← ClaudeAtlas

ad-creative-testinglisted

Use when the user wants to test ad creative - headlines, images, videos, hooks or offers - on Meta, LinkedIn, Google, TikTok or any paid channel, or wants to know whether a creative "winner" is real. Also use when the user mentions creative testing, ad testing, split test, A/B test an ad, which ad won, creative fatigue, hooks, or "should I scale this ad". Sizes the test before it runs so it can answer the question, then reads it out honestly - on the metric that pays, with uneven delivery, early stopping and multiple variants accounted for.
sarojkjha/aaj-marketing-skills · ★ 1 · Testing & QA · score 70
Install: claude install-skill sarojkjha/aaj-marketing-skills
# Ad Creative Testing Most creative "winners" are noise. Someone launches four ads in one ad set, looks after three days, sees one with a better click-through rate, and scales it. Three things went wrong at once: the ads were not shown to comparable people, the test stopped before it could tell a real difference from a lucky week, and the metric was clicks rather than the result the money is spent on. This skill fixes all three. It sizes the test before it runs, so you know whether your traffic can answer the question at all. Then it reads the result on the metric that pays and says "no winner" when that is the answer. ## When to use The user is about to test ads, is looking at a test and wants to know what won, or is about to scale an ad on the strength of a short run. Do not use it to rank ads that were never set up as a test. A report of five ads that ran at different times to different people is a performance history, not an experiment. Read it for ideas and then test the best idea properly. ## Before you start 1. **Pick the metric that pays.** Conversions per impression is the default, because it is closest to cost per result and it counts an ad that attracts the wrong clicks against itself. Use click-through rate only when conversions are too rare to test on, and confirm the winner on cost afterwards. 2. **Get the control's current rate** on that metric, from the last few weeks of the ad you would keep if nothing changed. 3. **Decide the smallest lift