09 // CREATIVE TESTING FRAMEWORKS
Generating one script doesn't matter. Generating 50 variants does.
The affiliates winning right now aren't the ones with the single best creative. They're the ones testing the most variations systematically. AI makes volume free. But volume without structure is just noise.
Here's how to test systematically, not randomly.
Generating Variants at Scale
Most affiliates generate one script, test it, generate another, test it. That's painfully slow and you're basically guessing.
The better approach: generate variants in structured batches where you control one variable at a time. Then you know exactly what's working and why.
The Variable Isolation Method
Every piece of creative has testable components. For a video script:
- Hook (the first 5 seconds)
- Problem framing (how you describe the struggle)
- Mechanism (how you explain why the solution works)
- Proof type (personal result, doctor authority, statistic)
- CTA style (soft, medium, direct)
- Tone (confessional, educational, frustrated, hopeful)
Don't change everything at once. Change one variable per batch.
Example: Hook Testing
You have a 30-second script that's performing okay. The body is solid but you think the hook could be stronger. Don't rewrite the whole script. Generate 10 hook variations and keep everything else identical.
I have this 30-second video script that's currently running:
[paste your current script]
The body, solution, proof, and CTA sections are performing well.
I want to test new hooks only.
Write 10 alternative hooks (5-7 seconds each) for this script.
Each hook should:
- Use a different approach (controversial, confessional, pattern-
interrupt, skeptic-conversion, statistic-lead, insider-knowledge,
fear-of-missing, social-proof, counter-intuitive, direct-challenge)
- Fit seamlessly with the existing body copy
- Create a curiosity gap that the problem section resolves
Keep the same tone and compliance level as the original.
Now you have 10 hooks to test against your control. The body copy is the same across all of them. When one wins, you know it was the hook that made the difference.
The Angle Expansion Method
Once you have a winning angle, expand it across formats.
This angle is working well for 30-second video scripts:
[paste winning angle/theme]
Adapt this same core angle into:
1. Three 15-second hook-only versions (for hook testing)
2. One 60-second extended testimonial version
3. Three Facebook primary text variations (for image ads)
4. One advertorial headline + opening paragraph
5. Three landing page headline variations
Same core message. Same compliance boundaries. Different formats
and lengths. Match the tone and style of each format.
One winning angle becomes 11 pieces of creative across 4 formats. That's how you scale without starting from scratch every time.
Systematic Iteration: The Feedback Loop
Random testing tells you what won. Systematic iteration tells you why it won. That's the difference between finding one winner and building a machine that produces winners.
The 3-Round Iteration Process
Round 1: Broad test. Generate 5 scripts with different angles (biology reframe, doctor authority, personal story, statistic lead, curiosity gap). Test them. Find the winning angle.
Round 2: Hook test. Take the winning angle. Generate 10 hook variations. Test them. Find the winning hook.
Round 3: Body optimization. Take the winning angle + winning hook. Generate 5 variations of the body copy (different problem framing, different proof type, different mechanism explanation). Test them.
After 3 rounds, you have a creative that's been optimized at every level. And you know exactly which elements are driving performance because you only changed one thing per round.
The Iteration Prompt
Use this between rounds to get Claude to analyze what's working and suggest next tests.
Here are my test results from Round [X]:
Winner: [paste winning creative]
Loser 1: [paste] - CTR: [X]%, CPA: $[X]
Loser 2: [paste] - CTR: [X]%, CPA: $[X]
Loser 3: [paste] - CTR: [X]%, CPA: $[X]
Analyze:
1. What specifically makes the winner different from the losers?
2. Which elements of the winner should I preserve in the next round?
3. What variable should I test next to improve it further?
4. Write 5 variations for the next test round, changing only
[the variable you identified in #3].
Claude becomes your creative strategist, not just your copywriter. It analyzes patterns across your test results and suggests what to test next. That's way more valuable than just generating random scripts.
Tracking What Works
You need a system for cataloging winning elements. Otherwise you're rediscovering the same insights every week.
The Winners Library
Every time you find a winning creative element, log it:
WINNING ELEMENT: [Hesitation hook]
EXAMPLE: "Okay so... I need to tell you something about GLP-1"
FORMAT: Video hook (5s)
VERTICAL: GLP-1 / weight loss
METRIC: 2.1% CTR (vs 0.8% average)
WHY IT WORKS: Creates conversational tone, doesn't feel like ad
DATE FOUND: [date]
STILL WORKING: Yes/No
After a few weeks, you'll have a library of proven elements. When you prompt Claude for new creative, you can include these as examples:
Here are hooks that have performed well in my account:
1. "Okay so... I need to tell you something about GLP-1" (2.1% CTR)
2. "My doctor told me something about weight loss that made me feel
like an idiot" (1.8% CTR)
3. "95% of diets fail within 5 years. Here's the biological reason
nobody talks about." (1.6% CTR)
Generate 10 new hooks that share the qualities of these winners
(conversational tone, specific details, curiosity gap) but use
different specific content. Do not repeat the same structures.
Now Claude is learning from your actual data, not just general copywriting principles. Your prompts get better as your testing data grows.
GLP-1 Example: 10 Headline Variants from One Prompt
Here's the prompt and the output. This is how you generate a full test batch in 60 seconds.
The Prompt
You are a performance marketer generating A/B test variants for
Facebook ads promoting a GLP-1 telehealth offer.
Generate 10 headline variations for:
- Offer: SlimRx telehealth GLP-1 (semaglutide)
- Audience: 35-55, chronic dieters, multiple failed attempts
- Compliance: No weight guarantees, no medical claims
Each headline should:
- Be under 10 words
- Use a different angle or emotional lever
- Work paired with a biology-reframe primary text
- Be specific enough to stand out (no generic "weight loss" headlines)
Mix of approaches:
- 3 curiosity-driven
- 3 authority-driven
- 2 personal-story-driven
- 2 reframe-driven
Avoid: "Transform", "game-changing", "real results", exclamation
points, questions, superlatives.
The Output
Curiosity-driven:
1. "What Your Hunger Hormones Have Been Hiding"
2. "The Reason Every Diet Eventually Stops Working"
3. "Why Your Body Reverses Weight Loss (and How to Stop It)"
Authority-driven:
4. "Why Doctors Are Prescribing This Before Another Diet"
5. "What Endocrinologists Wish Dieters Understood"
6. "The Medical Reason Behind Your Failed Diets"
Personal-story-driven:
7. "After 6 Years of Yo-Yo Dieting, My Doctor Explained Why"
8. "I Lost the Same 20 Pounds Four Times. Then This."
Reframe-driven:
9. "It's Not Willpower. It's Hormones."
10. "Stop Blaming Yourself. Blame Your Biology."
How to Test These
Don't test all 10 at once. You'll burn budget and learn nothing.
Week 1: Test headlines 1, 4, 7, 9 (one from each category) against each other. Same primary text, same image, same audience.
Week 2: Take the winning category. Test all 3 headlines from that category against each other.
Week 3: Take the winning headline. Now start testing primary text variations paired with it.
Three weeks. Structured progression. You end up with a validated headline backed by data, not a guess.
Volume plus structure equals winners. Generate in batches. Isolate variables. Test systematically. Log what works. Feed winners back into your prompts.
That's how you turn AI from a content generator into a creative testing machine.