Did your SEO change work?
Give it daily clicks for the pages you changed and for similar pages you left alone. It works out what the changed pages would have done without the change, then tells you the lift, how sure it is, and whether to keep waiting.
Try an example:
1 Your data
Drop a CSV here
or click to choose. One row per day with test and control columns, or one row per day per group.
Or paste the data
Test pages (the ones you changed)
Drop test CSV
e.g. a Search Console Dates.csv filtered to those pages
Control pages (left alone)
Drop control CSV
same date range, same metric
What data works?
- One row per day with a date column and at least one column for the test pages and one for the control pages, e.g.
date,test_clicks,control_clicks. - One row per day per group, e.g.
date,group,clickswith group values liketestandcontrol. Page-level rows work too: pick which values count as test. - Two files, one for each group, such as two Search Console performance exports (the Dates table) filtered to each page group.
- Dates like 2026-08-03, 08/03/2026, 03.08.2026 or Aug 3, 2026. Missing days are skipped. Duplicated dates are added up.
- Aim for at least four weeks before the change and two weeks after it.
2 The change
Model options
Lift vs expected
Extra clicks
Chance it is positive
Days counted
What happened vs what was expected
Daily difference
Running total of the difference
Show the numbers as a table
Before and after, side by side
A simple cross-check that does not use the model: the change in the test group minus the change in the control group.
Placebo checks
The same analysis at fake change dates before the real one, where nothing changed. They should come out flat.
How good is the baseline?
What it can detect
How it works
- Learn the relationship. Before the change, it fits how the test pages track the control pages day by day, with a weekday pattern and day-to-day carry-over in the errors. The fit starts from the assumption that both groups move in proportion and lets the data adjust it.
- Forecast the counterfactual. After the change it uses the control pages to forecast what the test pages would have done if nothing had changed, 1,000 times with the uncertainty included.
- Compare. Actual minus expected, day by day and in total, gives the lift and its 95% interval. The share of forecasts below the actual total is the chance the effect is positive.
- Check itself. A difference-in-differences comparison and placebo runs at fake dates show whether the result holds up.
Pick control pages that are similar to the test pages and were not changed. A sitewide update, a seasonal spike or a migration that hits only one group will look like an effect.