From day one — whether you’ve cracked a brand or you’re just starting — set up a claude project and make it your single source of truth.
The Claude Project (one brain for the brand)
Inside that project, keep everything:
- Market research
- Post-purchase surveys + analysis
- Creative analyses (winners/losers)
- Personas/avatars, angles, objections
- Prompts you and the team actually use
The rule: everything you do comes from that claude. It should be so dialed you can ask it questions and get new ideas only from your brand’s knowledge — not generic internet bullshit.
What it generates for us (pulled from the project only):
- Primary text
- Headlines
- New ad concepts
- Iterations on past winners
- Post-mortems on losers (why they failed)
- Checklists and briefs for editors/strategists
If the claude isn’t feeding your work, you’re winging it. One project. One brain. Everyone uses it.
Prompts live here (and you re-use them)
Below this section, I’ll drop the exact prompts we used — for iterating ads, making image ads, video briefs, and more. We used these religiously and made sure everyone on the team worked out of the same claude project so outputs stay consistent.
Analyzing video ads (use Google Studio)
For ad analysis, use Google Studio — that AI can actually watch the videos.
- Use the analysis prompt (I’ll paste it right under this) to break down every video.
- From those breakdowns, build two checklists:
- Loser checklist → what our ads must never do again.
- Winner checklist → patterns to repeat (hooks, angles, pacing, CTAs, visual beats).
Then store both checklists back in the claude so every new brief/iteration is measured against reality, not vibes.
Your losers teach guardrails. Your winners teach patterns. Write both down and enforce them.
How this actually produces winners
- Read your market research + surveys to educate yourself & discern winners.
- Ask the claude for angle → script → headline → primary text based on that exact persona.
- Generate image ad concepts fast (image is researcher-led; easy to do volume).
- Video is harder, but manageable if:
- Prompts are tight
- Briefs are specific
- Editors/strategists follow the same checklists
- Every new output is compared against what worked/what failed
This is how you dial in AI so it speeds you up at scale.
Guardrails (don’t let AI torpedo your brand)
- Double-check everything AI spits out. We used it for volume and direction, not blind publishing.
- Keep it brand-informed (claude-only context). If it starts sounding generic, you fed it trash.
- Image ads = great for AI-accelerated volume.
- Video ads = need stricter briefs + checklist enforcement.
- All outputs (texts, headlines, scripts) get saved back into the claude, so the next round is iteration, not from scratch.
AI is a force multiplier. If your inputs are dialed, you scale. If they’re generic, you burn time and money
⚡ Shoutout: Ecom Talent by Karlo
I have to give credit where it’s due. Ecom Talent by Karlo is where I learned a ton about creative strategy, prompting AI, and actually building systems around this.
- Many of the prompts in here came directly from Ecom Talent.
- If you want the full-length, 100/100 skill-level prompts, strategies, and systems — you need to go there.
- The diluted versions are what I’ve included here, plus my own originals. But the real sauce is in Karlo’s ecosystem.
👉 If you’re serious about mastering creative strategy and AI in ecom, go through the Ecom Talent courses, strategies, talk to Karlo, and use the prompts inside.
ecomtalent
ecomtalent isn't a course. It's a complete skill-to-income system that does 2 things: - Transforms you into a high-paid marketing specialist who creates ads that make people buy products online like crazy. - Connects you directly with ecommerce brands specifically looking for people with your skillset to make ads for their brand. Master the skill, then connect with ecommerce brand owners looking to hire - all within the ecomtalent.
whop.com