One product, twenty hooks: creative testing at scale
Most ad accounts fail at creative testing for a structural reason rather than a creative one: producing a variation is expensive, so they test four and call it a test. Four variations is not a test, it is a guess with extra steps. Here is how to get to twenty for one product, and how to run them so the result is worth acting on.
Change the hook, hold everything else
A hook test only means something if the hook is the only variable. Same product, same offer, same landing page, same body of the ad, same creator, same length. If you change the hook and the creator together, a winner tells you nothing about either.
This is easier to say than to do with shot footage, because reshooting twenty openings means twenty setups. It is straightforward when the rest of the ad is a record you can re-render against: you keep the creator, the scene and beats two through five, and you replace beat one.
Twenty hooks from six angles
You do not invent twenty hooks. You take six angles and write three or four variants of each.
1. The problem, named precisely. Not "tired all the time" but "you wake up at 4am and can't get back down". Specificity is what makes a viewer feel addressed. Write four versions, each naming a different symptom.
2. The number. Lead with a figure that is surprising and checkable. "Most people guess two hours. It's nine." Vary which number you lead with.
3. The mistake. "You're taking it at the wrong time of day." Accusatory hooks perform well because they promise the viewer is about to be corrected.
4. The contrast. Two things side by side, one obviously better. "$180 a month versus $12."
5. The overheard moment. Start mid-conversation with no context, so the viewer stays to work out what they have walked into. Street interviews are the pure form of this.
6. The refusal. "I'm not going to tell you this works for everyone." Hooks that withhold or qualify outperform hooks that oversell, particularly in skeptical categories.
Four variants each of angles one, two and three, and three each of four, five and six, gives twenty-one. Cut the weakest and you have your twenty.
Write them all before making any
Write the twenty as text, in one sitting, before rendering anything. Two things happen. You discover that four of them are the same hook in different words, and you discover which angles you find difficult — which is usually where the untested territory is.
Read them aloud. A hook that does not survive being said out loud will not survive being performed.
Production order
Do not render twenty. Render the hook beat for each — three to five seconds — and nothing else. Twenty hook beats is a fraction of the cost of twenty complete ads, and the hook is what you are testing.
Attach each hook beat to the same existing body, so you have twenty complete ads that differ only in the first five seconds. Then spend.
Structuring the test
Twenty variants in one ad set will not distribute spend evenly; the platform will pick two early and starve the rest, usually on the basis of a few hundred impressions. That is fine for outcomes and useless for learning.
Two workable approaches:
- Group testing. Four ad sets of five variants each, grouped by angle. Enough budget in each to give every variant a real sample. You learn which angle works, then test within the winner.
- Sequential rounds. Run ten, keep the top three, introduce seven new ones against them. Slower, but each round is cleaner and the control carries forward.
Either way, decide the budget per variant and the stopping rule before you launch. The most common failure in creative testing is calling a winner at 400 impressions because it looks good.
What to measure
For hook tests specifically, the early signal is 3-second view rate and hold to 50%. These separate faster than cost per purchase and they measure the thing the hook is responsible for.
But do not stop there. A hook can win attention and lose money — a shock opening that attracts the wrong audience will show a superb hold rate and a terrible cost per acquisition. Let the hold-rate leaders through to a spend test and judge the final decision on cost per result.
A practical read: hold rate tells you whether the hook works; cost per result tells you whether it works on buyers. You need both, in that order.
What you learn that outlasts the test
The point of testing twenty is not to find one winning video. Videos fatigue; the winner is dead in a month. The point is to learn which angle your audience responds to, because that survives.
If three of your four top performers are "the mistake" hooks, you have learned something about your market that will inform the next hundred ads. Write that down somewhere durable, because it is more valuable than the winning asset.
A realistic cadence
Twenty hooks against one product, tested properly, is a two-to-three week exercise including spend. A reasonable ongoing rhythm for an account spending meaningfully is one such round per product per quarter, with smaller five-variant refreshes in between to replace fatigued creative.
That cadence is achievable when a variation costs tens of dollars and minutes of work. It is not achievable when each one is a booking and a shoot, which is the actual reason most accounts test four.