Enough creative to make a test mean something
Three creatives is not a test. Produce the variations a real read needs, with one thing changed at a time so the result tells you something you can use.
Change one thing at a time
The reason most creative tests do not resolve is that every variation changes everything. Different hook, different edit, different offer, different length — and when one wins you have learned nothing that transfers.
The variables worth isolating
- The hook: first line, first frame, first sound.
- The format: problem-solution against listicle against testimonial.
- The offer framing: discount, bundle, guarantee, urgency.
- The length: the same cut at 15, 30 and 45 seconds.
- The presenter: who delivers it.
Batch mode produces the grid, so the test has structure instead of being a pile of assets.
What a structured batch looks like
Take one concept that already works — say a problem-solution cut for a skincare product. A structured batch produces four versions where exactly one thing differs: the opening line. Everything else is held constant, down to the music and the caption style.
Run those four against the same audience with the same budget. If one returns a materially better cost per result, you have learned something about openings that applies to the next ten ads you make. Run four ads that differ in every respect and you have learned that one of them won.
The second batch then changes a different variable — length, or where the offer lands — while holding the winning hook fixed. Two rounds in, you are not guessing about either.
The variables worth isolating, in order
The hook
First line, first frame, first sound. Cheapest to change, biggest effect, so always test it first.
The format
Problem-solution against listicle against testimonial. Changes who the ad speaks to, not just how.
The offer framing
Discount, bundle, guarantee, urgency. Often bigger than any creative change.
The length
The same cut at 15, 30 and 45 seconds. Compare hold rate, not just completions.
The presenter
Who delivers it, and whether anyone delivers it at all.
The ratio
9:16 against 4:5 in the same placement, which is a delivery test rather than a creative one.
Questions, answered.
How many variations can I generate at once?
There is no fixed cap. A batch is limited only by the credits on your plan.
How many creatives does a test actually need?
Enough that no single creative carries the read, and few enough that each gets meaningful delivery. In practice that is a function of your conversion volume: if an ad set cannot reach roughly 50 conversions inside the learning window, splitting it across eight creatives means eight sets of noise.
Should I kill the losers straight away?
Not before the ad set has left the learning phase. Results during calibration are unstable, and killing early is the most common way good creative gets thrown out.
Does running many variations at once cannibalise delivery?
It can. The platform will concentrate delivery on whichever creative looks best early, which is why a batch of twenty is usually worse than a batch of four followed by another four.