Landing-page audit for paid ads — 5-dimension weighted health score

A Claude Code skill that grades any landing page used by a paid-ads campaign on five weighted dimensions — message match 25%, page speed 25%, mobile experience 20%, trust signals 15%, form quality 15% — producing a 0–100 health score, an A–F letter grade, and a prioritized quick-wins list with quantified impact ranges. Triggers on natural-language signals like "landing page audit" or "conversion rate review" (not a slash command). Output is a LANDING-PAGE-REPORT.md with bar-chart-style scores per dimension and platform-specific notes for Google, Meta, LinkedIn, TikTok, and Microsoft Ads. Built on public Google standards (Core Web Vitals: LCP <2.5s, INP <200ms, CLS <0.1) and the click-ID set for paid-ads attribution (gclid, fbclid, ttclid, msclkid). Reads only — does not run Lighthouse itself; the operator or Claude views the page (e.g. via WebFetch) and the skill scores from the rubric.

The skill replaces "we should improve our landing page" with ranked actions and impact ranges per fix. Sample quick-wins shipped with the skill: primary CTA above the fold = +15–25% CVR; reduce form fields to essential only = +10–20% CVR; trust badges near CTA = +5–15% CVR; image optimization = –1 to –2s load time. Page-speed rule of thumb baked in: ~7% CVR drop per 1s delay. The fixed weighted rubric forces the operator to argue with the dimensions — "trust is only 15% of CVR? Then stop spending two sprints on testimonials" — instead of argue about subjective taste. Demoable in 5 minutes against any live page.

1. Collect landing-page URLs from active ad campaigns. One page per audit; the rubric is built for atomic scoring.
2. Load the skill's reference files — the bundled benchmarks.md (per-industry CVR ranges) and conversion-tracking.md (per-platform click-ID and pixel checks).
3. Score message match (25%). Does H1, offer, CTA, and hero visual match the ad creative and keyword? Scored 0% (mismatch), 30% (partial), 60% (close), 100% (mirror).
4. Score page speed (25%). Apply Google's public Core Web Vitals thresholds: LCP <2.5s, INP <200ms, CLS <0.1; plus TTI <3s and page weight <2MB.
5. Score mobile experience (20%). Tap targets ≥48×48px, body text ≥16px, form fields keyboard-typed appropriately, full-width CTA, no horizontal scroll, no interstitial popups blocking primary action, tel: links where appropriate.
6. Score trust signals (15%) and form quality (15%). Trust: logo, named-and-photographed testimonials, security badges near CTA, case-study metrics. Form: length sweet spot 1–3 fields (highest CVR); 9+ fields lowest; multi-step for 5+; pre-fill, inline validation, specific submit copy ("Get my estimate" beats "Submit").
7. Audit consent banner and tracking. Flag any banner that covers the CTA, delays interaction >1s, or pushes content below fold. Verify Consent Mode V2 status; verify click-ID capture (gclid, fbclid, ttclid, msclkid); verify conversion-tracking fires.
8. Output the ranked quick-wins table. Each row: priority, fix, expected impact range. Cap at 5–7 fixes; ranking by weighted-impact × ease, not by dimension.

Weighted health-score formula (verbatim):

Landing Page Health Score
 = (Message Match × 0.25)
 + (Page Speed × 0.25)
 + (Mobile × 0.20)
 + (Trust × 0.15)
 + (Form × 0.15)

Sample quick-wins table (verbatim from skill):

| Priority | Fix | Expected Impact |
| 1 | Move primary CTA above the fold on all devices | +15-25% CVR |
| 2 | Reduce form fields to essential only (name, email, one qualifier) | +10-20% CVR |
| 3 | Add trust badges near CTA (security, guarantee, reviews) | +5-15% CVR |

Install (verbatim — keep author path runnable):

/plugin marketplace add AgriciDaniel/claude-ads
/plugin install claude-ads@agricidaniel-claude-ads

• Auditing without seeing the page. The skill scores from what it observes; if the operator pastes only a URL and Claude can't fetch it, the report is rubric-quality not page-quality. Confirm WebFetch succeeded.
• Consent-banner blind spot for non-EU traffic. The consent-banner section assumes EU/EEA context. For US-only traffic, skip the banner check or it drags the score artificially.
• Treating the 5-dimension weights as universal. Weights reflect average B2C-paid-ads benchmarks. For high-trust B2B (legal, healthcare, fintech), trust signals deserve more than 15% — adjust weights consciously, don't fight the rubric blindly.
• Quick-win impact ranges are ranges, not promises. "Move CTA above fold = +15–25% CVR" assumes the prior state was below-fold; a page already compliant gains zero from that fix. Compose fixes against the current state, not the maximum.