The Ad That Wrote Itself Took 7 Prompts

📖 6 min read

The One-Prompt Ad Is Why Your CTR Is 0.8%

You open ChatGPT. You type “Write me a Facebook ad for my skincare brand targeting women 25-40.” You get back something that sounds like every other ad in the feed. Generic hook. Generic benefit. Generic CTA.

That’s not the AI’s fault. That’s a process failure.

The marketers spending $50K+/month on AI-assisted ads don’t use one prompt. They run a 7-layer pipeline that turns a blank page into a tested, compliant, conversion-ready ad – with variants. Here’s exactly how they do it.

Why One Prompt Produces Mediocre Copy

When you ask for “an ad,” the AI has to guess:

• Who’s the audience (really)?
• What pain point matters most?
• What stage of awareness are they at?
• What’s the competitive angle?
• What hooks work in this niche?
• What platform conventions apply?
• What compliance rules exist?

It guesses wrong on at least 3 of these. Every time. The result is safe, bland, forgettable copy that technically “sounds like an ad” but moves nobody.

The 7-Layer Ad Pipeline

Layer 1: Audience Deep Dive

Don’t describe your audience. Make the AI become your audience.

The prompt:

I sell [product] at [price point] through [channel].

Become my ideal buyer. Tell me:
1. What's their day look like when they first think about this problem?
2. What have they already tried that failed?
3. What words do they use to describe their frustration? (Not marketing words - real words they'd text a friend)
4. What would make them stop scrolling?
5. What's their objection at checkout?
6. What would they need to see/hear to override that objection?

Be specific. Use first person. No marketing jargon.

Why this works: You get language you’d never write yourself. Real phrases. Real frustrations. These become your hook fuel for Layer 3.

Layer 2: Competitive Gap Analysis

The prompt:

Here's what my competitors say in their ads:
[paste 3-5 competitor ad texts or describe their angles]

My product differs because: [your actual differentiator]

Find the gap. What is NOBODY saying that my audience actually cares about? What angle is underserved? What truth is everyone avoiding?

Give me 5 unexploited angles ranked by likely scroll-stopping power.

Why this works: Most AI ads fail because they say what everyone else says. This layer forces differentiation before you write a single word of copy.

Layer 3: Hook Generation

Now you combine Layer 1 (audience language) with Layer 2 (unique angles) to generate hooks.

The prompt:

Using this audience insight:
[paste Layer 1 output]

And these unexploited angles:
[paste Layer 2 output]

Write 20 scroll-stopping hooks for [platform]. Rules:
- Under 10 words each
- Use the audience's own language, not marketing speak
- Each hook should create an information gap or challenge a belief
- Mix formats: questions, statements, provocations, numbers
- Nothing that sounds like it was written by a copywriter

Rate each hook 1-10 on scroll-stop power and explain why.

Why this works: 20 hooks gives you options. The self-rating forces the model to evaluate quality, which actually improves the outputs. You’ll pick 3-5 winners for the next layer.

Layer 4: Full Copy Variants

The prompt:

Using these winning hooks:
[paste top 3-5 hooks from Layer 3]

Write a full [platform] ad for each hook. Structure:
- Hook (first line - the one that stops the scroll)
- Agitation (2-3 lines - twist the knife on the problem)
- Solution bridge (1-2 lines - introduce the product naturally)
- Proof element (social proof, stat, or specific result)
- CTA (clear, low-friction, matches the hook's energy)

Platform specs: [character limits, format rules]
Tone: [your brand voice]

For each ad, write a PRIMARY version and a CONTRARIAN version that takes the opposite emotional approach.

Why this works: The contrarian version often outperforms. If your primary is fear-based, the contrarian is aspiration-based. You’re pre-building your A/B test.

Layer 5: The A/B Filter

Make the AI pick winners before you spend money testing.

The prompt:

You are a performance marketer who has spent $10M+ on [platform] ads. Review these ad variants:

[paste all variants from Layer 4]

For each one, predict:
1. Estimated CTR range (low/mid/high) and why
2. Likely audience fatigue rate (how fast will it burn out?)
3. Comment section sentiment (will it attract haters, fans, or both?)
4. Biggest weakness that could tank performance

Then rank all variants. Pick your top 3 and explain why they'll win.

Kill any ad that's "fine but forgettable."

Why this works: This isn’t perfect prediction – but it catches the obvious losers before you burn budget on them. It also identifies the “boring middle” ads that get clicks but no conversions.

Layer 6: Compliance Check

Skip this layer and enjoy your ad account ban.

The prompt:

Review these ads for [platform] compliance:
[paste top 3 variants]

Check against:
- [Platform] advertising policies (especially [relevant categories])
- No income claims without disclaimers
- No before/after implications in restricted categories
- No second-person health claims ("you will lose weight")
- No engagement bait language
- No misleading buttons or fake UI elements in creative descriptions

For each violation found: flag it, explain the risk (warning vs. account ban), and rewrite the offending line to be compliant while keeping the impact.

Do NOT water down the copy unnecessarily. Only change what actually violates policy.

Why this works: The instruction “don’t water down” prevents the model from making everything bland “just to be safe.” You get compliant copy that still sells.

Layer 7: Creative Brief

The final layer connects copy to visuals.

The prompt:

For each winning ad:
[paste final compliant versions]

Write a creative brief for the visual/video:
1. What should the viewer SEE in the first 0.5 seconds? (thumb-stop visual)
2. What emotion should the image/video evoke?
3. Text overlay - what 3-5 words go on the creative itself?
4. Color palette and style (match the energy of the copy)
5. What should NOT be in the image (common mistakes for this niche)

Format as a brief a designer could execute without further questions.

What This Looks Like in Practice

A skincare brand running this pipeline went from:

• One-prompt ads: 0.9% CTR, $14 CPA, 3-day creative fatigue
• Layer Method ads: 2.7% CTR, $6 CPA, 12-day creative life

Same product. Same budget. Same AI model. The difference was 7 prompts instead of 1, taking about 25 minutes instead of 2.

Speed Hack: Template It

You don’t re-run all 7 layers for every ad. Once you’ve done Layers 1-2 for a product, save those outputs. They’re reusable for months until your audience shifts. Your weekly workflow becomes:

• Monday: Run Layers 3-5 (hooks, copy, filter) – 15 minutes
• Tuesday: Layer 6-7 (compliance, brief) – 10 minutes
• Wednesday: Creatives produced, ads launched
• Friday: Feed performance data back into Layer 5 for next week

Copy This Workflow