SEO tests

You changed something on some of your pages, such as adding FAQ markup, rewriting titles or adding internal links. Did it work? Traffic moves for many reasons (seasons, news, Google updates), so looking at the pages alone can mislead you. An SEO test compares the pages you changed with similar pages you did not change, so the ups and downs they share cancel out.

Open a site, choose the Tests tab and pick New test.

Setting up a test

1. Name it, and optionally write your hypothesis (what you expect and why).
2. Set the change date: the first day the change was live. A date in the past works. We measure from the data we already store, so you get an answer straight away.
3. Choose the test pages: the pages you changed. Use a URL rule (starts with, or contains, some text such as /tents/) or paste a list of URLs (up to 50). Rules ignore capital letters.
4. Choose control pages (optional but strongly recommended): similar pages you did not change. Any page that is also a test page is left out of the control group automatically.
5. Pick what to measure (clicks, the default, or impressions) and how many days to compare before and after the change (28 and 28 by default; 7 to 90 are allowed).

Reading the result. Each test shows a verdict:

• Positive: the test pages did better than the control pages say they would have, and we are fairly sure it is not chance.
• Negative: they did worse, and we are fairly sure.
• Inconclusive: the difference is small, or the range of possible answers includes zero, so it could be noise.
• Needs more data: there is not enough data yet (see below).

The big number is the effect: how much more (or less) the test pages gained than they would have if they had moved like the control pages. For example, +16% means the test pages ended up 16% above where the control pages suggest they would have been. Beneath it you see the average per day before and after for each group, the estimated extra clicks per day and in total, and the 95% range: where the true effect probably lies.

The chart shows both groups as a 7-day average, set so that 100 is each group’s average before the change. A dashed line marks the change date. If the lines run together before the change and split apart after it, that is the picture of a real effect.

How we work it out. We add up the clicks of each group for every day. We compare each group’s average after the change with its average before. The effect is the test group’s change measured against the control group’s change (a “difference in differences”). To see how much that answer could wobble, we reshuffle the days 1,000 times, keeping weekdays with weekdays so the weekly rhythm stays, and take the middle 95% of the results as the range. Positive or Negative needs the whole range on one side of zero and a change of at least 2%.

When there is not enough data. A test needs at least 14 days with data before the change and 14 days after it, and at least 100 clicks (or impressions) on each group in the before window. Until then the test says Needs more data and tells you what is missing. While the after window is still filling, the result is marked provisional and updates every day as new data arrives. A test whose change date is in the future waits for its data.

Keep in mind

• With no control pages we can only compare before with after, so seasonality is not removed. The result says so.
• Anything else you changed on the same day counts too. Run one change at a time on a set of pages when you can.
• The range assumes days are independent of each other. Treat the verdict as strong evidence, not proof.

Plans. SEO tests are on Pro and Agency. On Free the Tests tab explains the feature with a quiet “See plans” note. Every member can read tests; workspace owners and admins can create, edit and delete them. A site can have up to 50 tests.