projection to what happened. A forecast nobody validates is a forecast, whatever the page calls it.
Grade your own homework. Systems that propose the change, execute it, measure it and then award it a pass. Every part of that sentence is the same party.

Why ours is worth reading

We do not write your pages, place your links, or touch your site. That refusal costs us the easiest revenue in this category, and it buys the one thing a grade needs: the party marking the work did not do the work. It is the same reason an auditor does not keep your books.

Everything else we refuse to do

What a grade actually looks like

Three accepted changes, three outcomes. The figures below are sample inputs, and the verdicts are not: each one is computed at page load by the same function that grades a customer’s change, so this page cannot say something the product would not.

Rewrote the buying guide so each section answers its question on its own

We said you would see: Cited snippets match your passages verbatim rather than paraphrasing.

Prediction held

Citation share rose 20% of 480 to 36% of 480, beyond what sampling variation explains.

Added sourced statistics to the pricing page

We said you would see: AI answers begin quoting your figures for the target prompts.

Moved, below our floor

Citation share moved 20% of 30 to 27% of 30, but the interval on that change includes zero, so we cannot call it a real move yet.

Published an Organization schema block with sameAs profiles

We said you would see: The entity resolves in knowledge-graph lookups.

Not measurable by us

This indicator lives in Wikidata and the engines' knowledge graphs, which we do not measure. We will not report a result we did not observe.

The verdict nobody else ships

“It moved, and the move was too small for us to call.”

Look at the middle card again. The share went up. On a page selling you software, that is a win, and it would be trivial to print one. It is not a win, because the sample was small enough that ordinary sampling variation produces that move on its own.

The same proportional move on a large enough sample grades as held. Nothing about the change is different; the evidence is. That distinction is the whole product, and a tool that never draws it is telling you every fix worked.

Added sourced statistics to the pricing page

Citation share moved 20% of 30 to 27% of 30, but the interval on that change includes zero, so we cannot call it a real move yet.

The interval arithmetic behind that sentence

The part that only exists once you start

A single grade is a fact. A year of them is an argument: which changes pay on your site, which pillars stall, which advice we got wrong. That record cannot be shown to you now, and not because we are holding it back. It does not exist for anyone until a site has been measured, changed, and measured again.

Illustrative shape only, and it is blurred because it is unknowable rather than withheld. We never hide a finding we have already made.

Start with the measurement, not the invoice

The loop needs a before and an after, so it begins with one measurement of your site. That part is free, takes no account and no card, and returns the whole result including every failure it finds.

Run the free audit Create an account to track it

Tracking, history and grading are on the paid plans. See what each one includes.