Old creative tests split audiences and compared tiny variations. Under Andromeda that produces noise. Here is a creative testing framework built for 2026: what to test, how to structure it, how much to spend and how to decide winners.
Creative testing on Meta used to be a science of small differences: the same image with two headlines, split across matched audiences, until one reached statistical significance. Under Meta Andromeda, that approach struggles. Small variations are treated as the same creative, and because each different ad reaches different people, a clean like-for-like comparison is rare.
The goal of testing has changed. It is no longer to find the one best ad. It is to keep finding new concepts that open up new audiences, and to retire ones that have run their course.
What to test (and what to stop testing)
Test these
- Concepts: new reasons to care, such as a different pain, outcome or proof point.
- Personas: ads aimed at a role or segment you are not reaching yet.
- Formats: a format missing from your mix, like long video or carousel.
- Offers: a demo versus a free trial versus a resource, which changes who converts.
Stop testing these on their own
- Button colours and CTA labels.
- Small headline wording changes.
- Minor crops and edits of the same video.
Those are still worth iterating on inside a winning concept, but they should not take testing budget. For more on what counts as different, read creative diversification for Meta ads.
Three ways to structure creative tests
Option A: Test inside your main ad set (small budgets)
Add one or two new concepts to the live prospecting ad set every couple of weeks. Simple and keeps all learning in one place. The downside: established winners often hold the budget, so new ads can struggle to get spend.
Option B: A dedicated testing campaign (medium and large budgets)
Run a separate campaign with the same broad targeting and optimisation event, holding new concepts. Give it a fixed share of budget; Meta suggested 20 to 30% for testing in its late-2024 guidance for Advantage+ shopping. Concepts that win move into the main campaign, keeping their post ID so engagement carries over.
Option C: Meta's A/B test tool (big decisions)
For questions where you need a clean answer, such as whether a new offer beats the old one, use Experiments in Ads Manager. It splits audiences so each person sees only one version. Use it sparingly; it needs budget and time.
Budget and timing
Each new concept needs enough spend to show whether it works. A common rule of thumb is two to three times your target cost per result, over at least a week to smooth out day-of-week swings.
Illustration. If your target cost per qualified lead is 60, each concept needs about 120 to 180 before you judge it. With a testing budget of 600 a week, you can test four or five concepts each week. Our article on how many ads per ad set walks through this maths in more detail.
How to decide winners
- Use the right metric. Cost per real outcome: purchase, qualified lead, activated trial. Click-through rate and cost per click help explain results but should not decide them.
- Look past the first days. Early numbers swing a lot, and Meta's delivery is still exploring.
- Compare against your control, not against zero. A winner beats your current main-campaign average on cost per outcome, or reaches a clearly new audience at a similar cost.
- Check downstream. For B2B, check lead quality in your CRM after a few weeks. A cheap lead that never becomes pipeline is not a win. Our performance marketing system shows how to connect the data.
- Note what you learned. Record the concept, persona, format and result in a simple log. Patterns across tests are worth more than any one winner.
Self-assessment
Is your testing set up to learn?
Five questions about how you test creative today.
0 of 5 answered
A two-week testing rhythm
- Day 1: review last cycle's log and results. Pick two to four new concepts from your motivation and persona map.
- Days 2 to 5: brief and produce. Phone-shot video and statics are fine.
- Day 6: launch into the testing campaign or main ad set.
- Days 7 to 13: leave them alone. Check for broken links or disapprovals only.
- Day 14: judge, promote winners, pause losers, note what you learned, and start the next cycle.
Quick quiz
Creative testing check
Four questions on the framework.
1. Which is the best thing to spend testing budget on?
2. Which metric should decide a winner?
3. A rule of thumb for spend before judging a concept is:
4. When should you use Meta's A/B test (Experiments) tool?
0 of 4 answered
Frequently asked questions
If you spend enough to fund one, yes: it stops established winners from starving new ads. Smaller accounts can test inside the main ad set.
At least a week, and until each concept has spent around two to three times your target cost per result.
The new ad set goes through learning again, but if you use the same post ID, the ad keeps its likes, comments and shares.
It helps optimise combinations within a concept. It is not a good tool for comparing concepts, because it does not report cleanly which combination drove results.
