What's up, Marketers! This is Aazar.
This newsletter is all about giving you ad ideas to find winning ad ideas and help you scale.
This newsletter includes: Analysis of top internet ads, insights from other performance marketers, and my learnings (this issue).
You can make 100 AI static ads within an hour with Claude chat.
I’ve put together 10 agents (that work like a creative team) and trained them with 17 skills.
Making the images is the easy part now. What decides if those ads make money happens before a single image exists. It's one step almost everyone skips. You'll find it at Step 7.
Here's the whole thing, step by step.
Note: Claude code is powerful, but I intentionally wanted to make this method using Claude chat.
Caveat: While Claude provides a good number of static ads, I still think copywriting as a skill is not replaceable by AI. Copywriting, copythinking, and creative strategy are still the most valuable skills. For example, you might end up wasting 95% of your budget on the wrong ads. So judgement and taste are crucial.
Before you start
Claude paid plan (and desktop app)
Paid Higgsfield account, connected in Claude under Settings, then Connectors. The free tier won't carry 100 images
The agents and skills, free to download at the end
Which model to use:
Plenty of credits will be used: Fable 5.1. Sharpest angles and audits
Fewer credits: Opus 5.5. Opus 5 works too
Either way, be ready to burn a lot of credits and tokens. Research, 100 rows of copy and rework loops add up fast.
Step 1: Set up your workspace (30 min, once)
Create a project or a folder in Cloud Desktop for the working project.
You can run this two ways. Pick one:
Claude Project: everything lives inside Claude. Works on web, desktop and phone
Local folder: in the Claude desktop app, connect a folder on your computer. Claude reads your files from it and saves every output straight back into it
Using a Project:
Create a Claude Project called [Client] Statics
Paste the Project Instructions file into the instructions box
Upload the 10 agent files, the Playbook, Feedback Log and Headline Formula Library
Upload 3 to 5 clean product photos, the logo and 2 or 3 lifestyle shots
Install the skills
Using a local folder: put the same files in one folder per client. Start every chat with "Read 00 Project Instructions first."
Prompt:
Read 00 Project Instructions. Then list every file you can see in this Project or folder, and tell me what's missing before we start. Don't start any work yet.
Note:
Two ways to run this
Method 1: Step by step (best quality)
A new chat and a new prompt for each step. It takes longer, but every agent starts fresh and does one job well. If you want the best possible ads, use this.
Method 2: One big prompt (fastest)
Paste one prompt and Claude runs all the agents for you. It stops whenever it needs your input. Good when you’re short on time. Quality drops a little, because one long chat ends up grading its own work.
Whichever you pick, you still audit the headlines yourself (Step 7).
The Method 2 prompt
Run the full static ad system for [CLIENT], start to finish, in this chat.
Client: [CLIENT]
Website: [URL]
Product + price: [what it is + price]
Audience: [who buys]
Competitors: [3 to 5 names or URLs]
Round: [Give this round of statics a name]
Goal: 100 ads (5 personas x 5 angles x 4 templates)
First read 00 Project Instructions. Then work through the agent files, 01 to 10. Follow each file's prompt exactly. Before each one, write "Now acting as [agent name]".
Agents:
- The order 01 to 10 is the default, not a fixed rule
- You can switch to any agent whenever the work needs it. For example, go back to the Customer Researcher to find proof for a claim, or to the Template Curator to swap a layout that doesn't fit
- Tell me in one line which agent you're switching to and why, then carry on
Stop and wait for me at every one of these points:
1. During the Brand Brain interview: ask me one question at a time
2. After the Angle Map: I approve, cut or edit the angles
3. After the Creative Gatekeeper: show me only the headlines, one per line with the Ad ID, no scores or sources. Do not make any image until I type APPROVED FOR DESIGN
4. After the first 3 images: wait for my go
5. After QA: I pick the keepers on the review board
Rules:
- Work in batches of 20 ads, one persona at a time
- When you act as the Creative Gatekeeper or the QA Reviewer, judge the work as if someone else made it. Be strict. Never lower the bar
- Switching agents never skips a stop. Never skip a stop, even if I sound in a hurry
- If this chat gets long, write a hand-off note so I can carry on in a fresh chat
Start with the Brand Brain Builder.
Step 2: Build the brand brain (40 min, once)
Agent 1 interviews you about the brand. Answer by voice and ramble. More context, better ads
Agent 2 pulls real customer words from reviews, comments and Reddit
Read both outputs and fix mistakes now. Every later agent trusts them
Please start and continue the chats in the specific project or folder.
Prompt:
Agent: Brand Brain Builder. Follow the Agent 01 file.
Client: [CLIENT]. Website: [URL].
Read the site and my attached files first, then interview me one question at a time.
Prompt:
Agent: Customer Researcher. Follow the Agent 02 file.
Product: [what it is + price]
Audience: [who buys]
Competitors: [3 to 5 names or URLs]
Use Markdown as the output destination. Save the result as "03 Research Bank".
Step 3: Build your template library and add conext about the project (1 hour, once)
Collect 20 to 30 static ad screenshots: long-running ads from Meta Ad Library, the client's winners, native formats like notes app, iMessage and tweets
Run Agent 4. It turns each one into a reusable layout
Keep at least 15 strong ones. Weak templates mean weak ads
Here I used Atria’s AI to give me a context document that would give all the important learning about the client, things like:
What angles are working
Desires
Pain Points
Persona
Offers
Winning words and headlines
When you run this simple prompt with Atria (the Ask Raya), it goes and scrapes the actual Meta ad account. It then gives you a very organized context document that you can use for creating static ads.
I experimented with two versions:
1. One where I did not provide this context document.
2. Another where I did provide the context document.
The experiment with the context document won by a lot. It produced much better creatives compared to the one where I did not provide the document.
Prompt:
Agent: Template Curator. Follow the Agent 04 file.
I've attached [N] static ad screenshots. Score each one, drop anything under 3, and save the rest as "05 Template Library". I need at least 15 strong templates.
Thanks to our partners who support this newsletter.
I know you must have seen Atria mentioned several times in this newsletter, but they keep introducing features I can't help but love. Here are three hidden features that Atria launched, which make me look like a genius in front of my clients.
Rising ads tell me what is working right NOW, not 30 days ago.
I am on auto-alert if a format or video concept that is going viral for the brands I follow to make better ads.
Basically, I never run out of ideas because I can see who is pushing more volume and what new ideas they are trying.
If you’re an agency owner or a brand, you know how important it is to stay ahead. Meta is all about what is evergreen and what is working right now. So, steal these ideas with Atria AI.
Step 4: Pick your angles (Checkpoint 1)
Run Agent 3. For 100 ads, ask for 5 personas × 5 angles
Cut the angles every competitor already runs
Add the ones only you would think of
Prompt:
Agent: Angle Strategist. Follow the Agent 03 file.
This round needs 100 ads, so recommend 5 personas x 5 angles = 25 angles.
Show me the table and stop. I'll approve, cut or edit before you save "04 Angle Map".
Step 5: Write the copy in batches of 20 (Recommended)
Run Agent 5 one persona at a time. Ask for 4 templates per angle
5 angles × 4 templates = 20 ads per batch. Five batches = 100
Don't review it yet. Copy the table into a fresh chat
Prompt:
Agent: Static Copywriter. Follow the Agent 05 file.
Round: [R1OCT]. This batch is Persona [P1] only.
Use 4 templates per angle, so 5 angles x 4 templates = 20 ads.
Don't generate images. End with the hand-off note for the Gatekeeper.
Step 6: Let the Gatekeeper audit it
Run Agent 6 in a new chat, so it judges work it didn't write
It scores every ad, rewrites the weak ones and kills bad ideas
You get an "Approved for design" list
Prompt:
Agent: Creative Gatekeeper. Follow the Agent 06 file.
You did not write any of this. Here is the Copywriter's table and hand-off note for round [Round Name], Persona [P1]: [paste]
Audit, rework and give me the "Approved for design" list. Then stop and wait for my release line.
Step 7: Audit every headline yourself (Checkpoint 2)
This is the step that saves your budget. Don't skip it.
The Gatekeeper is good. But it's still AI grading AI.
AI headlines can be sharp. They can also be bland, or flat-out bad. Skip this and you could send a batch of really bad headlines straight to design.
One bad headline gets copied onto 4 images. You pay for all 4, review all 4, and maybe pay Meta to learn it was bad.
Check each headline (about 10 minutes per batch):
Can you get it in 2 seconds on a phone?
Does it sound like a customer, not a brand?
Is every claim true?
Would you stop scrolling for it?
Fix what's close. Cut the rest. Then type:
APPROVED FOR DESIGN
Prompt:
Show me only the headlines from the "Approved for design" list. One per line, with the Ad ID. No scores, no sources, no comments.
Step 8: Make the images
Make sure you have your Higgsfield account connected.
Make sure your cloud is connected with Higgsfield
Run Agent 7 with the approved list
It makes 3 first. Check spelling, the product and any leftover logos
Say go. It makes the rest of the batch in Higgsfield
Prompt:
Agent: Image Producer. Follow the Agent 07 file.
APPROVED FOR DESIGN
Here's the approved list: [paste]
Make the first 5 only and show me. Wait for my go before the rest.
Step 9: Pick your keepers (Checkpoint 3)
Run Agent 8 in a fresh chat. It grades every image and builds a review board
Star your keepers. Keeping a third to a half is a win
Prompt:
Agent: QA Reviewer. Follow the Agent 08 file.
You did not make these ads. Be strict.
Attached: the finished images and the "Approved for design" list for [R1OCT].
Score each one Pass, Fix or Cut, then build my review board.
Step 10: Launch and read results
Agent 9 builds your launch sheet: one ad set per persona, 5 to 10 ads each, everything paused
Turn the ads on yourself
Once spend hits 3× your target CPA, run Agent 10. It shows which angles and personas won
Write every lesson in the Feedback Log. Your next round starts at Step 4
Prompt 1:
Agent: Launch Packager. Follow the Agent 09 file.
Kept ads: [paste picks]
Landing page: [URL]. Daily budget: [].TargetCPA:[].
Build the launch sheet as a CSV. Everything paused.
Prompt 2:
Agent: Results Reader. Follow the Agent 10 file.
Attached: the Ads Manager export for [R1OCT], [date range], and the launch sheet. Target CPA: [$].
Roll up results by angle, persona, template and headline source. Draft updates for 04 Angle Map and 07 Feedback Log.
Get the agents and skills
Everything's in one folder: The Performers Agents and Skills
10 agent files with copy-ready prompts
Project Instructions, Playbook, Feedback Log and Headline Formula Library
16 skills, plus a README showing which agent uses which
To install a skill, go to Settings in Claude and upload the .skill file.
Happy Growing with Paid Social,
Aazar Shad
Since this newsletter is free, I do it to follow my curiosity. But I’d love it if you could leave some feedback so I know if I am helping you or not.
One way I can help you, whenever you are ready:
Want my creative strategy services for your Q4? Last time we crushed it for a brand and our ad spent more than $ 10 M. Book a call here. I’m open to more clients now in Q4.
