Connect AutoAdy to your AI assistant, then run these prompts in order: extract the brand, generate the first batch, multiply the winner, then build the full test matrix.
The bottleneck in Meta ads keeps moving. It used to be targeting, and Meta's algorithm mostly handles that now. Then it was tracking, and the Conversions API mostly handles that. Today it's creative volume. You need enough distinct concepts in market for the algorithm to find the winners, and most accounts can't produce them fast enough. A buyer spending real money burns through dozens of fresh creatives a week, and a designer in a queue can't keep up.
This playbook clears that bottleneck. AutoAdy ships an MCP server, so once you connect it to Claude or ChatGPT, your assistant can read a URL, extract the brand's DNA, write the copy, and produce the creatives. One landing page becomes a batch of test-ready ads in a single conversation, and you expand the winners from there. Creative production stops being the thing you wait on.
Key takeaways
- Creative volume is the modern constraint, not targeting or tracking. Solve it and the algorithm does the rest.
url_to_adsturns one URL into brand DNA, framework-driven copy, and up to 50 creatives with test hypotheses attached.- Don't start from scratch when something already works.
multiply_winnerclones a proven ad's DNA and rotates one variable at a time.- Every generated creative lands paused. Nothing goes live until you say so.
Setup: connect AutoAdy to your assistant (2 minutes)
- In AutoAdy, open Settings → API Keys and create a personal key.
- In Claude, go to Settings → Connectors → Add custom connector and paste the AutoAdy MCP endpoint with your key. ChatGPT supports custom connectors the same way.
- Ask: "Which ad account am I connected to?" If it answers with your account name and currency, you're live.
Creative generation runs on Pro and above. Generated ads are always created paused, so the assistant produces and stages them, and launching is a separate, deliberate step.
Use case 1: From URL to brand DNA to the first batch
This is the most useful prompt in the stack. You give it one of your own URLs (a landing page or a product page) and it runs the full pipeline: it pulls your brand's voice and visual identity, writes copy in a proven direct-response framework across short, mid, and long lengths, and generates a batch of creatives, each tagged with the hypothesis it's testing.
The prompt:
Take this URL: [LANDING PAGE / PRODUCT URL].
Extract the brand DNA: voice, visual identity, target avatar, the core
mechanism that makes the offer work. Then generate 27 ad creatives across
these angles: pain, gain, social proof, authority, urgency, curiosity.
Write copy in short, mid, and long lengths. For each creative, state the
test hypothesis (what specifically it's checking). Aspect ratio 4:5 for feed.
This calls url_to_ads, and you can request up to 50 in one run. The output isn't a pile of random images. It's a structured test: every creative is tied to an angle and a hypothesis, so when results come in you learn why one won, not just that it did.
For a full launch, swap in the launch variant:
Run a product launch blitz from [URL]: 50 ads across 6 messaging angles,
maximum coverage so I can find the winning angle fast.
That calls product_launch_blitz, the same engine tuned for breadth on day one of a launch.
Use case 2: Multiply your proven winner
Once an ad is clearly working, don't start the next batch from a blank page. The winner already encodes what your audience responds to. Clone its DNA and rotate one variable at a time, so each new variation is a clean test rather than a fresh gamble.
The prompt:
Ad [AD ID] is my current winner. Generate 10 variations that preserve its
winning DNA (the core message and what makes it convert) but each rotates
exactly one variable: hook, visual style, or angle. Tell me which variable
each variation changes so I can read the results cleanly.
This calls multiply_winner. Because every variation isolates a single change, the next round of data tells you something specific, like "the question hook beats the statement hook for this audience," instead of leaving you to guess which of five differences mattered.
Use case 3: Build the full angle × hook × style matrix
When you want systematic coverage instead of iteration on one winner, build a matrix. You name the angles, hooks, and visual styles, and the assistant produces every combination as its own creative. That's a full factorial test in one prompt.
The prompt:
Build me a creative test matrix.
Angles: [e.g. founder story, problem-agitate, before/after]
Hooks: [e.g. bold claim, question, pattern interrupt]
Visual styles: [e.g. UGC, clean studio, text-on-screen]
Generate one creative per combination so I can see which angle × hook ×
style wins. Keep it to the strongest combinations if it exceeds 27.
This calls creative_matrix, capped at 27 combinations. It turns "let's try some stuff" into a designed experiment. When the results land, you don't just have a winner, you have a map of which dimensions drive performance for this offer.
Use case 4: Tailor to placement and funnel stage
A 1:1 square built for the feed looks broken in Reels, and a cold-audience ad shouldn't carry a buyer-stage CTA. Once you have winning concepts, adapt them to where they'll run and who'll actually see them.
The prompt:
Take my winning concepts and generate placement-native versions:
4:5 for feed, 9:16 for Reels and Stories, 1:1 for right column, each
designed for how people consume that placement, not just cropped.
Then build a retargeting ladder: cold, warm, hot, and buyer-stage
creatives, each matched to where that audience is in the funnel.
The first half calls placement_optimize. The second calls retargeting_ladder.
How the steps stack
| Stage | Tools | What it does |
|---|---|---|
| Extract & produce | url_to_ads, product_launch_blitz | URL to brand DNA to framework copy to up to 50 staged creatives |
| Multiply winners | multiply_winner | Clones proven DNA, rotates one variable per variation |
| Systematic testing | creative_matrix | Every angle × hook × style combination as a designed experiment |
| Adapt to delivery | placement_optimize, retargeting_ladder | Placement-native and funnel-stage versions |
The win here isn't that you can make 50 ads. It's that you can make 50 ads that are each a real test with a hypothesis, faster than a design queue makes five. Once creative volume stops being the constraint, the algorithm finally has enough to work with.
What this means
If creative production is the real bottleneck in modern Meta ads, the operators who clear it pull ahead. Spamming variations doesn't do it. What works is producing structured tests fast enough for the algorithm to find winners, then multiplying those winners cleanly instead of starting over every week.
Connect AutoAdy to your assistant, paste your best landing page into the Use case 1 prompt, and you'll have a staged, hypothesis-tagged batch in minutes. Review them, launch the ones you like, and let the next prompt multiply whatever wins.