Most ad teams start with ideas.
A better system starts with evidence from the market, then turns that evidence into tests.
I call it the “Ad Signal Loop”:
1. Build the ad library
Tool: Facebook Ad Library
Collect active ads from the companies selling into the same category. Save the image, video, headline, offer, and landing page.
2. Turn creatives into notes
Tool: Apify + Gemini
Use Apify to gather the creative files. Have Gemini describe each one by its hook, promise, proof, objection, and expected outcome.
You’re looking for repeated patterns, not one clever ad.
3. Find the empty spaces
Tool: Claude
Give Claude the creative notes and ask it to group the market’s common messages.
Then ask it to identify claims, audiences, and problems that are being ignored.
4. Check whether the gap exists
Tool: Exa
Search discussions, reviews, and questions around those gaps.
Keep the ideas that show up in real customer language. Drop the ones that only sound interesting in a strategy document.
5. Make a small batch
Tool: Nano Banana + Seed Dance
Turn the strongest angles into several image and video variations.
Change one meaningful variable at a time. Keep the offer and audience clear enough to compare.
6. Put the tests in market
Tool: Meta Ads API
Launch the variations with a defined budget and a clear conversion event.
Don’t decide what worked from the ad that looks best in a folder.
7. Read the signal
Tool: Graphed MCP + Claude
Connect the performance data to Claude.
Have it compare spend, clicks, conversions, creative angle, and audience. Separate weak execution from a weak promise.
8. Reallocate with rules
Tool: Meta Ads API
Pause the ads that fail the agreed threshold.
Move the stronger ones into their own groups so new tests don’t hide their performance.
9. Keep the loop running
Tool: a server
Run the collection, analysis, launch, and review process on a schedule.
The system gets useful when every round adds to the next round’s market map.
The important shift is simple:
Research creates the hypotheses.
Ads create the evidence.
Performance decides what deserves more attention.