kochellenk-afk Meta Creative Engine
On Meta, creative is the primary performance lever, the way keywords and Quality Score are the lever on Google. Audiences are largely automated now (broad plus Advantage+), so what you make and how you test it decides the account. This skill plans creative, diagnoses why creative is failing, and iterates winners. It does not just write ads.
Step 1: Set the client type
Every downstream decision reads against the client's scoreboard. Confirm which one before anything else.
Ecommerce preset. Scoreboard: purchases, ROAS, cost per purchase, AOV. Assume catalog and Advantage+ Shopping are available. Destination is a product or landing page.
Content or creator preset. Scoreboard: cost per subscriber, cost per lead, or cheap qualified engaged traffic, plus watch-through. Assume creator-native Reels and lighter funnels. Destination is often a profile, a subscribe page, or a lead form.
If it is unclear which applies, ask once, then proceed.
Step 2: Pick the mode
State which mode you are in at the top of your response.
- Test Plan. They have a product or offer and no data yet. They need what to make and how to test it.
- Diagnostic. They have performance data and something is wrong, or they want a health read. This is the core mode.
- Iteration. They have a winner (or a fatiguing winner) and need the next batch.
Mode 1: Test Plan
Gather: product or offer, price, target buyer, primary objective, testing budget, and any existing angles or creative.
Build a testing matrix of angles by hooks by formats.
Angles are the distinct reasons to buy, the emotional or rational entry points. Aim for 3 to 5. Common angles: problem and agitation, social proof, founder story, before and after, price and value, speed and convenience, identity and belonging.
Hooks are the first three seconds. Write 2 to 3 hook variants per angle. The hook carries most of the performance, so spend the most thought here.
Formats map to production capacity: UGC video, founder talking head, static, carousel, testimonial. Match formats to what can actually be produced (statics in Canva, UGC video in Pletor).
Read-window rule. Define spend-per-ad and a minimum number of results before any verdict, so ads are not killed inside the learning phase. A directional starting rule: let each ad reach roughly a few times the target cost per result in spend, or a handful of results, before judging, and give the campaign 3 to 7 days to stabilize. Adjust up for higher-priced offers and down for cheap high-volume ones.
Output template:
## Meta Creative Test Plan: [Client / Offer]
Objective: [purchases / subscribers / leads] | Scoreboard: [ROAS / CPA / CPL]
Testing budget: [amount] | Read window: [spend or results before verdict]
### Testing Matrix
| Angle | Hook (first 3s) | Format | Production tool |
|-------|-----------------|--------|-----------------|
| ... | ... | ... | Canva / Pletor |
### Shot list / concepts
[Per row: the opening frame, the beat-by-beat or layout, the CTA]
Mode 2: Diagnostic (the core mode)
Do not guess "make better ads." Locate the single broken stage with a fault tree, then prescribe the specific fix. The point of this mode is to tell the client which lever to pull.
The metrics and what each one isolates:
- Hook rate (3-second video views / impressions): is the first frame stopping the scroll?
- Hold rate (thruplays or 15-second views / 3-second views): does it hold attention past the hook?
- Outbound CTR (outbound clicks / impressions): does it drive the click?
- Landing conversion rate (results / landing page views): is the click converting?
- CPM: auction and audience cost signal.
- Frequency: fatigue signal.
- Cost per result vs target: the bottom line.
The fault tree. Walk it in order and stop at the first break.
- Cost per result at or under target? The account is healthy. Move to scaling (Mode 3 or the Campaign Architect skill). Do not touch working creative.
- Hook rate low? The break is the opening frame. Fix the first three seconds: new visual hook, pattern interrupt, on-screen text. Everything downstream is starved of traffic until this is fixed.
- Hook fine but hold low? The break is the body. The hook earned attention the middle did not keep. Fix pacing and payoff, tighten the middle, deliver on the hook's promise faster.
- Hold fine but outbound CTR low? The break is the CTA or offer relevance. The viewer watched but did not act. Fix CTA clarity, offer framing, or the match between promise and click.
- CTR fine but landing conversion low? It is not the ad. It is the landing page. Hand off to landing-page-match-scorer. Do not iterate creative for a landing page problem.
- Everything upstream fine but cost per result still high? It is CPM or audience cost, or the offer economics do not support the target. Hand off to the Meta unit economics skill.
- Frequency rising and metrics decaying across the board? Creative fatigue. Go to Mode 3 and refresh, do not just raise budget.
Benchmarks are directional, not gospel. Ranges vary widely by vertical, placement, geo, and objective. Prefer the account's own trailing baseline whenever it exists (this ad set last month versus now). As loose directional anchors on cold prospecting: hook rate is often healthy around 30 percent and weak under roughly 20 percent; outbound CTR is often healthy around 1 percent and weak under roughly 0.5 percent; frequency above roughly 2 to 3 on a short cold window is worth watching. Treat these as starting points and defer to the account's baseline.
Output template:
## Meta Diagnostic: [Client / Ad or Account]
Verdict: [the ONE broken stage, in a sentence]
Evidence: [the metric(s) that locate it, vs target or baseline]
Fix: [the specific change, or the hand-off if not a creative problem]
Watch next: [what to re-check after the fix]
Mode 3: Iteration
Given a winner, iterate, do not restart. Ranked by leverage:
- Same body, new hooks. Cheapest, highest-leverage move. A working ad with three new openers is three new tests at low cost.
- Same angle, new format. Port the winning message to a new format (static winner to video, or the reverse).
- Winner's angle to adjacent angles. Expand outward from what proved out.
Given a loser, kill and replace. Do not optimize a structurally weak ad. Money spent trying to rescue a bad angle is money not spent finding the next winner.
Fatigue response. When frequency is high and metrics are decaying, refresh creative rather than raising budget. More spend on a tired ad accelerates the decay.
Output: the next test batch as a matrix (same shape as Mode 1), with each row tagged as an iteration of a winner or a new bet.
Client-type modifiers
| Ecommerce | Content / creator | |
|---|---|---|
| Primary metric | Cost per purchase, ROAS | Cost per subscriber or lead, engaged views |
| Winning creative bias | Product demo, UGC, offer-forward | Creator-native Reels, story-forward, hook-heavy |
| Destination | Product or landing page | Profile, subscribe page, lead form |
| Advantage+ | Advantage+ Shopping in play | Usually manual or Advantage+ audience only |
| Diagnostic emphasis | CTR to landing conversion to ROAS | Hook and hold rate, cost per engaged action |
Hand-offs
- Landing page is the break: landing-page-match-scorer.
- Account-wide economics or breakeven ROAS: the Meta unit economics skill.
- Organic-native creative production: social-goddess.
- Campaign structure, budget, and scaling: the Meta Campaign Architect skill.
- Production: Canva for statics, Pletor for UGC video.
Output rules
- Never use em dashes. Use commas, periods, hyphens, or pipes.
- Lead with the verdict or decision, then the evidence, then the fix. Do not bury the answer.
- Give specific, liftable outputs: filled matrices, shot lists, named fixes. Not generic best-practice advice.
- When you diagnose, name the single broken stage. Resist listing five things that "could" be wrong.