Good catch! You can't actually, it's only possible to have one version of a page in Google's index at one time.
But we also don't want to be making changes blind, using dated best practices and a finger in the air to guess what will work. So what's the solution?
We've got two:
- Test a single page and compare the results before and after
It's true, this is not strictly speaking a/b testing for seo. But comparing results specific to a period before and after a change is still a lot more useful than asking 'was traffic up this month?'
To make the data more precise we map your before and after periods to the same days in the past: eg we don't compare Tue-Wed against Thursday-Friday. Instead, if you start a test on a Tuesday and end it on a Sunday we compare it against Tuesday to Sunday the previous week. If your test lasts more than a week, we go back 2 weeks for comparison data, and so on. This helps even out differences caused by weekends and holidays periods.
- Group together similar pages - such as products, locations, or reviews - and test on a sample of these
When creating a test you can choose to test pages matching a particular pattern, such as /products/. On these multipage tests you can choose what percentage of pages to test. We will treat the remaining pages as a control group.