Mariana's toolkit for ANY conversion copy: ad creative, rail emails, teardown DMs, landing pages, VSLs. Two Claude skills installed globally (~/.claude/skills/copywriter-lite + ralph-wiggum-lite) — say "write copy" / "ralph wiggum this" to trigger. This note = the substance, so the framework survives even without the skills. Rio's voice rules always override: no em dashes, no AI-cliché words, his real talk. The framework is structure; the words stay his.
The 4-step pipeline (Schwartz → Georgi → Halbert → Bencivenga)
1. SCHWARTZ — pick the awareness level FIRST (mismatched opener kills conversion): Unaware → lead with the problem · Problem-aware → lead with agitation · Solution-aware → lead with the MECHANISM · Product-aware → lead with proof · Most-aware → lead with the offer. → Rio's barrels: local owners are mostly problem-aware ("slow months") or solution-aware ("tried ads once, didn't work" → mechanism!). The teardown IS a mechanism-lead.
2. GEORGI — RMBC before drafting: Research (real customer words — 11-audience) · Market (exactly who) · Big Idea (the ONE "wait, what?") · Concept (the angle that carries it). Any field empty → stop, research more.
3. HALBERT — write like a friend to ONE person: contractions, fragments, "you" 3x more than "we". Lead with conflict, not benefits. One sentence per line, white space is a weapon. Hook templates: "Here's something most [niche] owners will never tell you..." · "[Authority] hates this. But you'll love it." · "I almost didn't write this. Here's why I had to." · "Look, I'll be honest..." · "If you've tried [common fix] and it didn't work, this is for you."
Halbert DNA (from .brain/dna/halbert.md, kept in ~/Downloads/mega-brain-starter): write to ONE person · 80% of the time on the headline · write 50 headlines pick 1 · bucket brigades ("Here's the thing..." "Now check this out..." "But it gets better...") · hates adjectives (specific nouns instead), corporate speak, cleverness-over-clarity, unproven brags.
4. BENCIVENGA — 5 validation checks before shipping: Specificity (every vague word → a specific: "many"→"73%") · Proof (every claim gets a number/name/example or gets CUT) · Believability (would I buy this cold, or does it smell like marketing?) · Pace (read aloud, fix rhythm breaks) · Cut (can 20% go without losing meaning? then cut).
Ralph Wiggum 8-pass loop (the QA layer — run AFTER drafting)
1 Awareness match · 2 Hook (does line 1 stop the scroll?) · 3 Specificity · 4 Proof · 5 Voice (kill em dashes, triple structures, "leverage/robust/seamless", perfect symmetry) · 6 Pace (mix lengths, fragments, white space) · 7 Cut 20% · 8 Cold read (fresh eyes — would I act?). Loop back on any fail. Target 9.5+.
The Invisible Skeleton — 10 jobs, not frameworks (bonus lesson, added 7/6)
"The pros don't think in frameworks. They think in jobs." Every line of a winning ad performs ONE of 10 functional jobs, usually in this order:
- Pattern interrupt (stop the scroll — contrarian/surprise) · 2. Audience ID (who this is for) · 3. Problem statement (their named daily pain) · 4. Agitate (worse than they think) · 5. Villain/old way (blame an external factor — shifts responsibility off the reader) · 6. Mechanism/new way (the unique angle) · 7. Proof · 8. Benefit stack (3 concrete outcomes) · 9. Offer + risk reversal · 10. Urgency + CTA.
The reverse-engineering method (legal + ethical — copy structure, never words):
- Meta Ad Library (free): search the niche, filter ads running 30+ days (longevity = it's working), tag every line with its job number. Do 5 brands, one 60+ day ad each → the shared skeleton across all 5 = the niche's master skeleton → feed THAT structure + your product into the AI prompt.
- Master-prompt rules: "Replicate only the JOBS, in the same order" · one sentence per line, max 12 words · total ad under 90 words.
- Keyword recon: search specific terms not broad ("cold plunge" not "wellness"); 5,000+ results = saturated, <200 = emerging niche.
- Spy tools if ever needed: Foreplay.co ($49-99/mo, the standard). Brands worth studying for the DTC lane: Olipop, AG1, Liquid Death, Ridge, True Classic, Hims. Fits the stack as: the skeleton = WHAT each line does · copywriter-lite 4-master = HOW to write them · Ralph 8-pass = QA. Chain: skeleton → draft → Ralph.
Full-detail additions (Rio pasted the complete lesson 7/6 — deltas the first capture missed)
The 3 tagged example skeletons (study these shapes):
- GLP-1 clinic (local, pain-aware): "Houston women over 35, losing weight isn't about willpower" [interrupt+audienceID] → "hormone your doctor never tested for" [villain] → MDs+semaglutide+panels [mechanism] → "800+ patients, avg 22 lbs in 90 days" [proof] → "$49 this month, normally $299" [offer] → "FDA-approved, real MDs" [risk-rev] → "12 spots left" [urgency] → CTA. Skips agitate — audience already in daily pain.
- Cryo (audience doesn't want the service yet): starts at the pain they DO have ("Stretching isn't fixing your back — fix the inflammation") and bridges to cryo. When they don't want your thing, the MECHANISM does the heavy lifting — longest line in the ad ("3 min, drops temp 40°, shuts off inflammation at the source"). Proof = NFL/UFC + 1,400 locals. Offer = free first session + friend.
- Ecom skincare (dense): 9 jobs in 8 lines. "Stop using retinol at night. It's making your skin worse" [interrupt+villain] → "99% of retinols oxidize in 3 weeks" [problem+agitate] → encapsulated retinaldehyde 11x [mechanism] → 24K reviews + Vogue [proof] → "smoother in 14 days, fewer lines in 30, no peeling" [benefit stack] → 40% off [offer] → 60-day guarantee [risk-rev] → "sold out 3x" [urgency] → CTA. Visual does audienceID+interrupt, script skips ahead. Density rule: ecom = densest · local = slower, trust-heavy · info/high-ticket = slowest, needs a "big idea" up top. Same legos, different stacks. Master prompt ending (missed detail): "...write 5 different ads following this exact skeleton. Number each. After each, one short sentence on what's different about that variation." Keyword banks (Meta Ad Library, "Ad Content" toggle, US): GLP-1: semaglutide · tirzepatide · compounded GLP-1 · weight loss injection · ozempic alternative · stubborn belly fat · hormone weight loss. Wellness: cryotherapy · cold plunge · infrared sauna · red light therapy · recovery studio · chronic inflammation · biohacking · "first session free" (offer-based search — finds local tripwire structures!). Ecom: retinol serum · peptide serum · gut health · bloating · berberine ("natural ozempic") · non toxic cookware · as seen on shark tank. Hook-based (cross-niche): "stop using" · "the truth about" · "not your fault" · "why your X isn't" · "99% of" · "doctor approved". Scaled-ad filter workflow: sort oldest-first (60+ days active = making money; 90+ = a control) → same hook across 5+ creatives = their winner → "see ad details" (multi-platform = bigger budget) → brand's full library (50+ active ads = scaling, 5 = testing) → save to swipe file. Keyword result count = niche heat: 5,000+ hot · 500-2,000 healthy · <200 early/small. Homework loop: 5 brands → one 60+ day ad each → tag every line → find the COMMON skeleton across all 5 (that's the vertical's master skeleton) → feed to the prompt with your product → 5 ads → keep the strongest 2 = launch creative.
🎯 Rio's angle + YouTube-intent research prompt (saved 2026-07-07)
A niche → proven-angles → search-intent finder. Pairs with the Meta Ad Library recon above: Meta shows proven ad structures; this surfaces proven CONTENT angles + the exact 2-3-word terms people type when trying to fix the problem. The YouTube videos that already won on those terms = pre-validated attention = the angles/hooks to model into ad creative.
The prompt (copy-paste, swap the niche):
my niche: [INSERT YOUR NICHE HERE] what kind of angles are the most proven we could use it to target people online or the most angles that people would be the most interested in on this topic. What do you think are the primary keywords somebody would be searching that would make them, if they were trying to fix this problem in YouTube? What would they be searching so that we could see what the most popular videos are for each one of these topics in YouTube? Just give me, if they can only type in two to three words in a YouTube, what would they have, and what would be the biggest ones, the 80-20 ones?
The loop: run it → get angles + the 80/20 two-to-three-word terms → search each on YouTube, sort by views → the top videos = the proven angle/hook for that niche → feed the angle into the skeleton + 4-master pipeline above → write it in Rio's voice (model the angle, never copy words). Vortex/Ximena creative-research tool.
When to run what
- Quick DM/rail email: pipeline steps 1+3, then Ralph passes 2/3/5/7.
- Ad creative (once pilots run): full pipeline + full Ralph. Volume rule from the CMO doctrine: ship 10, kill 8, scale 2 — don't polish one ad for 3 days.
- Landing/sales pages: full everything + Rio reads it aloud.
- Pass 5's kill-list matches Rio's standing voice rules (04-voice + CLAUDE.md) — the frameworks agree with the house style.
🔍 Fidelity-audit additions (7/6 overnight — recovered from the verbatim archive)
Verified the 7/6 manual enrichment against the archive — the skeletons, keyword banks, filter workflow, and homework loop are all faithfully captured. Three small deltas:
1. The master prompt's DEFAULT skeleton is 9 jobs, not 10 — it deliberately drops Problem Statement, Agitate, and Benefit Stack and runs: interrupt → audience ID → villain → mechanism → proof → offer → risk reversal → urgency → CTA. (Matches the GLP-1 lesson point: pain-aware audiences don't need agitation.) Prompt input fields = PRODUCT / AUDIENCE / PAIN / OFFER / TONE, and one rule the note missed: "Write like a human texts a friend."
2. Annotated brand-study list (the note kept 6 names; the archive assigns each brand a specific skill to steal):
- Ecom: Hexclad = "old way vs new way" mechanism reveal · Magic Spoon = villain framing ("kids cereal but for adults") · Manscaped = humor as pattern interrupt · Lume = agitate masters ("problem you didn't know you had") · Tushy = boring product made scroll-stopping · True Classic = hook + offer stacking · Ridge = risk-reversal language.
- GLP-1/telehealth (maps to the med spa barrel): Hims & Hers = long-form curiosity hooks · Henry Meds = risk-reversal masters, low-friction signup · Ro = MD-led trust angle · Mochi = female-targeted soft sell · Found = "personalized/hormone testing" mechanism positioning.
- Local wellness (local-biz skeletons): Restore Hyper Wellness = "menu" ads for multi-service · iCRYO = city-by-city geo-hooks (the local-targeting play worth X-raying for Uprise niches).
3. Spy-tool alternates (note kept only Foreplay, correctly — but for reference): Atria ~$99/mo (AI ad-scoring, auto-surfaces hooks — for $5K+/mo spend) · Magic Brief ~$99/mo (UGC creative briefs) · AdSpy $149/mo (155M ads, 88 languages, search by CTA) · Minea $49-99/mo (drop a competitor URL → all their ads). Rio's Firecrawl + Playwright Ad Library sweep covers most of this free.