Did that SEO change actually work?

Most SEO changes are judged by someone opening Search Console a few weeks later and eyeballing a chart. An experiment writes the change down with its date, picks matching periods either side of it, and compares them on the same terms every day until you stop it.

What can an experiment cover?

Each experiment has one of three scopes:

• A single page, for a change to one URL, such as a rewritten title tag.
• A group of up to 50 pages, for a change rolled out to a set of URLs, such as a new template for every category page.
• The whole property, for site-wide work such as a redesign, new navigation or a domain migration.

You also record what kind of change it was (title, meta description, content, heading, schema, internal links, site redesign, navigation, technical, domain migration or other), and optionally a hypothesis and notes. Writing the hypothesis down before the numbers arrive is the part people skip, and the part that keeps the result honest.

How are the before and after periods chosen?

The after period runs from the day the change went live to its end date, or to the latest complete day of Search Console data if that comes first. Search Console runs about two days behind, so an open experiment never counts days Google hasn't finished filing. The before period is the same number of days, ending the day before the change. Equal lengths mean a 30-day experiment is compared with the 30 days that came before it, not with whatever window happens to be on screen.

For both periods Copilot SEO works out:

• average daily clicks
• average daily impressions
• average position, weighted by impressions
• click-through rate across the period (total clicks over total impressions)

The experiment page charts the daily figures across both periods, with your annotations and known Google algorithm updates drawn on the same chart.

When do results appear?

Running experiments are analysed every day after the Search Console sync. For the first few days the page says it is still building a baseline, then that results are settling, because a week of data swings with weekends and one-off spikes. When an experiment with an end date passes it, it is marked completed.

How do you know the result is real?

Two checks sit on top of the raw comparison.

Data quality

A page on 3 clicks a day can double or halve by chance. So until both periods have enough days and enough clicks behind them, the experiment tells you it can't say yet rather than showing you a confident arrow.

Is it real yet?

Copilot SEO tells you whether the change in click-through rate is clear enough to trust, or still too close to call. Until it is clear, it says so and asks you to keep gathering data, rather than calling a win early. A small rise in clicks is treated as noise, not a win.

What it can't tell you

This is a before and after comparison, not a controlled split test. Seasonality, a competitor's launch or a Google update can move the numbers in the same window as your change. That is why the chart shows algorithm updates and your own annotations: if a core update landed the same week, you'll see it next to the line rather than finding out later. The verdict is about click-through rate, so treat a change in position or impressions as a direction, not a proof.

Where experiments usually start

The quickest experiments to read are snippet changes on pages that already get impressions. The CTR opportunities list finds those: pages ranking well with a click-through rate below what their position should earn. Rewrite the title, record the experiment, and the CTR test above tells you within a few weeks whether it worked.

Refreshing a page flagged by content decay is the other common one. Record the refresh as a content change so you can see whether the traffic came back, or whether the decline was demand that left for good. Consolidating pages after a keyword cannibalisation check works the same way, as a URL group experiment over the pages involved.