Our Approach
Measure before assuming. Record the evidence. Say what was observed before saying what it means. Then do the work, measure again, and compare.
Why the order matters
Almost every mistake in this field comes from doing these steps in the wrong order.
The common pattern is to start with a recommendation — add this schema, publish that content, claim this profile — and to measure afterwards, if at all. That produces advice that cannot be wrong, because there was never a baseline to be wrong against. It also makes it impossible to tell which of the ten things you did was the one that mattered.
We do it the other way round. Measure first, so there is a position to improve on. Record the answers, so a later claim can be checked. Only then decide what to do, and only then do it. And afterwards, measure the same questions again, because the point of a baseline is to be compared with something.
It is slower to start and harder to sell than a checklist. It is also the only version of this work that can tell you whether it did anything.
How we work, step by step
Measure before assuming
Nothing starts with an opinion about why a business is not showing up. It starts with putting the questions a buyer would actually ask to the assistants, repeatedly, and seeing what comes back. An assumption that survives a measurement is worth acting on. One that has never been tested is just a guess with a confident tone.
Record the evidence
Every answer is recorded and kept, not summarised into a score and thrown away. That record is what every later figure traces back to, and it is what lets you check a claim we make rather than take it on trust. A number with no recorded answer behind it is not evidence.
Distinguish observation from interpretation
What was measured is stated first, and plainly. What we think it means is labelled as interpretation, and kept separate, so you can accept the first and argue with the second. Most of the confident advice in this field collapses that distinction, which is how a single observation turns into a rule about how AI works.
Identify findings the evidence supports
A finding is a statement the recorded answers will bear — that a business is absent from a set of questions, that it is described from information that is out of date, that the sources being cited are ones it has no presence on. Each finding carries the evidence behind it, and stops where that evidence stops.
Turn findings into actions
Findings are only useful if they say what to do. We set out the actions that follow, in priority order, based on what the evidence shows about this business rather than on a standard checklist. Where the honest answer is that no action is worth taking, that is the recommendation.
Implement where appropriate — with approval
Structured data, page and content changes, corrections where public details are inconsistent. You approve what happens before it happens, and every significant change is recorded with its date and what we expected it to do. We can do the work, or supply it for your own web developer to apply.
Remeasure against the same questions
The same questions go back to the same assistants, unchanged. Keeping the question set constant is what makes two measurements comparable — change the wording and the numbers move for reasons that have nothing to do with the business.
Compare against the baseline
The new measurement is set against the first one: what moved, what did not, and by how much. Visibility also drifts on its own, so a change is reported as an observation alongside the record of what was altered, not as proof that the work caused it.
Report, then repeat
A written record of the measurement, the answers behind it, what was changed and when, and how this cycle compares with the last. Then the cycle runs again, because what assistants can find out about a business keeps changing.
In the customer-facing shorthand used across the site, that is: Understand → Measure → Evidence → Find → Act → Remeasure → Compare → Report. Each stage is described at greater length on How It Works.
What that means in practice
- You see the recorded answers, not only a summary of them.
- Every figure we give you carries a date and the method behind it.
- You approve changes before they go live — nothing happens to your website unannounced.
- You get a written record of what was changed, when, and what we expected it to do.
- You are told when we think the answer is to do nothing.
- You can disagree with our interpretation and still rely on the measurement.
Not everything is measurable, and we say so where it is not. The point of the approach is not to produce certainty. It is to be clear about which parts are evidence, which parts are judgement, and which parts are neither.
The limits of the approach
Being disciplined about evidence does not make the evidence stronger than it is, so it is worth naming what this method still cannot do.
- It cannot establish that a change we made caused a change in what an assistant says. It can show what was changed, when, and what the measurement did afterwards.
- It cannot guarantee an outcome, in three months or in any other period. Our own research found that businesses we changed nothing about still moved between measurements.
- It covers the agreed question set, not the whole internet — and assistants change without notice, so every measurement describes the window it was taken in.
- It measures one business — yours. Our instrument does not currently measure competitors, so no comparison against them is offered.
The standards that hold that line, and the methodology behind our own published research, are set out on Evidence & Standards.
See the approach applied to your business
Tell us what your business does and which buyer questions matter, and we will tell you what measuring it would actually involve.