A creative testing method that gives you an answer
Most creative tests fail because they ask too many questions at once. Change five things, and when an ad wins you don't know why. This method keeps each test small enough to learn from.
The method in five steps
Pick one question
Which hook works best? Which angle? Which presenter? One question per test.
Change one thing
Make versions that differ only in that one thing. Keep the product, offer and length the same.
Give it time and budget
Let each version spend enough to get real results before you judge it.
Read the right numbers
Hook rate and hold rate tell you about attention. Cost per sale tells you about money.
Scale and repeat
Put budget behind the winner, cut the losers, and start the next question.
What to test, in order
Not every change matters equally. Test the big things first.
- The angle: the reason to buy. A discount, a problem solved, a review, a demo, a joke. This usually makes the biggest difference.
- The hook: the first second or two. Same ad, different opening line or first shot.
- The format: UGC-style, product demo, avatar presenter, meme.
- The details: captions, music, voice, length. Smaller wins, best tested once the bigger questions are settled.
Our process works the same way. We read your live ads and competitors' longest-running ads to choose the angles, make every angle, then test, cut the losers and put budget behind the winner.
Reading the numbers without fooling yourself
Attention numbers
Hook rate is the share of people who watch past the first few seconds. Hold rate is the share who keep watching. They tell you whether the ad grabs and keeps attention, which shows up quickly.
Money numbers
CPA, cost per acquisition, is what you pay in ads for each sale. ROAS, return on ad spend, is the revenue you get back for each dollar spent. These are what decide the winner, but they need more sales before they mean anything.
Common traps
- Calling a winner after a day. Early results swing a lot, especially during Meta's learning phase.
- Judging on likes or cheap clicks. An ad can be popular and still not sell.
- Comparing ads that ran at different times, such as one during a sale and one after.
- Testing tiny changes before the big ones. A new caption font won't rescue the wrong angle.
Meta and TikTok differences
On Meta, the system takes some time to learn who to show new ads to. Group test ads so each one gets a fair share of spend, and wait for results to settle before you judge.
On TikTok, in our experience, ads tend to tire faster, so the test cycle is shorter and you need new versions more often. A hook that worked last month may need replacing this month.
Performance analytics shows both platforms side by side, ranked by spend, CPA and ROAS, and by hook and format. It reads results only and never changes a campaign.
Making enough versions to test
Testing needs volume, and volume is where most brands get stuck. Batch variations make several versions of one ad with different hooks. Clone ads rebuild a proven structure around your product. In Claude, our connector can make Instant ads and new hook versions from a prompt.
If you'd rather hand it over, our team can make the ads for you, across offer, UGC-style, demo, meme and testimonial angles.
Questions, answered.
How many versions should a test have?
Enough to answer one question, and few enough that each version gets real spend. A handful of hooks on one ad is a common starting point.
How long should a creative test run?
Until each version has spent enough to produce a meaningful number of sales. Judging after one day usually misleads.
Should I test on Meta and TikTok at the same time?
You can, but read them separately. The same idea can win on one and lose on the other.