Set Up SEO

Discoverability as a checked property, not a vibe. Audit the served HTML, fix in order: crawlability → metadata → structured data → sitemap/robots → answerability.

1. Gate — is this a public page?

Run the gate from ../_shared/page-types.md. App surfaces behind auth: stop, wrong catalogue. For this skill the gate doubles as the first audit: if the main content fails the view-source test, that is finding #1 and everything else waits — see ./crawlability.md.

2. Audit current state

curl -s "$URL" -o /tmp/page.html
grep -oiE "<title>[^<]*</title>" /tmp/page.html
grep -ciE "<meta name=\"description\"" /tmp/page.html
grep -ciE "rel=\"canonical\"" /tmp/page.html
grep -ciE "property=\"og:(title|description|image)\"" /tmp/page.html
grep -ciE "application/ld\+json" /tmp/page.html
curl -s "$ORIGIN/robots.txt"; curl -sI "$ORIGIN/sitemap.xml" | head -1

Audit each template (home, article, product), not just one URL. Apply only what's missing; a conforming page is a no-op. (The -c counters count matching lines, not tags — minified heads collapse to 1 — and the patterns assume double-quoted attributes; treat them as presence checks, not totals.)

3. Metadata per page

• <title> — unique per page, the page's promise, brand last ("Outcome the visitor gets — Brand"). ~60 chars is a display heuristic: Google truncates by pixel width and may rewrite the title; it is not a rule.
• <meta name="description"> — the snippet pitch containing the page's one job. Google often substitutes its own snippet; write it for the click, not for a length.
• <link rel="canonical"> — absolute URL, on every public page (self-referencing is correct); one canonical per duplicate-reachable content.
• Open Graph / Twitter — og:title, og:description, og:image (1200×630), twitter:card — this is what the link looks like when shared.
• All of it must be in the served HTML (step 1), not injected client-side.

4. Structured data (JSON-LD)

Pick types from the table in ./structured-data.md; the hard rule is schema mirrors visible content, never invents it. Example for an article page:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Exactly the visible h1",
  "description": "Matches the meta description",
  "datePublished": "2026-05-01",
  "dateModified": "2026-06-10",
  "author": { "@type": "Person", "name": "The visible byline" }
}
</script>

FAQPage only for real, visible FAQ sections — Google stopped showing FAQ rich results on 2026-05-07 for all sites, so the markup earns no SERP feature; it is optional (./structured-data.md). No HowTo markup: that rich result went away in 2023.

5. sitemap.xml + robots.txt

# robots.txt — at the origin root
User-agent: *
Allow: /
Sitemap: https://example.com/sitemap.xml

• Sitemap: every canonical, public URL — no redirects, no noindexed pages, no params. Generate at build; lastmod only if it's real.
• Submit once via Search Console (the old ping endpoints are retired); after that the Sitemap: line is discovery enough.
• Disallow is not noindex — to de-index, serve <meta name="robots" content="noindex"> on a crawlable page (./crawlability.md).
• AI crawlers are a policy decision, not a default. Training bots and search/answer bots are separate user agents — allow or block each on purpose (./crawlability.md). Landing pages that want to be cited keep the search bots allowed.

6. Answer-engine readability

Answer engines quote pages that answer cleanly. The moves are content structure, not tags:

• A section heading phrased as the question users ask, followed immediately by a direct answer in 2–3 sentences — elaboration after, never before.
• Visible author, published/updated dates, and (for health/finance/legal topics) a named reviewer — trust signals machines and humans share.
• Enumerable facts as lists or tables — extractable beats prose.
• One idea per paragraph; the page answers its <title>'s promise.

This is the generic core of GEO/AEO; project-specific scoring belongs in a rubric (../audit-content-quality/SKILL.md). No special tag, file, or schema is known to make answer engines cite a page — llms.txt included (./crawlability.md). Clear, visible, crawlable content is the whole lever.

7. Verify

curl -s "$URL" | grep -c "ld+json"        # ≥1
curl -s "$ORIGIN/robots.txt" | grep -ci "^Sitemap:"   # 1

Paste the URL into validator.schema.org (0 errors) and, for Product/Article/Breadcrumb, the Rich Results Test. Fetch with JS disabled: title, description, content all present. Share the URL in a chat client: the OG preview renders.

References

• ./crawlability.md — the view-source test, the empty-shell symptom, Disallow vs noindex, remediation pointers.
• ./structured-data.md — type table, the mirror rule, the FAQ caveat, validation.
• ../_shared/page-types.md — the gate and the priority inversion.
• ../audit-content-quality/SKILL.md — rubric-driven scoring on top of this baseline.
• ../../frontend/set-up-document-head/SKILL.md — in-app (SPA) head management; it handles tabs and a11y, not crawlability.