On this page
TLDR:
- Test ideas, not files. Three to five ideas at a time, each with two to three versions (different hooks or visuals), is a load a small team can read. Dovey's daily agent works at that pace.
- Give each creative 3 to 7 days or about 50 results, whichever comes first, before judging it. Under that, the numbers are noise; Dovey's own bar for calling something a pattern is 20 ads.
- Most ads in the Meta Ad Library have one version. The ones that run for months have many: the median long runner in Dovey's index carries several.
- Read cost per result first, click-through second, and always against the account's own average. Dovey never quotes a benchmark CTR, because Meta publishes none.
- Keep the winner running and test against it. Dovey's remix makes the next versions from the winner, never from scratch.
Test ideas, not files
A creative test that compares twelve files with no idea behind them tells you which file won and nothing about why. Group by idea: a problem-led ad, an offer-led ad, a proof-led ad. Inside each idea, two or three versions: a different hook, a different picture, the same everything else. When the problem idea wins, you know what to make more of. When a hook wins inside it, you know the cheapest thing to vary next.
Three to five ideas at a time is the most a small team can make well and read honestly. Ten ideas at once spreads the budget too thin for any of them to get a fair reading.
How long does a creative need?
Long enough for Meta to leave the learning phase and for the result count to mean something:
- 3 to 7 days, or
- about 50 results (purchases, leads, whatever you optimise for),
whichever comes first. Below that, a 30% difference between two ads is usually noise. Two ads at 8 and 11 purchases are tied.
Do not edit a live ad; duplicate it. Editing resets learning and you lose the comparison.
What did the long runners do?
Meta's Ad Library shows how many ads share a creative. Dovey keeps that number. In the index on September 24, 2026, most ads carry a single version, and the ones that have run for months rarely do:
| Ad | Days running | Versions |
|---|---|---|
| Buildify, "Build apps with AI" | 341 | 129 |
| Peak Footwear, "Walk pain free" | 401 | 24 |
| Feelingirl, shaper bodysuit | 280 | 21 |
| Fitbod, form tips | 176 | 18 |
| RatePunk, "Real deals" | 120 | 13 |
| Try Strawberry, brightening cream | 301 | 12 |
| Alevia, "Try Amla" | 169 | 8 to 11 |
Versions are how an idea survives fatigue. The test is not "which of these twelve wins"; it is "which idea wins, and how many ways can we keep it fresh".
What do I read, and against what?
- Cost per result, first. It is the thing you pay.
- Click-through rate, second. It tells you whether the creative stops people, separate from whether the page converts them.
- Frequency, to know when to refresh (see ad fatigue).
Compare against your own account's trailing average, not against a benchmark from a blog post. Categories, prices and audiences differ so much that a "good CTR" from someone else is not information.
What is a routine a small team can keep?
- Monday: two new versions of last week's winner, one new idea against it.
- Thursday: read the numbers. Anything under the account average for two weeks is retired.
- Every month: look at what runs in your category in the Ad Library and take one angle you have not tried.
Dovey makes the creatives for this loop: ask what is working in your category and for new creatives in your brand, delivered to Slack, and you approve the ones you like.
When should I not test at all?
When the budget is under about $30 a day. Below that, nothing reaches 50 results in a week and every "test" is a coin flip. Run your best guess as one ad, let it earn, then test.
FAQ
How many ad creatives should I run at once?
Three to five ideas, each with two to three versions, for a small team. More than that and no creative gets enough budget to be judged.
How long should I run an ad before deciding?
Three to seven days or about 50 results, whichever comes first. Under that, differences of 30% are usually noise.
Should I test one variable at a time?
Inside an idea, yes: change the hook or the picture, not both. Between ideas, everything changes, and that is fine because you are testing the idea.
What is a good CTR for Facebook ads?
The one above your own account's average. Category, price and audience move CTR so much that a general benchmark is not useful for a decision.
See every ad that works. Then make yours.
New creatives from what works for you and your category, in your brand. You pick what runs.
See the testing loop