メインコンテンツへスキップ

ARC 2.0.0-beta.25

SEO, links, and accessibility

Use generated SEO and link checks as a floor, then finish with human accessibility inspection.

Web Device can assemble canonical and multi-locale hreflang links, Open Graph/Twitter metadata, and JSON-LD from site/page/content configuration. That automation makes the right structure available; it does not make the resulting page discoverable or accessible by itself.

Give each level its own metadata

ScopePut it here
Site-wide title template, description, site identity, sharing defaults.web/site.yaml seo: block
Page title/description and locale-specific variantpages/<name>/seo/ such as title.zh and description.zh
Content-object title, summary, date, and relationshipsthe object’s own content/front matter

A page can also choose an seo/og-type. Do not compensate for missing page or object metadata by putting every title in the site-level configuration.

A denied access shell (preview-shell or promo-shell) always sends X-Robots-Tag: noindex and a robots meta tag. Its description may only use a declared SEO description or excerpt, never a summary derived from the protected body. See Content access reference.

Know what automated checks cover

The static publishing doctor can block a static build with a missing build index, sitemap, robots file, or recognizable internal page dead link. Missing title is currently advisory rather than a block. The page-level SEO check is intentionally lightweight: it looks for title, description, H1 count, an alt attribute on img, and basic Open Graph fields.

That does not mean an image alt is meaningful, keyboard paths work, focus is visible, color contrast is adequate, external links work, or an embedded service endpoint is reachable. Those are author and reviewer responsibilities.

A practical acceptance pass

  1. Build the site and open the default locale plus every declared locale.
  2. Inspect title, description, canonical, hreflang, and a sharing preview on representative pages.
  3. Follow primary navigation and a content detail/list path; distinguish a known internal page from an external URL that cannot be statically verified.
  4. At wide and narrow widths, use keyboard navigation, inspect focus, read image alternatives, and look for overflow or clipped text.
  5. Review the page’s actual content hierarchy; one literal H1 is not the same as a coherent reader-facing heading structure.

These checks are intentionally small enough to repeat. For route formation, see Routes, navigation, and locales; for runtime/build failure categories, see Diagnose a site.