Free tools

Did the SEO change work? A before/after calculator with a control group

"Clicks went up 30% after we rewrote the titles" means little if the whole site went up 10% that month. This calculator compares the pages you changed with pages you did not touch, and reports the part of the change your edit can claim.

By Adem ÇakalPublished 3 sources

Measure a change

Use the same date ranges for both groups. Leave the control fields empty to see the raw change only. Numbers stay in your browser.

Short answer

Expected clicks = changed pages' clicks before × (control after ÷ control before). The effect of your change is actual clicks after minus that expectation. With the example numbers, the changed pages rose 30%, the control rose 10%, so the edit can claim about 18%, or 200 clicks, not 30%.

Why a raw before/after misleads

Changed pages rise from 1000 to 1300; expected without change is 1100 based on control trend; effect is 200 clicks Before: 1,000 After: 1,300 Expected from trend: 1,100 +200 effect Example numbers. Dashed: what the control group's +10% trend predicts.
The gap between the solid and dashed lines is what your change can claim.

Google's own guide to traffic drops lists the forces that move clicks without any action on your side: algorithm updates, seasonality and changing interest, technical and security issues, and site moves. Any of them can make an edit look better or worse than it was. A control group is the cheapest way to subtract them.

Choosing a good control group

Good controlPoor control
Same page type (other guides, other product pages)The homepage or brand pages
Similar traffic level and position rangePages with a handful of clicks
Not edited, not newly linked during the testPages that got new internal links from the changed ones
Chosen before you look at the resultsPicked afterwards because they happened to fall

The idea is the same one behind Google's open-source CausalImpact package, which estimates the effect of an intervention on a time series using control series that were "themselves not affected by the intervention". This calculator is the simple two-period version; CausalImpact models the whole daily series and gives an uncertainty range.

Reading small effects

The calculator flags effects under 10% as unclear. That is a rule of thumb, not a statistical test: on a few pages with a few hundred clicks, normal week-to-week movement easily exceeds it. Larger groups of changed and control pages, and longer periods, make smaller effects trustworthy.

Getting the numbers from Search Console

  1. Note the date the change went live and wait until URL Inspection shows the changed pages were recrawled.
  2. In Performance, filter Page with "URLs containing" or a regex that matches the changed pages. Set two equal date ranges with Compare.
  3. Read total clicks for each range. Repeat with a filter for the control pages.
# Search Console page filter, Custom (regex): changed pages
/guides/(invoice-template|quote-template|receipt-template)

# control pages: other guides you did not touch
/guides/(expense-report|timesheet|purchase-order)

This is the comparison SiteTiller is built to make for every version of the pages it creates: the edited page against the rest of the site over the same period, with losing versions rolled back.

Questions people ask

What is a control group in SEO?

A set of pages you did not change, similar in type and traffic to the ones you did. They experience the same seasonality, Google updates and brand campaigns, so their trend estimates what would have happened to your changed pages without the edit.

How long should the before and after periods be?

Equal length, long enough to cover full weeks, and starting after Google has recrawled the changed pages. Four weeks each is a reasonable default for pages with steady traffic; low-traffic pages need longer.

What if my result is +5%?

Treat small effects as unproven. Week-to-week noise on a few pages easily reaches several percent. Repeat the change on another set of pages, or run it longer, before you call it a win.

Is this the same as an A/B test?

Not quite. In a true test you assign pages to groups at random before the change. Here you pick the control after the fact, so choose pages that are as similar as possible and do not cherry-pick ones that happened to fall.

Sources

  1. Google: CausalImpact (Brodersen et al., Annals of Applied Statistics, 2015), accessed 2026-11-12.
  2. Google Search Central: Debug Google Search traffic drops, accessed 2026-11-12.
  3. Search Console Help: Performance report (comparing date ranges), accessed 2026-11-12.