Last updated:
Quick answer
Write 3-5 copy variants around one variable, run them inside one campaign with even spend for at least 7 days, then keep the winner and kill the rest. Meta's own creative testing flow caps the test at about 20 percent of the campaign budget so the test does not starve what already works.
At a glance
All outcome figures are estimates from cited sources, not independent verification by us.
- What it is allowed to do
- Drafts for you
Writes drafts (emails, posts, plans, replies). You review and send. Nothing goes out without you.
- Time saved (estimate)
- Roughly 1-2 hours per test round versus writing variants from scratch.
- Difficulty
- Intermediate
- Tools needed
- ChatGPT, Meta Ads
- Typical cost
- ChatGPT and Claude free tiers work for drafting. Ad spend is separate; local small businesses commonly start around $15-25 per day.
Reviewed Oct 19, 2026
Basis (estimate): Estimate only: drafting 4-5 disciplined variants takes over an hour by hand; AI produces the set in minutes and the owner spends the time on review.
What does this setup do?
Generate a small set of ad copy variants from one brief, run a disciplined 7-day test with capped spend, and pick a winner by cost per result, with you approving every variant.
Who is it for?
Small business owners running their own Meta ads who want to test copy without burning the budget.
How do you set it up?
Tools
- A Meta ad account with at least one running campaign
- A ChatGPT or Claude account (free tier works) for drafting variants
- One clear offer and landing page per test
- A decision rule written down before launch: metric, duration, and what happens to the winner
Steps
- Pick one variable to test: the hook, the offer angle, or the call to action. Never test all three at once.
- Have the AI write 3-5 variants changing only that variable, from your offer facts.
- Set up the test inside your existing campaign using Meta's creative test flow, capping test spend at about 20 percent of the budget.
- Run for at least 7 days without touching it; shorter tests produce inconclusive results.
- Judge by one metric decided in advance, usually cost per result.
- Keep the winner running, pause the losers, and write down what won before the next test.
Example output
Illustrative. Your results will differ.
- This month's spending sorted into 6 categories
- 3 subscriptions you forgot about, flagged for review
- You decide what to cut. Nothing moves on its own.
Where to paste this prompt
ChatGPT: ChatGPT → Project instructions, or ChatGPT Tasks for scheduled reminders (once a day on Free)
Copy-paste prompt
Adapt bracketed fields to your situation.
Write Meta ad copy variants for a creative test.
Offer facts:
- Product/service: [what it is]
- Offer: [price, deal, or lead magnet]
- Audience: [who]
- Landing page promise: [one line]
- Variable to test: [hook, offer angle, or call to action - pick one]
Write 4 variants changing ONLY that variable. Each variant needs: primary text (under 125 characters for the visible part), headline (under 40 characters), and call to action.
Rules: one idea per variant, no invented claims or statistics, no fake urgency. Mark anything uncertain with [VERIFY]. Show me all four first; I will pick which ones run.Newsletter
Get the free starter pack
Free starter pack with five beginner copy-paste prompts after you confirm, plus three sourced setups in your inbox each week. Unsubscribe anytime.
What usually goes wrong?
Where agents tend to stall on this job. More patterns in Where AI gets stuck.
- Changing the budget, audience, or creative mid-test restarts learning and voids the result.
- Testing six hooks on the same visual is not a real test; the algorithm reads them as one ad.
- Ending the test early because one variant "looks" better produces false winners; wait the full 7 days.
- AI can invent testimonials or statistics; every claim must be one you can prove.
Which chat tools work with this prompt?
Paste the prompt into any of these assistants (free or paid tiers).
- ChatGPT
- Claude
Common questions
- How much budget does a copy test need?
- Cap the test at about 20 percent of the campaign budget so winners keep most of the spend. Each variant needs enough results to judge, which is why tiny budgets should test on clicks first.
- How long should the test run?
- At least 7 days on Meta; up to 30. Shorter tests produce inconclusive results, and touching the campaign mid-test restarts learning.
- What should I change between variants?
- One variable only: the hook, the offer angle, or the call to action. Changing everything at once teaches you nothing about what worked.
What tips make it smoother?
- Test concepts before hooks: a new angle ("saves time" versus "builds confidence") moves results more than a reworded first line.
- Write the decision rule before launch: which metric wins, how long it runs, what happens to the winner.
- On tiny budgets, test on clicks first and offers second; you need enough results per variant to learn anything.
Where do the claims come from?
Summaries use public sources linked below, not independent verification. How setups are picked.
- Read Everix’s write-up (everix.ai)
Everix: How Meta's creative testing tool works: 2-5 test ads, even spend, about 20 percent budget cap, 7-day default, decision left to the advertiser.
- Read Supernodes’s write-up (www.supernodes.ai)
Supernodes: One variable per test, one platform, one decision rule written before launch; 10-20 percent of spend reserved for testing.
Related setups
- Plan a seasonal promotion without the scramble
Plan a seasonal promotion end to end: the offer, the math check, the emails, and the social posts, all drafted before the season starts.
- Score inbound leads so the best ones reach you first
Build a simple lead scoring rubric in a spreadsheet, then use AI to score each inbound lead and route the hot ones to immediate follow-up.
- Build a local prospect list with AI help
Build a researched prospect list of local businesses: who they are, what is missing from their online presence, and a personalized opener for each.