How do you actually measure if an SEO change worked?
- Published November 6, 2026
- Linked from 3 answers
- In a cluster of 77
Decide the metric, the success threshold and the end date before making the change, record the starting value, change one thing, compare against similar pages you left alone, and judge at the end date you set, once Google has recrawled the page. Watching rankings rise after an edit is not measurement, because rankings move for reasons that have nothing to do with the edit: updates, seasons, competitors and demand.
The steps
- Pick the metric that matches the goal. Clicks for a traffic goal, click-through rate for a title change, impressions for a coverage change, indexing status for a technical fix. Read it for the same queries, device and country, since a page-level average can hide a shift in which queries the page appears for.
- Record the baseline. The metric’s value for the affected pages over the previous weeks, with the date. Search Console’s Performance report keeps 16 months, so you can also compare the same period a year earlier for seasonal pages.
- Change one thing. A new title and new content in the same week leave no way to tell which did what.
- Choose a control group. Similar pages you do not change, watched over the same dates.
- Confirm the change was seen. URL Inspection shows when Google last crawled each changed page.
- Wait for the end date you set. Fix the length in advance, in whole weeks counted from the recrawl; ending the test when the numbers happen to look good, or extending it until they do, builds the result you wanted into the measurement.
- Compare. The change in the edited pages, minus the change in the control group, is your estimate of the effect, not proof of it: unless the control pages were picked at random, differences in their queries, competitors or seasonality can account for part of the gap.
Reading the comparison
| What you see | What it suggests |
|---|---|
| Edited pages rise, control holds | The change likely caused the rise |
| Both rise together | Something site-wide or external, such as an update or seasonal demand |
| Both fall together | The same, in the other direction; the change may still have helped relative to the control |
| Edited pages fall, control holds | The change likely hurt |
The control group is what turns a before-and-after into evidence, though still not proof. Without it, a change made during a core update or a seasonal shift measures the shift, not the change.
When the result is unclear
Pages with little traffic may never show a difference bigger than their normal week-to-week swing. Record that as no result, and either repeat the change across more pages of the same kind or make the call for reasons other than the numbers. Test length is its own decision; settle it before the change ships, with how long a test should run before you trust the result. If several changes went live together, how to isolate which of three simultaneous changes moved rankings applies, and if gathering more data would delay the decision too long, weigh that with when more data is not worth the time to gather. Write the measurement plan before the change, not after the numbers have moved.
Add a comment