Creative testing workflow
A workable creative testing loop is weekly, not quarterly: ship a set of 8 to 20 variations against your current winner, read hook rate first and conversion last, kill the bottom half, and refill from the library. The constraint most teams hit is production throughput, which is the constraint a generation tool removes.
Last updated
How to do it in Genyad
- Monday: brief one campaign around a single argument you have not tested.
- Monday: generate 8 to 20 variations and reject anything whose angle is wrong.
- Tuesday: ship the set into one ad set alongside the incumbent winner. Do not split into 20 ad sets.
- Thursday: read hook rate. Kill anything below half the set median. It will not recover.
- Following Monday: read completion and conversion on survivors, keep the top two, refill the set.
What good looks like
| Metric | Answers | Read it at |
|---|---|---|
| Hook rate (3s views / impressions) | Did the opening earn attention? | 48 to 72 hours |
| Completion rate | Did the argument hold? | 3 to 5 days |
| Click-through rate | Did the offer land? | 3 to 5 days |
| Cost per result | Is it worth scaling? | 7 days or one conversion cycle |
| Frequency | Is fatigue arriving? | Continuously |
The order matters. Judging a hook on cost per result confuses two different questions and usually kills the wrong ad. Decay curves by placement are in the 2026 fatigue benchmark.
Mistakes that cost you the test
- Testing everything at once. If the hook, offer and format all changed, no result is attributable.
- Waiting for significance on every variation. You are not publishing a paper. Kill the obvious losers early and spend the budget on the survivors.
- Refilling only after performance drops. The drop is the cost. Refill on schedule.
- Keeping only the winner. Keep two, because the winner fatigues and you want the runner-up warm.
Frequently asked questions
How often should I refresh ad creative?
Weekly for TikTok and every two to three weeks for Meta in an active campaign. The exact point depends on audience size and frequency, and there are decay curves by placement in our 2026 fatigue benchmark.
How many variations should be in a test set?
8 to 20 live at once. Fewer gives automated allocation too little to work with, more usually means near-duplicates that fatigue together.
Should I test in separate ad sets?
Usually not. One ad set with the whole creative set lets the platform allocate, which is what its optimisation is for. Separate ad sets split your data and your budget.
What is the first metric to read?
Hook rate. If the opening does not earn attention, nothing downstream is interpretable.