Split testing, called A/B testing in Ads Manager, is Meta's built-in way to compare versions of your advertising fairly. Meta divides the audience into random groups that don't overlap, shows each group one version, and reports which one won on the metric you chose, with a confidence level. It answers "which is better", not "does advertising work at all", which is a conversion lift study.
Why use Meta's A/B test instead of just running two ads?
When two ads sit in the same ad set, Meta shifts spend toward the one it predicts will win, often within hours. The loser may never get a fair chance. A formal A/B test splits the audience first, so each version reaches comparable people with comparable budgets.
| Two ads in one ad set | Meta A/B test | |
|---|---|---|
| Spend split | Uneven, Meta decides | Controlled |
| Audience overlap | Yes | No, random groups |
| Confidence reading | No | Yes |
| Speed | Fast | Slower, needs a set duration |
| Best for | Day-to-day optimisation | Decisions you'll build on |
What you can test
- Creative: a new hook, angle, format or offer.
- Audience: Advantage+ audience vs a lookalike, for example.
- Placement: automatic vs a manual selection.
- Campaign set-up: such as bid strategy, optimisation event or campaign structure.
Change one thing at a time. Testing a new image and a new headline together tells you which pair won, not why.
How to run one
- Create the versions, or start from an existing campaign and choose A/B test.
- Pick the variable and the key metric, usually cost per purchase.
- Set the budget and schedule. Meta's guidance suggests running tests for at least several days, and around a week or more is common, so weekday and weekend behaviour are both included.
- Don't edit the versions during the test.
- Read the result and its confidence level when the test ends.
Worked example
Example: testing a UGC video against a studio image.
| A: studio image | B: UGC video | |
|---|---|---|
| Spend | ₹25,000 | ₹25,000 |
| Purchases | 50 | 60 |
| Cost per purchase | ₹500 | ₹417 |
B is about 17% cheaper per purchase. If Meta reports high confidence, roll B's approach into more ads. If confidence is low, the difference may be noise, so extend the test or run it again. The conversion rate calculator helps compare rates between versions.
Common mistakes
- Stopping early when one version leads after a day.
- Tests too small to conclude. A handful of purchases per version rarely produces a confident winner.
- Testing trivial changes, such as a button colour, that can't move results enough to detect.
- Not acting on results. Write down what you learned so the next test builds on it.
For a full testing system, read the creative testing framework, and see creative testing for faster, less formal methods.
Frequently asked questions
How do I run an A/B test on Facebook ads?
In Ads Manager, create an A/B test from an existing campaign or from the Experiments area. Choose one variable to change, such as creative or audience, set the key metric, budget and schedule, and don't edit the versions until the test ends.
How long should a Meta A/B test run?
Long enough to collect a meaningful number of results for each version, and usually at least a week so different days are included. Meta's guidance suggests several days at minimum. Ending a test as soon as one version leads is the most common mistake.
What is the difference between an A/B test and a lift test?
An A/B test compares versions of your ads to find which performs better, and every group sees an ad. A lift test compares people who can see your ads with a holdout who can't, to measure whether the ads cause sales at all.
Can I A/B test by putting two ads in one ad set?
You can compare them, but it isn't a fair test. Meta moves spend toward the ad it predicts will win, so the other may not get enough delivery. Use Meta's A/B test feature when the decision matters.
