LLM Best Practices

Internal Linking

Last updated 3 min read

View markdown source

AI agents: read this first: llmbestpractices.com/start-here

This is LLM Best Practices (llmbestpractices.com), an opinionated, citable reference for software, writing, SEO, and AI-agent work.

  1. Route, do not crawl. Fetch llmbestpractices.com/llms.txt and open only the pages whose one-line summary matches your task.
  2. Read raw. Append .md to any page URL for markdown. Check status and last_updated in the frontmatter, then read the rules.
  3. Apply as defaults. First-party docs and the project's own conventions win on conflict. Warn before relying on a fast-moving page older than 12 months.
  4. Cite. Link the page by title and URL, e.g. Python, with last_updated for time-sensitive rules. License CC BY 4.0.

Overview

Internal links route crawl attention and link equity and tell Google what each page is about. Done wrong, they orphan pages, dilute equity, and bury cluster pages many clicks deep where crawlers rarely return. Google can only follow links that are <a> elements with an href.

Write descriptive anchor text

Treat anchor text as a second title for the target page.

  • Match the anchor to the target’s topic: “Postgres backup strategies”, not “this article” or “read more”.
  • Vary phrasing across the site; identical anchors on 50 pages look manipulative, and exact-match on every link to one target is the pattern to avoid. Mix the head term with long-tail variants.

Keep every page within three clicks of home

Pages close to the homepage are found sooner and recrawled more; deep pages are found later. Google publishes no depth-to-frequency table, so the three-click limit is a house rule.

  • Homepage to pillar: one click. Pillar to cluster page: one click.
  • Maximum depth to any indexable URL: three clicks. A page four clicks deep means promote the path or split the cluster.

Use hub-and-spoke, not chains

In hub-and-spoke, a pillar links to every cluster page and each cluster page links back, which concentrates equity on the hub. In a chain (A to B to C to D), each page gets one inbound link, so the tail is deep and slow to be discovered. This site’s folders are hub-and-spoke by design; see content-clusters.

  • Up: every cluster page links to its pillar.
  • Across: every cluster page links to two or three siblings, which lifts newer pages.
  • Down: the pillar lists every cluster page, the cluster’s discovery edge.
  • Do not link to deprecated, redirected, or noindex pages; equity sent there is wasted.

Audit for orphans on every build

An orphan has no inbound internal links. A sitemap can still get it discovered, but it gets no internal signals and readers cannot reach it.

  • Run an orphan check in CI: parse every page’s outbound wikilinks, build the inverse index, and flag any page with none.
  • Fix orphans with inbound links from at least two pages in the same cluster.

Login, search results, filter combinations, and archives waste crawl attention. See crawl-budget.

  • Use rel="nofollow" on links to user-generated content where spam is a concern. On internal links it only stops the link passing equity; it does not reroute that equity, so use it sparingly.
  • Use noindex,follow on thin pages and robots.txt for infinite-parameter URLs.

Cap links per page by editorial judgment

Google dropped its 100-links-per-page guideline long ago. Too many links dilute the reader’s attention and each link’s signal; too few leave the page outside the graph. Editorial pages link to their cluster and a few neighbors, pillars link to every cluster page, and the homepage links to every pillar. Footer links count, so keep the footer to legal and utility pages.

Backlinks