Paid-Traffic Landing Page skill
Researches a niche's pains and builds a single-screen, one-decision landing page for paid ad traffic (Meta/Google/YouTube) — headline-first, message-matched to the ad, with one CTA and a risk-reversal/trust element, nothing else.
by oathdriven·MIT license·★ 28 Stars on the repo·GitHub ↗
npx degit oathdriven/rymac-skills/skills/rymac-paid-traffic-landing-page#main ~/.claude/skills/rymac-paid-traffic-landing-pageChecked ·commit main
Inside this skill
SKILL.md · the author's own words# Paid-Traffic Landing Page Build a landing page whose only job is converting one ad click into one action. Headline quality is ~80% of the outcome; simplicity is the… Reference files (read at the step that needs them): - `${CLAUDE_SKILL_DIR}/references/niche-research.md` — 12 intake questions, the 20-minute pain-mining sprint, sophistication stages, pain→c… - `${CLAUDE_SKILL_DIR}/references/headlines.md` — awareness levels, 13 rules, 16 formulas, subheadline formulas, message match - `${CLAUDE_SKILL_DIR}/references/page-structure.md` — structure rules with conversion data, single-screen anatomy, CTA/trust rankings, mob… - `${CLAUDE_SKILL_DIR}/assets/page-template.html` — the single-screen HTML shell with `{{TOKENS}}` to fill (read it, fill it, save as the d… ## Workflow Copy this checklist and tick it off: - [ ] 1. Intake — parse `$ARGUMENTS`; collect what's missing - [ ] 2. Mine the niche (research sprint, parallel subagents)
Word-for-word excerpt, not a result. Read the full SKILL.md ↓
Files of Paid-Traffic Landing Page
Show the full text81 lines
Paid-Traffic Landing Page
Build a landing page whose only job is converting one ad click into one action. Headline quality is ~80% of the outcome; simplicity is the other 20%. Everything else is deleted, not designed.
Reference files (read at the step that needs them):
${CLAUDE_SKILL_DIR}/references/niche-research.md— 12 intake questions, the 20-minute pain-mining sprint, sophistication stages, pain→copy mapping${CLAUDE_SKILL_DIR}/references/headlines.md— awareness levels, 13 rules, 16 formulas, subheadline formulas, message match${CLAUDE_SKILL_DIR}/references/page-structure.md— structure rules with conversion data, single-screen anatomy, CTA/trust rankings, mobile rules${CLAUDE_SKILL_DIR}/assets/page-template.html— the single-screen HTML shell with{{TOKENS}}to fill (read it, fill it, save as the deliverable; don't rebuild the layout from scratch)
Workflow
Copy this checklist and tick it off:
- 1. Intake — parse
$ARGUMENTS; collect what's missing - 2. Mine the niche (research sprint, parallel subagents)
- 3. Diagnose awareness + sophistication
- 4. Write the copy elements, headline first
- 5. Build the page from the template
- 6. Run the conversion gate — fix anything that fails
- 7. Deliver page + copy doc with A/B alternates
1. Intake
Needed before writing: niche/avatar, offer, traffic source + the ad's promise (for message match), CTA goal (lead form / booking / click-through), brand name + accent color. Take what $ARGUMENTS and context give; ask the user only for genuinely unknowable items (the ad copy, the brand color). If no ad exists yet, note that the H1 should become the ad headline later — write them as one unit.
2. Mine the niche
Read ${CLAUDE_SKILL_DIR}/references/niche-research.md and run its sprint. Fan out 2–3 parallel research subagents (Reddit/forums · reviews of competitors or current solutions · Meta Ad Library/competitor headlines) and have each return verbatim quotes tagged pain / desire / objection / failed-alternative / trigger. Rank by frequency × emotional charge. Output: 1 dominant pain, 1 dream outcome, top 3 objections, top 2 failed alternatives — in the prospect's exact words.
3. Diagnose
- Awareness of the traffic (cold social ≈ problem/solution-aware; search ≈ solution-aware; retargeting ≈ product/most-aware) — sets which headline formulas apply.
- Sophistication of the market (stage 1–5 from competitor ads) — write one stage ahead of the loudest incumbent; never repeat a worn-out claim.
4. Write the elements (headline first)
Read ${CLAUDE_SKILL_DIR}/references/headlines.md. In order:
- Headline — write 20 candidates from the 3 best-fitting formulas; if ad copy exists, the verbatim ad promise is always candidate #1. Cut to the strongest 3 by the rules (≤8 words, specific, visualizable, outcome-led, prospect's own words). Pick one; keep the other two as A/B variants.
- Subheadline — mechanism + proof + kills the #1 objection. 10–30 words.
- 3 bullets max — ≤7 words each, specific numbers, mined language. Skip entirely if headline+sub carry the page.
- CTA —
[verb] + my + [outcome]("Get My Free Quote"). Never "Submit"/"Learn More". - Risk reversal — one line under the button, aimed at the #1 mined objection ("Free — no credit card" / guarantee / "Takes 30 seconds").
- Trust line — ONE element: stars+count, micro-testimonial with name + number, or 3–4 known logos. Only what's real — never fabricate reviews, counts, or testimonials; if nothing real exists, use effort reassurance instead.
5. Build the page
Fill ${CLAUDE_SKILL_DIR}/assets/page-template.html (copy it into the project/output directory first). Lead form → Variant B with the fewest fields follow-up can survive on (3 max); booking/click-through → Variant A. Set the accent color; keep it the only saturated hue on the page. No additions: no nav, no sections below the fold, no images unless one visual directly shows the deliverable/outcome, no external scripts except the ad pixel if provided.
6. Conversion gate
Verify every line; fix and re-check anything that fails:
- H1 restates the ad's promise (message match) — or is flagged to become the ad headline
- H1 ≤8 words, outcome-led, passes the 5-second test (what / for whom / what to do)
- Copy reads at 5th–7th grade level
- Exactly ONE decision: one CTA goal, zero other links (Privacy/Terms ghost-text footer only)
- Headline + subhead + CTA + risk line all visible in a ~390×844 viewport without scrolling; CTA ≥48px tall, full-width on mobile, never grey
- Form (if any) ≤3 fields, correct input types
- One trust element, specific and true, adjacent to the CTA
- No render-blocking or third-party requests; page is a single self-contained file
7. Deliver
Hand back: the HTML file path, the chosen headline with the two A/B alternates, and a short copy doc (all elements + the mined quotes that justify them) so ads can be written with the same language. Recommend one page per ad angle — offer to spin variants by swapping H1/eyebrow.
Rules
- The headline is the product. Never spend more effort on layout than on headline candidates.
- One page, one pain, one decision. Cut a second idea rather than fit it.
- Prospect's verbatim words beat anything invented — if a phrase wasn't mined or given, question it.
- Never fabricate proof (reviews, star counts, names, "as seen in").
- Delete before adding: when the fold is tight on mobile, bullets go before the CTA does.
| 1 | |
| 2 | name rymac-paid-traffic-landing-page |
| 3 | description Researches a niche's pains and builds a single-screen, one-decision landing page for paid ad traffic (Meta/Google/YouTube) — headline-first, message-matched to the ad, with one CTA and a risk-reversal/trust element, nothing else. Use when the user asks to "make a landing page for ads", "build a page for this campaign", "squeeze page", "lead capture page", "opt-in page", or wants to convert paid traffic for a niche. Make sure to use this skill for any ad-destination page; for multi-section marketing websites use website-builder or build-premium-website instead. |
| 4 | argument-hint [niche/offer, traffic source, and CTA goal — e.g. "HVAC owners, Facebook ads, book a call"] |
| 5 | |
| 6 | |
| 7 | # Paid-Traffic Landing Page |
| 8 | |
| 9 | Build a landing page whose only job is converting one ad click into one action. Headline quality is ~80% of the outcome; simplicity is the other 20%. Everything else is deleted, not designed. |
| 10 | |
| 11 | Reference files (read at the step that needs them): |
| 12 | `${CLAUDE_SKILL_DIR}/references/niche-research.md` — 12 intake questions, the 20-minute pain-mining sprint, sophistication stages, pain→copy mapping |
| 13 | `${CLAUDE_SKILL_DIR}/references/headlines.md` — awareness levels, 13 rules, 16 formulas, subheadline formulas, message match |
| 14 | `${CLAUDE_SKILL_DIR}/references/page-structure.md` — structure rules with conversion data, single-screen anatomy, CTA/trust rankings, mobile rules |
| 15 | `${CLAUDE_SKILL_DIR}/assets/page-template.html` — the single-screen HTML shell with `{{TOKENS}}` to fill (read it, fill it, save as the deliverable; don't rebuild the layout from scratch) |
| 16 | |
| 17 | ## Workflow |
| 18 | |
| 19 | Copy this checklist and tick it off: |
| 20 | |
| 21 | [ ] 1. Intake — parse `$ARGUMENTS`; collect what's missing |
| 22 | [ ] 2. Mine the niche (research sprint, parallel subagents) |
| 23 | [ ] 3. Diagnose awareness + sophistication |
| 24 | [ ] 4. Write the copy elements, headline first |
| 25 | [ ] 5. Build the page from the template |
| 26 | [ ] 6. Run the conversion gate — fix anything that fails |
| 27 | [ ] 7. Deliver page + copy doc with A/B alternates |
| 28 | |
| 29 | ### 1. Intake |
| 30 | |
| 31 | Needed before writing: **niche/avatar, offer, traffic source + the ad's promise (for message match), CTA goal (lead form / booking / click-through), brand name + accent color.** Take what `$ARGUMENTS` and context give; ask the user only for genuinely unknowable items (the ad copy, the brand color). If no ad exists yet, note that the H1 should become the ad headline later — write them as one unit. |
| 32 | |
| 33 | ### 2. Mine the niche |
| 34 | |
| 35 | Read `${CLAUDE_SKILL_DIR}/references/niche-research.md` and run its sprint. Fan out 2–3 parallel research subagents (Reddit/forums · reviews of competitors or current solutions · Meta Ad Library/competitor headlines) and have each return verbatim quotes tagged pain / desire / objection / failed-alternative / trigger. Rank by frequency × emotional charge. Output: 1 dominant pain, 1 dream outcome, top 3 objections, top 2 failed alternatives — in the prospect's exact words. |
| 36 | |
| 37 | ### 3. Diagnose |
| 38 | |
| 39 | **Awareness** of the traffic (cold social ≈ problem/solution-aware; search ≈ solution-aware; retargeting ≈ product/most-aware) — sets which headline formulas apply. |
| 40 | **Sophistication** of the market (stage 1–5 from competitor ads) — write one stage ahead of the loudest incumbent; never repeat a worn-out claim. |
| 41 | |
| 42 | ### 4. Write the elements (headline first) |
| 43 | |
| 44 | Read `${CLAUDE_SKILL_DIR}/references/headlines.md`. In order: |
| 45 | |
| 46 | **Headline** — write 20 candidates from the 3 best-fitting formulas; if ad copy exists, the verbatim ad promise is always candidate #1. Cut to the strongest 3 by the rules (≤8 words, specific, visualizable, outcome-led, prospect's own words). Pick one; keep the other two as A/B variants. |
| 47 | **Subheadline** — mechanism + proof + kills the #1 objection. 10–30 words. |
| 48 | **3 bullets max** — ≤7 words each, specific numbers, mined language. Skip entirely if headline+sub carry the page. |
| 49 | **CTA** — `[verb] + my + [outcome]` ("Get My Free Quote"). Never "Submit"/"Learn More". |
| 50 | **Risk reversal** — one line under the button, aimed at the #1 mined objection ("Free — no credit card" / guarantee / "Takes 30 seconds"). |
| 51 | **Trust line** — ONE element: stars+count, micro-testimonial with name + number, or 3–4 known logos. Only what's real — never fabricate reviews, counts, or testimonials; if nothing real exists, use effort reassurance instead. |
| 52 | |
| 53 | ### 5. Build the page |
| 54 | |
| 55 | Fill `${CLAUDE_SKILL_DIR}/assets/page-template.html` (copy it into the project/output directory first). Lead form → Variant B with the fewest fields follow-up can survive on (3 max); booking/click-through → Variant A. Set the accent color; keep it the only saturated hue on the page. No additions: no nav, no sections below the fold, no images unless one visual directly shows the deliverable/outcome, no external scripts except the ad pixel if provided. |
| 56 | |
| 57 | ### 6. Conversion gate |
| 58 | |
| 59 | Verify every line; fix and re-check anything that fails: |
| 60 | |
| 61 | [ ] H1 restates the ad's promise (message match) — or is flagged to become the ad headline |
| 62 | [ ] H1 ≤8 words, outcome-led, passes the 5-second test (what / for whom / what to do) |
| 63 | [ ] Copy reads at 5th–7th grade level |
| 64 | [ ] Exactly ONE decision: one CTA goal, zero other links (Privacy/Terms ghost-text footer only) |
| 65 | [ ] Headline + subhead + CTA + risk line all visible in a ~390×844 viewport without scrolling; CTA ≥48px tall, full-width on mobile, never grey |
| 66 | [ ] Form (if any) ≤3 fields, correct input types |
| 67 | [ ] One trust element, specific and true, adjacent to the CTA |
| 68 | [ ] No render-blocking or third-party requests; page is a single self-contained file |
| 69 | |
| 70 | ### 7. Deliver |
| 71 | |
| 72 | Hand back: the HTML file path, the chosen headline with the two A/B alternates, and a short copy doc (all elements + the mined quotes that justify them) so ads can be written with the same language. Recommend one page per ad angle — offer to spin variants by swapping H1/eyebrow. |
| 73 | |
| 74 | ## Rules |
| 75 | |
| 76 | The headline is the product. Never spend more effort on layout than on headline candidates. |
| 77 | One page, one pain, one decision. Cut a second idea rather than fit it. |
| 78 | Prospect's verbatim words beat anything invented — if a phrase wasn't mined or given, question it. |
| 79 | Never fabricate proof (reviews, star counts, names, "as seen in"). |
| 80 | Delete before adding: when the fold is tight on mobile, bullets go before the CTA does. |
| 81 |
Discussion
Alternatives
Browse more free Claude skills or everything in Marketing.