The Job
- You are paid for shipped ads that win, not for research, analysis, or taste. Output is the metric.
- Examples saturate the brain — consumption has diminishing returns. When inputs exceed output, stop consuming and write.
- Ideation and judgment are the skill; writing is the last 10-20%. Let AI write, you decide.
Core Mental Models
- Angle is the argument, format the delivery: reformatting a loser won't save it, but a proven message earns a real test in each new format.
- The hook IS the targeting — broad audience, creative self-selects who stops.
- A sale costs ~100 persuasion units; awareness + sophistication set the bill, surfaces split it (ad, page, VSL).
- Demonstrability decides mechanism: showable result = proof + identity lane; invisible result = mechanism carries everything.
- One idea per ad. The competing idea isn't cut, it becomes the next ad.
- Andromeda collapses ads over ~60% similar into one entity — a tweak is not a test.
Volume & Diversity Doctrine
- Spread bets like Battleship: diverse concepts across segments, awareness levels, and sources — never 10 shots in one cell.
- Budget 5-6 losers per winner. A 15-18% hit rate is the game working, not failing.
- Steady cadence beats binges: new concepts every week, refresh before fatigue (2-4 weeks).
- 60-70% of volume on unaware/problem-aware — it is harder to write, which is why it is underfished.
- Pull from multiple spy sources; one source optimizes one local maximum.
- Trunk before leaf: confirm awareness + core desire first; hook and copy tweaks are wasted on a wrong trunk.
- An untargeted segment is the biggest unlock — mediocre copy wins where there's no competition. Mine reviews to find it.
Winner Doctrine
- Don't copy — beat it. Model the structure, then out-charge, out-specify, or out-prove the original.
- 70/30: 70% of each batch iterates YOUR proven winners, 30% hunts new angles.
- Iterate winners one CASH dimension at a time: new hook, new angle, new style — never a full rewrite.
- A winning MESSAGE scales through new formats and actors; a winning AD just fatigues.
- Kill on thresholds, not vibes; never before 72 hours; judge aggregates, not single-ad CPA.
Judgment Bar
- Shrug or shake — if you're comfortable reading it, it's dead.
- Would you repeat it to a friend? Do you need to know what happens next?
- Name the belief the ad shifts in one sentence, or there is no ad.
- Skeptic pass: at every claim ask 'why does that follow?' — close the gap or delete the claim.
- The best hook is usually buried mid-draft. Find it, drag it to the front.
- A shake needs all four: hyper-relevant, hyper-specific, unscratchable curiosity, emotionally vicious. Missing one = shrug.
⚡ When Stuck
- Drowning in inputs → close the spy tools, write 10 ads from what's already in your head.
- Can't pick between two ideas → ship both as separate ads; the auction is the judge.
- Nothing feels good enough → you're judging drafts against shipped winners; ship the best 3 and let data grade them.
- Account stalling → check concentration before quality: same angle everywhere means one bet, not twenty.
- Unsure what to do next → open the Ad Engine Process Map and re-enter the pipeline where you left it.