Decide what to measure by working backward from the outcome the change is meant to produce, then lock that in before you touch anything. The method has four parts. First, define the specific metric the change should move, tied to the real goal rather than a vanity number. If the change is meant to win more qualified visits, measure that, not raw impressions or a position that looks good but converts nothing. The metric has to be the thing you actually care about moving, because that is the only thing whose change will tell you whether the work mattered.
Second, record the baseline before you change anything. A metric means nothing without the starting value to compare against, and you cannot reconstruct the “before” once you have altered the page, so the baseline has to be captured up front. Third, identify a control if you can, something comparable that you are not changing, so that if both move together you know it was the wider environment, not your change. A control is what separates “my change worked” from “everything moved that week,” and without one you are often crediting your work for an update or a seasonal swing.
Fourth, set the window and the success threshold in advance. Decide how long you will wait before you read the result and what level of movement counts as success, so you are not staring at noise on day three or moving the goalposts after the fact. The discipline is define-the-target-metric-and-baseline-first, and it is distinct from measuring the result afterward. This guards against the common habit of making the change and then casting around for whatever number happened to move, which lets you tell a flattering story from whichever metric cooperated rather than learning what the change actually did.
To apply the method, start from the intended outcome and name the single metric that captures it. Write down its baseline, pick a control if one exists, and fix the window and the threshold for success, all before you make the change. Then make the change and read that metric against its baseline over the window you set. Define what success looks like first, and the measurement afterward becomes a clean answer instead of a guess.