Copy Analysis — reading headlines, hooks and claims

How to judge the words of an ad (static headline, video hook, body line, stat, offer line), how to classify them, how to bank them, and where the compliance traps are. For writing full video scripts the master document is /root/claude-vault/knowledge/copywriting/COPY-CONSTITUTION.md; this file is the analysis lens.

---

1. The seven truths (judge every line against them)

1. You sell a desire, not a product.
2. You never create desire; you channel it.
3. The market beats the copy (a starving crowd beats a brilliant ad).
4. The customer is the best copywriter (their words beat yours).
5. One ad, one idea: one sub-avatar, one angle, one promise, one mechanism, one action. Two promises don't add up, they compete.
6. Specificity sells; generalities are discounted. ("600 mg, 31 formulas tested, 12,400 subscribers" is believed; "powerful, premium, revolutionary" drags everything down.)
7. Only the numbers judge — spend, CPA, ROAS, and month-2 rebill for subscriptions.

---

2. Static headline tests (apply to every headline you bank or brief)

80 % of a static's result is wording (Crafter). The best wording comes from review mining and golden nuggets, not from the copywriter.

---

3. Headline pattern library

3.1 Universal patterns (H1–H16, from the blueprint skill)

3.2 Observed winning headline families in supplements (verbatim examples, testosterone market)

3.3 Time-contrast amplifiers (Crafter static #8)

Amplify the pain with one of: time · effort · money spent · consequences if nothing changes · comparison with intense efforts · repetition of the problem · a quantified visual measure · felt injustice. Then step week by week using the exact SECONDARY symptoms ("heavy, swollen legs in week 2"), never generic claims.

---

4. Video hooks

4.1 The triple test (one failure = throw it away)

1. Does it name the audience? 2. Does it imply a benefit? 3. Does it open a curiosity that isn't closed?

4.2 Nineteen hook types (Crafter, adopted in the Constitution)

1 Direct benefit · 2 Humour · 3 Audience call-out · 4 Direct objection ("Can this really…?") · 5 Answer an objection ("…so we tested it") · 6 Pain point · 7 Question (ideally one that recurs in your comments) · 8 Big claim (about the mechanism/input, never a body-result number) · 9 Fast claim · 10 Authority · 11 Before/after (verbal, or of an object) · 12 Comparison with the category · 13 Destroy a limiting belief · 14 Information · 15 Novelty · 16 Exclusivity · 17 Belief reversal ("I thought it was a scam until…") · 18 Solution call-out ("don't buy another X before…") · 19 Call-out + curiosity, with a physical scroll-stop.

Six classic leads by awareness: Offer (most aware) · Promise (product-aware) · Problem–solution (problem-aware) · Secret (named hidden mechanism) · Proclamation · Story (unaware). One lead per script.

4.3 Hook anatomy

• Second 1 = capture (movement, rupture). Second 2 = selection (the sub-avatar and their situation). Second 3+ = sobriety.
• Three layers vary separately: image (first frame/motion), on-screen text, voice. The feed starts muted: the first word READ isn't the first word SAID. Sound-off test.
• Intensity is its own variable: factual → dramatised → shock. The best is rarely the strongest.
• The hook is targeting: a problem-specific hook stops fewer, better people. Judge on spend/CPA.
• A hook must meet an existing state ("she already dreads bedtime"); if you can't write that state, you're educating — and paying for competitors.

4.4 Verbatim hooks worth studying (bison market)

• "If you think two capsules can't flatten a beer belly you've had for 10 years, watch what happens inside your gut over the next 90 days."
• "If your gut feels like a bag of concrete you carry around all day…"
• "Your wife says snack less. Your doctor says it's just your age. They're both partially right."
• "If you're swallowing 12 supplements every morning and your gut still hasn't moved, the answer isn't another bottle."
• "The cheaper organ supplement is the better one. That shouldn't be possible."
• "These two bottles look almost identical," (Primal Edge, authenticity UGC)
• "I quit TRT, started Mars Men. Got leaner than I ever did on injections." (Mars Men, 149 days, 45 variants) Pattern: condition callout ("If you…") + a felt metaphor or a contradiction + an open loop.

---

5. Claims, stats and proof lines

• Stats are benefit claims phrased like the template's stat, with the survey footnote: "92 % reported clearer minds" → "93 % saw their beer belly shrink" (12-week customer survey). Use only stats the product sheet allows.
• Specificity moves from the result to the input (Constitution): "600 mg per capsule", "31 formulas tested", "12,400 active subscribers", "$1.60 a day" — believed and compliant. Body-result numbers ("lose 12 kg", "+40 % testosterone") are the risky kind.
• Numbers in a source ad are never copied as our facts. Replace with our inputs or remove.
• Proof placement: right after the mechanism, where the viewer asks for it.
• Deadly sincerity: a real flaw before the benefits ("It takes three weeks, not three days") raises belief in everything else.
• Duration is the mechanism: staged benefits (day 7 / 30 / 90); the subscription explained as a protocol phase.
• Who notices (identity): the wife, the colleague, the kids — derived from real reviews.
• Weird credible numbers that worked for competitors: 1979, 40 years of research, 400,000 years, 2,000 lb vs 350 lb, $1,750, "14 carrots / 24 oysters / 12 Brazil nuts / 9 ribeyes or 2 capsules". Credible because specific — but each must be TRUE for our brand before use.

---

6. Offer lines and CTA

• One CTA, one action; say what happens after the click.
• One booster at most (time, stock, scarcity, social proof, guarantee, bonus) — only if real. On subscriptions, no fake scarcity (recruits deal-hunters who churn at the second charge).
• The guarantee is the climax, restated against the angle: "Flat gut or your money back", "Chest back or every penny back", "She'll notice or you won't pay".
• Reason-why for any discount (production surplus, packaging change, new market test). Naked discounts aren't believed.
• Change the price reference: the $120 consultation, the four dead bottles, the daily coffee, $/day — never another supplement.
• Statics: never mention autoship/subscription mechanics; use "50 % off for life" style framing.

---

7. Compliance — the two-level policy

Level 1 — Landing pages / website (Stripe/Shopify-safe): structure-function language only ("formulated to support healthy testosterone*" + FDA disclaimer). Never cure/treat/block as hard claims on-site. Ad-level claims are not carried to the landing page without a landing-page check.

Level 2 — Ads: aggressive is allowed and expected (the founder will say if it isn't aggressive enough), but the following stay off-limits in OUR ads even when competitors use them:

Observed in competitors and NOT to copy: fake media partnerships (NY Post / Thrillist co-branded listicles, Mars Men), "doctors profit from your symptoms" conspiracy, celebrity endorsements we don't have, "+60 % T" result numbers, unverifiable "results in 24 hours".

When analyzing a competitor line, always add the compliance flag column: OK / risky (why) / banned (why) + the nearest compliant rewrite with the same function. Compliance is a slot filler, not a final check: in the transplant study it was the most frequent change (36 hits, 10/10 ads).

---

8. Copy diagnosis grid (use on any ad that "doesn't work")

---

9. Which AI to use for copy (study, 30 Sep 2026)

Study: ~8,200 X + Reddit posts, 820 relevant, 344 from credible pros (files: /root/workspace/copy-ai-study/quotes.md, workflows.md, stats.json).

Findings:

• Usage: Claude dominates (~5:1 among pros) for hooks, advertorials, emails, VSLs — because of Projects/Skills and review mining.

• Raw text quality: since July, many top operators rate GPT-5.6 Sol then GPT-6 Astra (Sept) above Claude for tight direct-response copy (Stefan Georgi: Astra "probably the best out of any frontier model"; Dan Shipper: Astra main writer, Sonnet 5.5 better at revision; Lorenzo Pravata ranks Sol above Fable for native ad copy). Complaint about Claude: too cautious / wordy / "Ted Talk" for aggressive DR. Opus 5.5 / Sonnet 5.5 (late Sept) regain some praise; little data yet.

• How pros actually work: research first (cluster 100–500 reviews into pains/desires/objections, pull exact phrases) → one Project per product (brand doc, ICP research, winners, top VSL transcript) → skills for the process (hook skill: extract the winning angle, write 9 hooks across relevance / curiosity / emotion, cut the weak) → an anti-slop rules file (banned words, rhythm, own voice samples) → blind A/B of models on the same brief.

• Skeptics: raw output is "AI slop" without research and winners; audiences now spot ChatGPT-style AND Claude-style scripts.

House recommendation: Claude for research, decoding and structure (this skill); test Astra vs Opus/Sonnet 5.5 on the same brief for the final copy lines; always feed the model the brand's winners and verbatim customer language. Plain chat writes better prose than coding-agent system prompts.

---

10. Banking copy from research (format)

Every line banked must carry its source. One row per line:

Categories (blueprint report): Pain & mechanism · Comparison · Product · Story · Urgency · Video hooks. Top 5 headlines are chosen from the bank and stay near-verbatim, adapted only for our facts (price, offer, guarantee, no competitor names).