Most marketers ask ChatGPT “build me the best angle for my product” and wonder why their Meta ads flop. Our AI ad creation system flips that script. It is a two-phase research and creation framework that feeds Claude and ChatGPT structured market data, competitor intelligence, and expert knowledge before a single ad concept is written. The result on one client account: $23K in spend at almost 5x ROAS (Return on Ad Spend), driven entirely by ads built with this system.
This guide breaks down the exact prompts, documents, and order of operations we use inside Skaleit to produce winning Facebook ads on demand in 2026. No fluff prompts, no “act as a marketer” nonsense. Just the workflow that actually moves ROAS.
TL;DR
– The AI ad creation system has two phases: research (market data, competitor research, expert knowledge, reviews) and analysis plus creation (brand memory, audience psychology extraction, ad briefs). – Build four foundational documents: social listening doc, winning ads doc, expert knowledge doc, and competitor reviews doc. – Use ChatGPT to compile a brand memory from your own winning assets, then upload everything to Claude for psychology analysis and ad brief generation. – Real result: $23K spend at nearly 5x ROAS on one account using ads produced through this system. – AI handles research and structure. Human creativity handles execution and visual delivery. Both are required.
Why Generic AI Prompts Fail at Meta Ads
Generic AI prompts produce generic ads. Asking ChatGPT for “angle ideas” without context gives you the same output every competitor is getting. The model has no understanding of your audience pain points, no exposure to your category’s winning ad patterns, and no memory of what has historically converted for your brand.
Winning Meta ads in 2026 require three layers of input the AI does not have by default:
- Real audience language scraped from where buyers actually talk
- Decoded competitor winners with transcripts, hooks, and angles extracted
- Expert frameworks specific to your niche and scaling stage
Without these, the AI is guessing. With them, the AI becomes a research analyst that can synthesize patterns no human has time to read through manually. According to Meta’s own performance guidance, creative quality is now the single largest variable in ad performance post-Andromeda, which is why front-loading research matters more than ever.
Phase 1: The Four Research Documents
Phase 1 builds the knowledge base the AI will pull from. You will create four documents, each from a different source. Skipping any of them weakens the entire system.
1. Social Listening Doc (Market Data)
Research your audience across Reddit, YouTube, Instagram, and Facebook using site-specific Google searches like `site:reddit.com probiotics` or `site:instagram.com [your niche]`.
For each platform, export: – Reddit threads in full – Facebook and Instagram comments under relevant posts – YouTube video transcripts AND the comment sections
Why multiple sources? Instagram and Facebook commenters are often not the same people who actually buy. Reddit and YouTube comments tend to surface higher-intent, higher-quality language. Pulling from all four gives you a broader, more accurate picture of how your buyer thinks.
Dump everything into one file: `social_listening.doc`.
2. Winning Ads Doc (Competitor Research)
Use Foreplay, Trio, or the Meta Ad Library directly. As of 2026, the Ad Library now auto-orders ads by winner, so manual research is faster than it used to be.
Filter by active or longest running ads. Export at least 20 to 30 transcripts of winning video ads in your category. Save them in `winning_ads.doc` with consistent naming so the AI can reference them later.
If you want to go deeper on what specifically makes static creative outperform video in many ecom categories, we cover that in our breakdown of the static ads strategy that hit 4x ROAS without a single video.
3. Expert Knowledge Doc
Find YouTube videos, LinkedIn posts, and case studies from operators who have scaled brands in your specific niche. Pull transcripts. Save case studies. Drop everything into `expert_knowledge.doc`.
The key here: be specific to your product, not broad. “Supplement brand scaling” beats “ecommerce marketing.” “Toy brand under $30K/month” beats “DTC growth.”
4. Competitor Reviews Doc
Go to Amazon and Trustpilot. Use a tool like Instant Data Scraper to export every review your top three to five competitors have. Reviews are the cleanest source of buyer language because they come from people who actually paid money. Save as `competitor_reviews.doc`.
Phase 2: Brand Memory and AI Synthesis
Phase 2 turns research into ads. This is where ChatGPT and Claude do the heavy lifting.
Step 1: Build the Brand Memory
Scrape your own website. Pull your past winning static ads, winning video transcripts, top-performing angles, and existing personas. Upload it all to ChatGPT and ask it to create a `brand_memory.doc`.
This document summarizes: – Angles that have worked – Personas that have converted – Hooks and openers that drove CTR – Script structures from your top performers
The brand memory becomes the persistent context layer for every future ad you brief.
Step 2: Switch to Claude for the Heavy Reasoning
Claude handles long-context reasoning and copywriting noticeably better than ChatGPT for this stage. Upload all five documents to Claude:
– `brand_memory.doc` – `social_listening.doc` – `competitor_reviews.doc` – `winning_ads.doc` – `expert_knowledge.doc`
Then run a synthesis prompt that extracts:
- Audience psychology (what do they actually believe?)
- Pain points (in their own words)
- Objections (what stops them from buying?)
- Language patterns (the exact phrases they use)
- Competitor perception (how do buyers see the alternatives?)
- Hooks and emotional triggers (what stops scroll?)
If you stop the system here, you are already 10 steps ahead of any competitor who is using base ChatGPT prompts. This output alone is enough to brief a copywriter or creative team.
Step 3: Generate Ad Briefs
With the synthesis complete, run two follow-up prompts in Claude:
– One prompt for static ad concepts (angles, headlines, visual direction, persona match) – One prompt for video ad briefs (hook, problem, solution, proof, CTA structure)
These briefs go directly to your creator or static designer. No more vague “make us a Meta ad” requests.
Where AI Stops and Human Creativity Starts
Execution is where AI breaks down. The system above gives you angles, personas, language, and structural briefs. It does not give you the creative leap that makes an ad stand out in feed.
Ads that win for our clients 99% of the time share one trait: they look and feel different from every other ad in the category. New concept, unexpected visual, message and headline locked tightly to the image.
This is why we avoid fully AI-generated static ads. The AI will give you a competent, on-brief static. It will rarely give you the weird, scroll-stopping visual that breaks pattern. That part is your job.
The formula: AI handles research and structure, you handle the creative execution. Match your headline, your message, and your visual into one tight unit. The smarter the integration, the higher the CTR and the higher the ROAS.
For brands operating in the post-Andromeda Meta landscape, this human creative layer is no longer optional. The algorithm rewards creative diversity at a scale that pure AI output cannot sustain on its own.
What Results Look Like When the System Is Run Properly
On the client account we referenced at the top: $23,000 in Meta ad spend, almost 5x ROAS, all driven by ads briefed through this exact two-phase system.
The pattern we see across accounts:
– Ads briefed with the full research stack convert at 2x to 3x the rate of ads briefed with surface-level prompts – Time from concept to live ad drops by roughly 50% because the creative team gets clearer briefs – Winning angle hit-rate goes up because the AI is pattern-matching against real buyer language, not generic marketing copy
This approach pairs well with the creative cadence we describe in our 7.02x ROAS creative strategy breakdown, where volume of well-briefed concepts is the lever, not random testing.
FAQ
FAQ
What is the AI ad creation system in one sentence?
It is a two-phase workflow where you build four research documents (social listening, winning ads, expert knowledge, competitor reviews), compile a brand memory from your own winners, then use Claude to synthesize audience psychology and generate static and video ad briefs.
Why use both ChatGPT and Claude instead of one?
ChatGPT is faster for compiling the brand memory from your own assets. Claude handles long-context reasoning, audience psychology extraction, and copywriting at a higher quality level when you upload multiple large documents at once.
Do I need paid tools like Foreplay or Trio?
No. The Meta Ad Library now auto-orders ads by winner, so you can do competitor research manually. Paid tools speed up transcript extraction and filtering, but they are not required to run the system.
How long does the full system take to set up the first time?
Expect 8 to 12 hours for the initial document build across the four research files plus the brand memory. After that, refreshing the system for new campaigns takes 1 to 2 hours.
Can AI replace my creative team entirely?
No. The system gives you angles, briefs, hooks, and structural copy. It cannot reliably produce the scroll-stopping visual execution that separates winners from losers. Human creativity sits on top of the AI output, not under it.
Does this work for low-AOV ecom or only premium brands?
It works across price points. The research depth matters more than the AOV. Lower-AOV brands often benefit more because the system surfaces emotional triggers that pure spec-driven copy misses.
About the Author
Antonio Ventre is the founder of Skaleit, a Meta ads agency for ecommerce brands. Skaleit has managed over $10M in ad spend and specializes in scaling DTC brands using research-driven creative systems and post-Andromeda account structures.
Want Skaleit to Build This System for Your Brand?
If you want our team to run the full AI ad creation system on your account, build your research docs, brief your creatives, and scale your Meta ads, book a call with Skaleit here.