Write Ad Variants That Test One Variable
Added 2026-11-01
Change the headline or the image, never both, or the data tells you nothing.
I am running a set of ads and want to know which works better. Do: 1) ask what hypothesis I am testing; 2) give three variant sets, each changing exactly one variable — headline, hero image or first line — with everything else held constant; 3) for each set, state what a win for A versus B would mean; 4) suggest a minimum sample and observation window; 5) warn me about the most likely misreading, such as high clicks and poor conversion.
Site original
Chinese original
How to use
广告测试一次只改一个变量，否则赢了也不知道为什么。
Try tweaking it
- Write one selling point for three readers: beginner, expert, decision maker.
- Replace "compelling" with a structure: pain, proof, action.
What you should get
- Several versions by channel or reader
- The claim and proof behind each version
Common mistakes
- Asking for "more compelling copy" and getting adjectives instead of facts.
- Letting it invent user counts or reviews — a risk the moment you publish.
How to check the output
- Claims are specific and checkable, without empty adjectives.
- No invented sales, user counts or reviews.
Where it works
- Good for: writing one selling point in several channel voices, then comparing which is most specific.
- Watch out: never let it fabricate sales, user counts or reviews — leave unknown numbers blank.