What to measure, and what to ignore

Six lines, one page

The design principle is that a measure exists to change a decision. A measure that has never changed one is a number being collected out of habit, and collecting it costs attention that the useful measures need.

Twenty minutes, on the same day each month. The discipline is the date rather than the depth: a shallow review done every month is worth several times a deep one done twice a year, because it is the sequence of readings that carries the information.

What each line tells you

Note what is absent: revenue, which belongs in the accounts rather than in a marketing review; social media figures, which for a clinic rarely relate to bookings; and anything expressed as a score.

Three measures to ignore

Average position. It averages across every query a page appeared for, including queries you do not want, and it moves when the mix changes rather than when your page does. A page can improve for every query it cares about while its average position falls.

Bounce rate. A visitor who arrives at an answer page, gets the answer and leaves has been served perfectly, and is recorded as a bounce. For a publication of answers, a high figure is frequently a sign of success.

Impressions on their own. Useful as context, misleading as a target. Impressions rise when you appear for more queries, including irrelevant ones, and a rise with no change in clicks usually means you have become more visible for things you cannot serve.

Noise and signal

Clinic numbers are small, and small numbers move about. Eleven enquiries following nine enquiries is not a twenty-two per cent improvement; it is two enquiries and it may well be weather.

Two habits handle this. Compare the same month across years rather than consecutive months where seasonality is strong, which in aesthetics it is. And require a change to persist for three readings before treating it as real, unless it is large enough to be unmistakable.

Most of what a clinic responds to in month-on-month figures is noise, and most of the responses make things worse by changing pages that were working.

When something really drops

A genuine, sustained drop has a small number of causes and they are distinguishable. Google's guidance on debugging drops in search traffic separates them methodically: a technical problem, a manual action, a change you made, a change to the results page itself, seasonality, or a shift in what people are searching for.

Work through them in that order, because the first three are yours to fix and the last three are not. In practice, most clinic drops trace to something in Part V of this course: a migration without a complete map, a template change that introduced a canonical fault, or a noindex left behind.

Record the date of every change you make to the site, in the same document as the review sheet. Without that record, diagnosis becomes archaeology.

The annual questions

Once a year, four questions that the monthly sheet cannot answer.

1. Which pages produced consultations, and are they the pages you expected?
2. Which pages have produced nothing in twelve months, and should they be improved, merged or removed?
3. Is the proposition from Module 06, Writing the clinic's proposition still true, and is it still on the pages?
4. Has the work outgrown the hours available, which is the subject of Module 32, When to stop doing it yourself, and what to hand over?

The second question is the one clinics avoid. A page that has produced nothing for a year is either wrong for its subject or wrong for the site, and leaving it in place because it took an afternoon to write is the sunk cost fallacy with a URL.