Creative testing fails when it is random. The 5x3 framework gives you a one-week sprint that isolates variables and produces a decision.
The setup
- 5 hooks — different psychological families (interrupt, claim, callout, demo, confession)
- 3 bodies — the same core message delivered through different frameworks (PAS, AIDA, before-after-bridge)
- 1 CTA — held constant so it never confounds results
The sprint
Day 1-2: Ship all 15 combinations at equal budget. Kill nothing yet — you are buying data, not conversions.
Day 3-4: Read hook rate (3-second views / impressions). Kill the bottom two hooks. Reallocate budget to the survivors.
Day 5-7: Read hold rate and CPA on the surviving six. Your winner is the lowest CPA above minimum viable spend.
Then iterate
The winning combination becomes your control. Next sprint, keep the winning hook and test three new bodies, or keep the body and test five new hooks. Never change both.
With AI generation, producing 15 variants is an afternoon, not a quarter. That is the real unlock — testing velocity, not just production cost.