An A/B test shows some visitors one version of a page and the rest another, then counts which version gets more people to act. It's how you settle a question like "does the shorter headline actually work better?" with evidence instead of opinion.
Setting one up
- Duplicate the page you want to test. The copy is an ordinary page, so you edit it in the builder exactly like any other.
- Change one thing on the copy — the headline, the button wording, the order of sections. Changing several things at once tells you the combination won, not which change did it.
- Under A/B Tests, create a test naming the original as the control and the copy as the variant.
- Pick your traffic split (50/50 by default), what counts as a conversion, and optionally an end date.
- Start it.
What counts as a conversion
Either any form submission on the page, or one specific form — useful when a page carries both a newsletter box and a booking enquiry and only one of them is the point.
How visitors see it
Both versions are served from the same URL — there's no second address to share, and no flicker, because the choice happens on the server before the page is sent. A visitor who is assigned one version keeps seeing that version on later visits. Search engine crawlers are always shown the control and are never counted.
Reading the results
The results screen shows visitors and conversions for each arm, and the conversion rate for each. Counting is per visitor rather than per page load, so the two numbers are directly comparable.
Let it run until each arm has a few hundred visitors before drawing conclusions. A test with twelve visitors will happily tell you something that turns out to be noise.
While a test is running
The traffic split and the goal are locked once a test starts — changing the rules mid-flight would make the numbers meaningless. Stop the test, adjust, and start it again if you need to. Only one test can run on a given page at a time.
Finishing
When you end a test, mark the winner. If the variant won, the simplest thing is to move its content onto the original page so the winning version lives at the original URL.