Free grader, no signup
Did my SEO change work?
Compare the 28 days after the change with the 28 days before it, on the one page you changed, against your untouched pages over the same days. Call it a win only if the move clears 10% and the 90% interval excludes zero. Anything weaker is inconclusive, not a win.
Grade one change
The verdict comes from the same functions Growth Radar's scorer runs on every recorded change.
- before the change
- 28days
- after the change
- 28days
- the move has to clear
- ±10%
- interval must exclude zero
- 90%
- Won / Lost
- The interval excludes zero and the move clears ±10%, against matched untouched pages.
- Flat
- The interval rules out any move of ±10% or more.
- Inconclusive
- No control, too few pages, an interval that includes zero, or an underpowered test.
- Too early / No data
- The after window isn't 28 days old yet, or one side has nothing to measure.
How to grade a change, step by step
Export the daily series
In Search Console open Performance, filter to the changed page, set a range covering 28 days either side of the change, export, and open Dates.csv.
Paste it and set the change date
Paste the file or two columns of date and value, and enter the first day the change was live.
Add the untouched pages
For a page-level verdict, export the Pages tab with Compare on for the same two 28-day ranges and paste it, then enter the changed page's path.
Grade
Press Grade. The verdict, the before and after sums, and the 90% interval come from the same functions Growth Radar's scorer runs.
Why two equal windows around the change
Search traffic has a weekly rhythm, so the before and after windows must hold the same number of whole weeks. Growth Radar sums the 28 days before the change went live and the 28 days from that day on. Shorter windows are noise; longer ones let an unrelated update leak in. Until the after window is complete the answer is "too early", and if one side wasn't measured, it's "no data". Neither is ever counted as a loss or a win.
Why a ±10% bar
A page's clicks wobble by several percent from one month to the next with nothing changed. Treating a small wobble as a win is how a site teaches itself superstitions, so a move under 10% in either direction is never a win or a loss. The bar is a floor, not the test: a 15% move on a page with a handful of clicks still fails, because its interval is too wide to rule out zero.
Why untouched pages, not a plain before and after
Seasons, core updates and a competitor's launch move every page at once. Growth Radar compares the changed page with its 39 most similar untouched pages over the same two windows, and calls a win only when the page beats 95% of them. On Growth Radar's own backtest, a raw before and after on one page read "win" in 39% of cases where nothing had changed, and in only 47% of cases with a real +30% change. That's close to a coin flip, so this grader marks it inconclusive.
What the 90% interval means
The interval is the range of effects the data can't rule out at 90% confidence. If it includes zero, the change might have done nothing, and the verdict is inconclusive however large the point estimate looks. If it sits entirely inside ±10%, that is a finding too: the change didn't move the page enough to matter. A test that would catch a real +30% change less than 50% of the time is called underpowered, and it decides nothing.
Split testing on a small site
Classic SEO split testing assigns hundreds of similar pages to two groups, which a small site doesn't have. Comparing one changed page with its own untouched pages works with a few dozen pages, provided at least 19 of them had search traffic in both windows. With fewer, this grader says so and refuses to call a winner. Most single changes on small sites end up inconclusive. That's the honest reading, and it beats a dashboard arrow pointing up.
Does the same method work for GEO?
Yes, if the outcome is measured the same way. For AI answers Growth Radar asks the same questions before and after the change and grades the share of answers that cite the site, against questions the change did not target, with the same ±10 point bar and 90% interval. This calculator covers the Search Console side: clicks, impressions, click-through rate and position. The citation side needs repeated probes of AI assistants, which a paste can't provide.
Questions
How do I measure if an SEO change worked?
Sum the target metric over the 28 days before the change and the 28 days after, on the changed page only, and compare the move with the same move on pages you didn't touch. A win needs at least 10% and a 90% interval that excludes zero.
How long should I wait before judging an SEO change?
28 full days after the change went live, plus the two or three days Search Console takes to settle its numbers. Before that the honest answer is "too early".
Is there an SEO split testing tool for small sites?
This grader is one: it uses your own untouched pages as the control instead of a randomised page group, and needs at least 19 of them with traffic in both windows.
Why does my change say inconclusive when clicks went up?
Either there were no untouched pages to compare with, or the 90% interval still includes zero, or the test was too weak to catch a real change. If every other page rose too, the change didn't cause it.
Is my data uploaded?
No. The grader runs in your browser; the pasted data isn't sent anywhere or stored.
Want this run on every change automatically? How Growth Radar works