Internal links: crawlable, descriptive, in context

Google's documented position: every important page should be linked from at least one other page, links should be crawlable <a href> elements, and anchor text should be descriptive, reasonably concise and relevant. It publishes no ideal number of internal links per page. The skill finds pages that fail the first rule and fixes anchors, without inventing quotas.

Inputs

• The site URL (public, with an XML sitemap; links compares the sitemap with what it crawls).
• Optionally a query+page export (gsc) to choose which pages deserve links first.
• Pages the user owns and may edit.

Steps

1. Find orphans and weak pages:

   npx sitetiller links https://example.com --pages 100 --json

   Output: sitemapUrl, listed, crawled, orphans and weak, each row {url, linkedFrom, anchors}. linkedFrom counts crawled pages linking to it. weak means linked from 1 or 2 crawled pages. The crawl is a spread sample of the sitemap, so a page listed as orphan is "not linked from the sample". Raise --pages and re-run before acting on important pages. The command fetches one page at a time with delays; do not run several at once.
2. Rank candidates. Use impressions from gsc (striking for positions 4-20) or business value. A page with queries at positions 4-20 and no links is the first target.
3. Check the page itself:

   npx sitetiller page https://example.com/guides/invoice-template --json

   Fields: internalLinks, links[] (href, internal, anchor). Look for empty anchors, generic anchors ("click here", "read more"), image links without alt, and links that are not <a href> (JavaScript-only, onclick, span). Google cannot reliably follow those.
4. For each target page, find 3-5 existing pages where a reader of that page would plausibly want the target. Pick pages on the same topic first (topic-clusters), and pages that already have traffic.
5. Write the link into a sentence where it helps the reader, then use anchor text that:
   • describes the target page and makes sense read alone (Google's own test),
   • is reasonably concise, relevant to both pages, and varies naturally,
   • is not a repeated exact-match keyword. Example: instead of click here, write see the <a href="/guides/invoice-template">invoice template with payment terms</a>.
6. Add navigation links where structure calls for them: breadcrumbs, hub page lists, related-pages blocks written by hand or from data, not by keyword string match.
7. Re-run links after the edits. Orphans should drop to zero for the pages touched. Track the effect with measure-rollback: pages that gained links against matched pages that did not, 28 days.

Anchor review table

Without spam

• Link only where the destination helps the reader at that moment.
• No hidden links, no links in footers stuffed with keyword lists, no auto-inserting links on every occurrence of a word.
• Do not add links to pages the user marked noindex or that redirect.
• Do not chain links back to back or put many links in one block.
• Pages not worth linking (thin, duplicate, outdated) should be improved, merged or removed instead (cannibalization, content-decay).

Output to the user

1. Orphans and weak pages with linkedFrom and the crawl coverage (crawled of listed).
2. For each target: the source pages, the sentence, the anchor and the HTML to add.
3. Anchors to rewrite, with current and proposed text.
4. Measurement plan and a reminder that the crawl is a sample.

Guardrails

• Do not state a target link count per page. Google gives none.
• Do not edit pages the user did not name. Provide the proposed text for them to apply.
• Do not use nofollow on internal links to "sculpt" signals; use it only for untrusted external sources, with sponsored for paid links and ugc for user-submitted links, per Google.
• Treat orphan results as sampling output until confirmed with a larger crawl or the user's own CMS data.

References

• examples/links.json: real output of this skill's command on sitetiller.com (2026-11-13), see examples/README.md.
• references/sources.md