How does creative testing work on Meta?

Every creative test asks which ad does better. The method decides how fair the comparison is.

The first way is informal. You launch several ads in one ad set, let Meta deliver them, and compare the results. It needs no setup, and it mirrors how ads run once a test is over. The second way is controlled. A Meta tool gives each version its own share of people and budget, so the ads do not compete for delivery and the gap between them means more.

Meta warns against one informal habit in particular, which is switching ad sets or campaigns on and off by hand to compare them. Its help pages say this can lead to overlapping audiences and unreliable results, while its A/B test keeps audiences evenly split and statistically comparable.

Ads in the same ad set share its audience, placements, budget and schedule. The delivery system then decides which ad to show each time. Meta's help pages say the system learns over time which ads perform best, so some ads are delivered more than others, and that an ad earning more engagement runs more often to get the best results for the budget.

That turns an ad set into a contest the system judges as it goes. It forms a view as results come in and concentrates spend on the ads it favours. An ad that gets little spend has not lost a fair test. It has lost the system's prediction, which may rest on very little data. An ad that wins has shown it performs well under the conditions it got, and those conditions included most of the budget.

This method still answers a useful question, namely which ad the system chooses to spend on. For many accounts that is the question that matters, because it is how ads run after the test. It cannot tell you whether a low-spend ad would have done well with equal delivery.

Meta's A/B test compares versions of an ad strategy that differ in one variable, such as the creative, the audience or the placement. Meta shows each version to a separate segment of the audience, makes sure nobody sees both, and then picks a winner.

You can create one from the Ads Manager toolbar by duplicating an existing campaign, ad set or ad and changing one thing, or you can compare existing campaigns or ad sets in the Experiments tool. Duplicating copies the budget, while comparing two existing items divides the budget you already have. Meta recommends the same budget for each version and one variable per test. It suggests a minimum of seven days, some of its help pages suggest two weeks, and a test can run for up to 30 days. A test can include up to five versions.

At the end, Meta names the version with the lowest cost per result for the campaign's objective and reports a confidence percentage, its estimate of the chance of getting the same winner if the test ran again. The article on statistical significance in ad tests explains how to read that number.

Meta added a second controlled option in 2025. Its help centre documents it as setting up a creative test in Meta Ads Manager, and it lives in the ad editor rather than in Experiments. Jon Loomer, a Meta ads consultant who documented the feature in October 2025, describes how it works:

• You start from an existing or draft ad and create two to five test ads, each a duplicate of it that you then change. The original ad is not part of the test.

• You set aside part of the campaign or ad set budget for the test. Meta recommends no more than 20%, so the existing ads keep most of the spend.

• Each person sees only one of the test ads, and Meta aims to spend similar amounts on each.

• The campaign must use the highest volume bid strategy.

• You pick the metric that decides the test, such as cost per result, and view results in Experiments.

This design answers a narrower question than a full A/B test. It shows which of the new ads performs best when each gets a fair share, while the rest of the campaign carries on as normal. It suits teams that want to test new creative without building a separate campaign. Later reports say some accounts now allow up to ten test ads, so check what your account shows.

Each method trades setup effort against how fair the comparison is and which question it answers.

Dynamic creative and Advantage+ creative enhancements are left off this list on purpose. They mix components or variations and report combined results. Meta says that because you only see aggregate performance, it does not recommend dynamic creative as a substitute for split testing.

A controlled test isolates one change. If the only difference between two ads is the hook, a reliable gap points to the hook. If they differ in image, copy and offer at once, you learn which ad won but not why. Meta's advice is to start with a hypothesis and test one variable at a time.

A test also only covers the people, placements and dates it ran on. A winner from a two-week test in spring may not hold during a December sale. A winner also does not stay a winner. Once it runs at scale, its frequency rises and it starts to fatigue.

The last limit is volume. Small tests produce noisy results, and the budget a test needs depends on how many purchases or leads each version must gather. The guides on running a creative test and setting a testing budget turn this into steps.