What Is a Variation in A/B Testing? Definition

First published Jun 11, 2019Updated August 21, 20268 min read

Founder & CEO, Omniconvert · Author, The CLV Revolution

Published: Jun 11, 2019Updated: Aug 21, 2026

Reviewed by Cristina Stefanova, Head of Content

Quick Answer

A variation is an alternative version of a page or element that you test against the control, the unchanged original. In an A/B test the control (A) is your current page as it stands, and the variation (B) is the same page with a deliberate change made to it. You split live traffic so some visitors see the control and others see the variation, at random and at the same time, then measure a goal metric for each. Any difference in results can be credited to the change in the variation, so the variation is the challenger and the control is the baseline it has to beat. The discipline that keeps a test honest is changing as little as possible per variation, ideally one thing, or one coherent idea, so you know what caused any difference; a variation that changes ten things wins or loses without telling you why. An A/B/n test runs several variations against one control at once, but each needs its own sample to reach significance, so more variations mean more traffic and a longer test. Omniconvert Explore builds and serves variations, keeps each visitor in one variation, and reports uplift at significance, averaging 23.2% across 70,000+ experiments.

Key Takeaways

A variation is an alternative version of a page or element that you test against the control, the unchanged original; the variation (B) is the challenger and the control (A) is the baseline it has to beat.
Both run at the same time on split, random halves of live traffic, so any difference in results can be credited to the change in the variation rather than to chance, timing, or seasonality.
Change one thing, or one coherent idea, per variation so the result is attributable; a variation that changes ten things at once wins or loses without telling you which change did it.
The exception is a deliberate full-redesign test, a whole new page against the old one; it is a legitimate all-or-nothing bet but does not give element-level insight.
An A/B/n test runs several variations against one control at once, but each needs its own sample to reach significance, so more variations mean more traffic and a longer test.

7,000+ websites 15+ industries 70,000+ experiments 23.2% avg uplift

In A/B testing, a "variation" is the challenger, an alternative version of a page or element that you test against the control, the unchanged original. The word sounds trivial, but the discipline behind a good variation is where most of the value in testing lives: a variation that changes one thing tells you exactly what caused a win or a loss, while a variation that changes ten things at once moves the numbers without ever telling you why. This guide explains what a variation is, how it differs from the control, why you isolate a single variable, how A/B/n tests run several variations at once, and how to build a good variation, drawing on the experimentation practice behind 70,000+ experiments across 7,000+ websites in 15+ industries [CROBenchmark Report 2026, Omniconvert].

The core idea: a variation is only as useful as it is disciplined. Keep it clean, keep the control running, and let statistical significance, not a good-looking first day, decide the winner.

What is a variation?

A variation is an alternative version of a page or element that you test against the control, the unchanged original. In an A/B test the control (A) is your current page as it stands, and the variation (B) is the same page with a deliberate change applied, a new headline, a clearer call to action, a shorter form. You split live traffic so some visitors see the control and others see the variation, at random and at the same time, then measure a goal metric for each. Because the groups are similar and shop under the same conditions, any difference can be credited to the change in the variation, not to chance or timing. The variation is the challenger; the control is the baseline it has to beat.

You split your live traffic so that some visitors see the control and others see the variation, at the same time and at random, then measure a goal metric such as conversion rate for each group. Because the two groups are similar and shop under the same conditions, any difference in results can be credited to the change in the variation rather than to chance or timing. Put simply, the variation is the challenger and the control is the baseline it has to beat. If the variation wins at statistical significance, it becomes the new page; if it loses, you keep the control. To see why that split works, it helps to place the two side by side.

Control vs variation

The control is the unchanged original; the variation is the changed challenger. The control is your current page exactly as visitors experience it, left untouched as the baseline everything is measured against. The variation is the same page with one deliberate change, the idea you want to prove or disprove. Both run at the same time on split, random halves of live traffic, so conditions are identical. The control tells you how the page performs today; the variation tells you how it would perform with your change; the difference, once significant, is your answer. You keep the control running rather than comparing to last week, or seasonality and campaigns would muddy the result.

The distinction is simple but load-bearing: the control is left exactly as it is, and only the variation carries your change. You keep both running at the same time on split traffic rather than switching the page and comparing to a past period, because comparing to last week would mix in seasonality, campaigns, and traffic changes, and you would no longer know whether your change or the calendar moved the numbers. The table sets out how the two differ.

Source: Omniconvert. Control versus variation in an A/B test.
Aspect	Control	Variation
What it is	The unchanged original page, exactly as it stands today	The same page with one deliberate change applied
Role in the test	The baseline; the standard everything is measured against	The challenger; the new idea that has to beat the baseline
What it tells you	How the page performs as it is now	How the page would perform with your change
How many	Always exactly one per test	One or more (several in an A/B/n test)

Read across the rows and the relationship is clear: there is always exactly one control, and the variation, or variations, are measured against it. What turns that comparison into a trustworthy answer is how tightly you control the change itself.

The one-change discipline

A variation should usually change one thing, so you know what caused any difference. If a variation changes the headline, the button, and the layout at once and wins, you have learned the bundle beat the control, but not which change did the work, or whether one hurt and the others carried it. Isolating a single variable keeps the result attributable: a win means that change worked, a loss means it did not, and either way you carry a clear lesson forward. The exception is a deliberate full-redesign test, a whole new page against the old one. That is legitimate, but it answers a different question, which page is better, not which element made the difference, and is judged as an all-or-nothing bet.

The single most important habit in building a variation is to change as little as possible, ideally one thing, or one coherent idea. The reason is attribution. If your variation changes the headline, the button colour, and the layout all at once and it wins, you have learned that the bundle together beat the control, but you have no idea which of the three did the work, or whether one of them actually hurt and the other two carried it. Isolating a single variable keeps the result readable: a win points clearly at that change, and a loss does too.

There is one honest exception. Sometimes you deliberately pit a completely new page against the old one, a full-redesign test. That is a legitimate thing to do, but it is judged differently. It answers "which whole page is better?", not "which element made the difference", so you go into it as an all-or-nothing bet, knowing you will not get element-level insight from it. Everywhere else, one change per variation is the rule. That same logic shapes what happens when you want to test more than one idea at once.

Testing several variations: A/B/n

An A/B/n test runs more than one variation at the same time: the control plus several challengers (B, C, D...), each a different single-idea alternative, all against the same control simultaneously. It is useful when you have several distinct ideas for one element, say three headlines, and want the strongest without testing them one after another. The catch is traffic: each variation is its own group and needs its own sample to reach significance, so splitting traffic across four versions instead of two gives each a smaller share, and the test needs more total traffic and more time. More variations, more traffic, longer test, which often makes a simple two-way A/B test the practical choice on lower-traffic sites.

An A/B/n test is an A/B test with more than one variation running at the same time: the control plus several challengers, B, C, D, and so on, each a different single-idea alternative to the original, all tested against the same control simultaneously. It is the right shape when you have several distinct ideas for the same element, three different headlines, say, and want to find the strongest rather than testing them one after another.

The catch is traffic. Each variation is its own group and needs its own sample of visitors to reach statistical significance. Splitting your traffic across four versions instead of two means each version receives a smaller share, so the test needs more total traffic and more time to reach a reliable result. The rule of thumb is blunt: more variations, more traffic, longer test. On lower-traffic sites this trade-off often makes a simple two-way A/B test the more practical choice, while high-traffic pages can support several variations at once and answer questions about all of them in a single split test. Whether you run one variation or several, building each one well follows the same steps.

How to build a good variation

Build a good variation by starting from a hypothesis, changing one thing, keeping the control running, setting your metric and sample up front, and running to significance before you judge it. Base it on a testable idea grounded in evidence, a heatmap, a recording, a survey, not a guess. Change one thing, or one coherent idea, so the result is attributable. Keep the control running alongside it on split live traffic so the comparison is like with like. Decide your goal metric, confidence level, and sample size before you start, so you are not tempted to stop early. Run to that sample and to significance, then decide: keep the winner, keep the control if the variation lost, and carry the lesson forward.

A good variation is less about a clever idea and more about a disciplined process. Follow these five steps and each test, win or lose, produces a trustworthy answer.

Start from a hypothesis
Base the variation on a clear, testable idea: if we change X, then Y will improve, because Z. Ground it in evidence, a heatmap, a session recording, or a survey, rather than a random guess.
Change one thing (or one coherent idea)
Alter a single element, or one coherent idea, and keep everything else identical to the control, so any difference in results can be attributed to that change and nothing else.
Keep the control running
Run the variation alongside the unchanged control on split, random, live traffic at the same time, so you compare like with like under the same conditions rather than against a past period.
Set metric, confidence, and sample up front
Decide your goal metric, your confidence level (95 percent is a common standard), and the sample size you need before you start, so you are not tempted to stop the moment the numbers look good.
Run to significance, then decide
Let the test reach that sample and statistical significance, then decide: keep the variation if it beat the control, keep the control if it did not, and carry the lesson into the next test.

Notice that nothing here depends on the variation winning. A disciplined variation that loses still tells you something true, because the single change it made points clearly at the cause. That is the whole payoff of testing this way, and it is what a good tool is built to protect.

Building and serving variations with Omniconvert Explore

Omniconvert Explore is an A/B testing and experimentation platform that builds, serves, and measures variations for you. You form a hypothesis with its heatmaps, session recordings, and surveys, then build one or more variations in its visual editor without code. Explore splits live traffic between control and variations and keeps each visitor in the same variation across visits, so the comparison stays clean. It measures conversion rate and revenue per visitor for each version, segments results by audience, and calculates statistical significance so you know when a result is reliable. It supports A/B/n tests, several variations against one control at once, and reports each one's uplift. Across 70,000+ experiments it has averaged a 23.2% conversion uplift.

When you are ready to run a variation, you need a tool that handles the build, the split, and the statistics for you. That is what Omniconvert Explore is built for. You form a hypothesis using its research tools, heatmaps, session recordings, and on-site surveys, then build one or more variations in its visual editor without needing to code. Explore splits your live traffic between the control and the variations and, crucially, keeps each visitor in the same variation across their visits, so a person who saw one version on Monday is not shown a different one on Wednesday and the comparison stays clean.

Explore measures conversion rate and revenue per visitor for each version, segments results by audience so you can see which variation works for which customers, and calculates statistical significance so you know exactly when a result is reliable rather than stopping on a good-looking day. It supports A/B/n tests, several variations against one control at once, and reports the uplift of each. Across more than 70,000 experiments Explore has produced an average 23.2% conversion uplift. Build one variation on one high-traffic page, prove the win on your own traffic, and let the habit compound.

Build a variation, keep the control running, and read the uplift of each version at significance.

See how Omniconvert Explore serves variations →

Frequently Asked Questions

1What is a variation in A/B testing?

A variation is an alternative version of a page or element that you test against the control, which is the unchanged original. In an A/B test, the control (often called A) is your current page exactly as it stands, and the variation (often called B) is the same page with a deliberate change made to it, a different headline, a clearer call to action, a shorter form. You split your live traffic so that some visitors see the control and others see the variation, at the same time and at random, then measure a goal metric such as conversion rate for each group. Because the two groups are similar and shop under the same conditions, any difference in results can be credited to the change in the variation rather than to chance or timing. The variation is the challenger; the control is the baseline it has to beat. If the variation performs better and the result is statistically significant, it becomes the new version of the page; if it performs worse, you keep the control and you have learned what not to ship.

2What is the difference between a variation and the control?

The control is the unchanged original, and the variation is the changed challenger, and the whole point of a test is to compare them. The control is your current page exactly as visitors already experience it; it is the baseline, the standard everything is measured against, and it is left untouched for the duration of the test. The variation is the same page with one deliberate change applied, the new idea you want to prove or disprove. Both run at the same time on split, random halves of your live traffic, so conditions are identical for each. The control tells you how the page performs as it is today; the variation tells you how it would perform with your change. The difference between their results, once it reaches statistical significance, is your answer. You always keep the control running alongside the variation rather than simply switching the page and comparing to last week, because comparing to a past period would mix in seasonality, campaigns, and traffic changes, and you would no longer know whether the change or the calendar moved the numbers.

3Why should a variation change only one thing?

A variation should usually change one thing so that you know what caused any difference in results. If your variation changes the headline, the button colour, and the layout all at once and it wins, you have learned that the bundle of changes together beat the control, but you have no idea which of the three did the work, or whether one of them actually hurt and the other two carried it. Isolating a single variable keeps the result attributable: a win means that specific change worked, and a loss means it did not, and either way you can carry a clear lesson into the next test. The exception is a deliberate full-redesign test, where you intentionally pit a completely new page against the old one. That is a legitimate thing to test, but it answers a different question, which whole page is better, not which element made the difference. It is judged as an all-or-nothing bet rather than as a way to learn what specifically to change, so you go in knowing you will not get element-level insight from it.

4What is an A/B/n test?

An A/B/n test is an A/B test with more than one variation running at the same time: the control plus several challengers (B, C, D, and so on), each a different single-idea alternative to the original, all tested against the same control simultaneously. It is useful when you have several distinct ideas for the same element, for example three different headlines, and want to find the strongest rather than testing them one after another. The catch is traffic. Each variation is its own group and needs its own sample of visitors to reach statistical significance, so splitting your traffic across four versions instead of two means each version receives a smaller share, and the test needs more total traffic and more time to reach a reliable result. More variations, more traffic, longer test. On lower-traffic sites this trade-off often makes a simple two-way A/B test the more practical choice, while high-traffic pages can support several variations at once and get answers about all of them in a single test.

5How do you build a good variation?

You build a good variation by starting from a hypothesis, changing one thing, keeping the control running, deciding your metric and sample up front, and running to significance before you judge it. First, base the variation on a clear, testable idea, if we change X, then Y will improve, because Z, ideally grounded in evidence such as a heatmap, a session recording, or a survey rather than a random guess. Second, change one thing, or one coherent idea, so the result is attributable. Third, keep the control running alongside the variation on split live traffic, so the comparison is like with like under the same conditions. Fourth, set your goal metric, your confidence level (95 percent is a common standard), and the sample size you need before you start, so you are not tempted to stop the moment the numbers look good. Fifth, run the test until it reaches that sample and statistical significance, then decide: keep the variation if it beat the control, keep the control if it did not, and carry the lesson into the next test.

6Does a variation always have to win to be useful?

No, a losing variation is still useful, because a clean test that changed one thing tells you something true whichever way it goes. When a variation beats the control at statistical significance, you have found an improvement worth rolling out and a real uplift. When a variation loses, you have learned that the change you believed in does not work, or actively harms the page, which is genuinely valuable: it stops you shipping a change that would have cost you conversions, and it sharpens your understanding of what your customers respond to. This is only true if the variation was disciplined. Because it changed a single, isolated element, a loss points clearly at that element; a variation that changed ten things at once wins or loses without telling you why, so even its result teaches you little. The value of testing is not that every variation wins, it is that every well-built variation, win or lose, replaces a guess with evidence and makes the next hypothesis smarter.

7How does Omniconvert Explore handle variations?

Omniconvert Explore is an A/B testing and experimentation platform that builds, serves, and measures variations for you. You use its research tools, heatmaps, session recordings, and on-site surveys, to form a hypothesis, then build one or more variations with its visual editor without needing to code. Explore splits your live traffic between the control and the variations and, crucially, keeps each visitor in the same variation across their visits, so a person who saw one version on Monday is not shown a different one on Wednesday and the comparison stays clean. It measures conversion rate and revenue per visitor for each version, segments the results by audience so you can see which variation works for which customers, and calculates statistical significance so you know exactly when a result is reliable rather than stopping on a good-looking day. It supports A/B/n tests, several variations against one control at once, and reports the uplift of each. Across more than 70,000 experiments Explore has produced an average 23.2 percent conversion uplift.

The takeaway

A variation is simply the challenger, an alternative version of a page or element that you test against the control, the unchanged original. Everything that makes a variation trustworthy comes down to discipline. Change one thing, or one coherent idea, so that a win or a loss points clearly at that change and you know what caused the difference; a variation that changes ten things at once moves the numbers without telling you why. Keep the control running alongside it on split, random, simultaneous traffic, so you are comparing like with like rather than against a past period muddied by seasonality and campaigns. If you run several variations at once in an A/B/n test, remember each one needs its own sample to reach significance, so more variations mean more traffic and a longer test. And judge every variation, winner or loser, by statistical significance, not by the first good-looking day. Build variations this way and each test, whichever way it goes, replaces a guess with evidence. That is exactly how Omniconvert Explore serves and reads them.

Valentin Radu

Founder & CEO, Omniconvert

Valentin Radu is the founder and CEO of Omniconvert. He is an entrepreneur, data-driven marketer, CRO expert, CVO evangelist, international speaker, father, husband, and pet guardian. Valentin is also an Instructor at the Customer Value Optimization (CVO) Academy, an educational project that aims to help companies understand and improve Customer Lifetime Value.

Build and test your variations with Omniconvert Explore

Omniconvert Explore lets you build a variation in a visual editor, keep the control running, split your live traffic, and read the uplift of each version at statistical significance, so every variation you ship is a validated win rather than a guess.

Learn more about Omniconvert Explore Book a demo