In an A/B test, the control is the version you already have and the variant is the changed version you are testing against it. Visitors are split at random between the two, and the variant only wins if it performs measurably better than the control on the goal you chose before the test started.
How control and variant work
The control is your baseline. It might be your current landing page, your existing ad headline or your usual email subject line. The variant, sometimes called the challenger or version B, is identical except for the change you want to test: a different headline, a shorter form, a new main image, a price shown upfront.
A testing tool assigns each visitor to one version and keeps them there on later visits, usually by setting a cookie. It then records how many people in each group complete the goal, whether that is a form submission, a booking or a purchase. When enough people have been through both versions, you compare the two conversion rates and check whether the difference is likely to be real or just chance, which is what statistical significance measures.
A test can have more than one variant (an A/B/n test), but each extra variant splits your traffic further and makes a clear result take longer. Testing several elements at once in different combinations is a multivariate test, which needs far more traffic again.
The same idea runs through the ad platforms. Google Ads experiments split traffic between an original campaign and a trial version, and Meta’s A/B testing tool splits audiences between ad sets. In each case one side is the control and the other the variant.
Why it matters
A control is what turns a change into evidence. If you redesign a landing page and enquiries rise the next month, you cannot tell whether the page caused it or demand rose anyway: a bank holiday, a competitor closing, a spell of weather that sends people looking for a roofer. Running the old and new versions side by side, at the same time and to the same kind of visitor, removes those other explanations.
For UK businesses there is a practical constraint. Most website testing tools set cookies or similar storage, which under PECR generally needs consent in the same way analytics cookies do, so only visitors who accept are in the test. That shrinks the sample, and on a small site it can mean a test needs months to reach a reliable answer.
Common mistakes
- Changing too much in the variant. If the headline, image and form all change, a win tells you nothing about which change mattered.
- Stopping when the variant first looks ahead. Early leads often disappear as more data arrives; this is the peeking problem.
- Editing the control mid-test. Once the test starts, neither version should change.
- Uneven exposure. If one version loads more slowly, breaks on mobile or is left out of some traffic, the comparison is unfair.
- No goal agreed in advance. Picking the metric afterwards, from whichever one happens to favour the variant, is not a test.
How to act on it
Before building a variant, write one sentence: if we change this, we expect that to improve, because of this reason. It keeps the variant to a single idea and fixes the goal before any data arrives.
Next, check you have the traffic. A site with a few hundred visitors a month may never reach a clear result on a small change. In that case, test bolder changes, test in your ads where traffic is easier to buy, or make the change and watch it against the same period last year, accepting that the evidence is weaker.
When the test ends, put the numbers through an A/B test significance calculator rather than judging by eye. If the variant wins, it becomes the new control and the next test starts from there.
Landing pages are the usual place to begin, because they sit between your ad spend and your enquiries. Building landing pages that can be tested properly is part of my PPC landing page design service.