Most ad creative testing is undisciplined — you launch a few variants, one performs better, you scale it, and three months later you have no idea why it worked or what to test next. This stack imposes structure. Define your hypothesis with Claude before you test, set pre-determined evaluation criteria so you know what counts as a win, run the test, and interpret the results through a structured output prompt. Every test gets logged in Notion. The log compounds into institutional knowledge — not a graveyard of inconclusive experiments.
The Stack
The Prompt
This stack is built around the A/B Test Hypothesis Generator Prompt. Here's the abbreviated version — the full prompt with all variables and usage notes is on its own page.
You are a B2B paid media specialist designing a structured ad creative test. Review the current campaign performance data and the test objective provided below. Generate a structured test hypothesis: what you're testing, what you predict, and why. Define the evaluation criteria: minimum sample size, test duration, and the metric that determines the winner. Identify 3 creative variables to test (headline angle, visual format, CTA language) with a hypothesis for each. Output the test plan in a format ready to log in a structured test tracking document.[ ... continued — see full prompt ]
The Workflow
-
Review current creative performance
Pull the past 30 days of creative performance from Google Ads and Meta Ads — CTR, CPL, and conversion rate by ad creative. Identify the control (current best performer) for each test.
-
Generate the test hypothesis
Paste your campaign objective and current performance data into the A/B Test Hypothesis Generator prompt. Claude outputs a structured test plan with hypothesis, evaluation criteria, and 3 creative variables to test.
-
Set up the test
Launch the test variants in your ad platform with the same targeting, budget, and settings as the control. Set the scheduled review date based on the minimum sample size Claude recommended.
-
Interpret results at the review date
Pull results and run them through a test result interpretation prompt. Claude evaluates statistical significance and flags whether the declared winner is real or a noise artifact from insufficient sample size.
-
Log and build the knowledge base
Record the full test — hypothesis, result, confidence level, and transferable insight — in your Notion test log. Post the winner and key insight to Slack. This log becomes your creative strategy reference.
What This Replaces
- Creative tests launched without a hypothesis that can't be learned from even when they show a winner
- Tests called as winners at sample sizes too small to trust the result
- Creative decisions made from memory and gut feel because there's no structured test record
Related Stacks
New stacks drop weekly.
Each one includes the tools, the Claude prompt, and the workflow logic. Free — built for in-house B2B demand gen managers.