An A/B test compares two complete versions of one page at a single public address. Visitors are split in half, and you keep the version that converts better.
Before you start
The page must be published. If it isn’t, the test screen tells you so and offers nothing else.
Creating a test
In the page list, open the three-dot menu next to a published page and choose to create an A/B test. The test screen for that page opens, where you choose one of two starting points:
- Start from this page. Variant B is created as an editable copy of the current page.
- Test two existing pages. Variant B is another published page, which you choose from a list.
Then click to create the test. The test starts as a draft and nothing is split yet.
Setup
The test settings card holds:
- The test title.
- The winning metric: clicks, sign-ups, conversions or revenue (GMV). If the page has no form or checkout, choose clicks.
- The test duration in hours, from 1 to 720 hours (30 days).
- The fallback variant, served to the remaining visitors if the result stays inconclusive.
- The traffic split, shown as A 50% and B 50%. It is fixed and can’t be changed.
The variants card links to editing each version, so edit B until it genuinely differs from A.
Starting
Click to start the test. At that point the settings are locked. While the test is still a draft, you can click cancel instead, which deletes the test and, if B was a clone, the clone too. A page chosen as an existing variant B is left untouched.
Reading the result
The results card shows views, clicks, sign-ups, conversions and GMV for each variant plus a total row, with your metric highlighted. A banner above it says that one variant is leading, with a confidence percentage, or that it is too close to call and worth collecting more data. The chart below switches between views, conversions and GMV. The numbers are approximate, because cookie blockers hide some visitors.
Finishing
Click to declare a winner and choose Auto (metric leader), A or B. The content of the chosen version is copied to the live page, so 100% of traffic gets the winner, the test moves to the finished state and the decision is frozen. This can’t be undone. Choosing Auto while the result is still inconclusive is refused, so pick a side or keep testing.
Test list
Website → A/B tests shows every test with its status (draft, running, finished), the leading or winning variant, the number of views and the dates. If this item isn’t in your menu, A/B tests aren’t part of your current plan.