Marketing Pages Agent Skill

Marketing Pages

A rulebook, not a workflow: every rule below is simultaneously a constraint to

build against and a check to audit against. Classify the page first — every

public page belongs to exactly one page class, and the class decides its

indexing, linking, chrome, and wiring.

Page classes

| Class | Indexing | Reachability | Chrome |

|---|---|---|---|

| Campaign lander | noindex, follow; never in a sitemap | never linked from nav/footer | minimal: logo + ONE CTA, legal-only footer, no site nav |

| Product page | indexed | nav Product section and/or footer | full chrome |

| SEO tree page (programmatic: per-persona, per-industry, per-competitor) | indexed | footer link grid + hub pages | full chrome |

| Editorial (blog/guides) | indexed | header + footer | full chrome |

| Legal/utility | indexed | footer | full chrome |

• A page that is noindexed AND unlinked is a zombie — promote it to a

linked, indexed class or make it a campaign lander. Never leave the halfway

state.

• Campaign landers live in a dedicated URL prefix (e.g. /l/<slug>): one

startsWith covers blanket noindex and "never nav-linked" checks, and slug

collisions with real pages become impossible.

Campaign lander rules

Campaign landers are alternate main landers — one audience, one promise, paid

traffic. They trade SEO for message-match freedom; indexed near-duplicates

split authority and freeze copy that should be A/B-able daily.

• Message match end-to-end: ad copy → H1 → post-signup first screen, one

unbroken promise. Where the product supports it, the campaign identifier

flows through signup into onboarding and analytics, so the first-run

experience continues the promise and conversions attribute per lander.

• One CTA, repeated down the page; social proof beside it; friction-reducing

microcopy ("no credit card"); capture only what qualifies later (usually

just email).

• Static text/image hero — no full-viewport video or heavy JS above the fold.

LCP is a conversion input on paid traffic.

• A lander you cannot measure cannot be iterated: conversion-per-lander

(lead → signup → purchase) must be readable somewhere before ads spend.

• Killed campaign pages get no redirects: deleted routes 404, and the 404s in

Search Console afterward are the intended signal, not a defect.

Indexed page rules

Each of these has one owner in the codebase — find the current owner rather

than trusting remembered paths:

• The route appears in whatever closed-world route registry the site's gates

check against.

• Indexed ⇔ sitemapped, always both or neither: a noindexed URL in the

sitemap is a contradictory signal, as is an indexed page missing from it.

lastmod is honest (content date, not build time).

• Agent-readable surfaces, if the site ships them (llms.txt, markdown twins),

render from the same copy source the page renders from — never from built

HTML, never hand-drifted duplicates.

• Robots allows reviewed indexed trees for AI crawlers; disallows only

campaign landers and internal surfaces. robots, meta robots, and the

sitemap must agree on every URL.

• Every indexed page is reachable from header or footer — the footer link

grid is the crawl rail. Pages discoverable only via the sitemap sit in

"Discovered — not indexed" for weeks and receive no internal authority.

• Product facts (pricing, limits, tier names) derive from the single source

the product code owns — restated numbers drift silently and nothing turns

red.

• Internal links name paths, never origins, so a closed-world check can

verify every authored target resolves.

• Locale-prefixed links to pages that exist in only one locale must resolve

(redirect to the canonical locale, not 404) — sitewide chrome prefetches

every link from every page.

• Deleting or renaming a page deletes every reference in the same pass: route

registry, sitemap, agent-readable twin, links from kept pages, visual

baselines, and e2e/journey anchors (grep the test suites — a rename that

keeps local gates green can break a live journey hours later).

SEO rules

• Contextual in-body links from editorial to money pages move more authority

than footer anchors — add them where content genuinely relates.

• Comparison/competitor pages carry an honest visible updated-date and get

refreshed on a real cadence — a stale dated page is worse than an undated

one.

• Programmatic tree pages stay substantive (concrete jobs, specific copy) —

thin templated pages are doorway-page bait.

• Hub pages exist so leaves inherit authority: every tree leaf is linked from

its hub, and the hub from the site chrome.

Visual coverage rules

• Reusable sections and components have an isolated, deterministic fixture

(playground/storybook entry) that owns their visual coverage; the fixture

menu doubles as the component catalog new pages are composed from.

• Representative full-page screenshots stay at ONE key route per template.

Routes churn; fixtures are the stable oracle. New template → one new

representative route; new section → a fixture, not more route shots.

• Visual baselines are accepted only after an unprimed second opinion — eyes

primed by the change pass defects fresh eyes catch.

Audit sweeps

Auditing a site = evaluating every rule above, plus these cross-cutting

sweeps; each must come back empty or explained:

• Zombie pages (noindexed AND unlinked).
• Sitemap/robots/meta disagreements.
• Campaign lander linked from any indexed page, or present in a sitemap.
• Indexed page unreachable from header/footer.
• Component with no owner in the visual coverage system.
• Marketing copy restating a product fact that has a canonical owner.
• Broken locale-crossed links from sitewide chrome.