LLM Best Practices

How to make content discoverable by AI assistants

3 min read

View markdown source

Overview

To make content discoverable by AI assistants, publish atomic pages in topical clusters, title each with the literal query it answers, open with a self-contained answer, ship complete metadata and JSON-LD, index everything in /llms.txt, and follow every rule yourself. This page is the step-by-step playbook behind the pillar llm-seo-best-practices; it is also the exact method used to build this site. For the implementation mechanics on a static site, see write-llm-friendly-content.

Step 1: Organize into topical clusters

Group pages into clusters with one pillar per topic and atomic single-concept children. Pillars link down; children link up and across. The cluster signals topical authority that one long page cannot, and each atomic page is a clean chunk an assistant can retrieve and quote. See content-clusters.

Step 2: Target the literal query

Make the title and H1 the exact phrasing a user or model would type. Capture variant phrasings as aliases so the page resolves under several query forms. Write a description that carries the query phrasing plus a one-line answer, because descriptions surface in results and get pulled into AI summaries.

Step 3: Open answer-first and quotable

Lead each page with a direct answer to its core question, self-contained enough to stand alone when an assistant lifts the chunk. Name the subject; do not rely on the title for context. Keep the lead declarative and specific. See answer-first-content for the full pattern.

Step 4: Ship structured metadata

Give every page complete frontmatter and JSON-LD so crawlers parse it as structured data, not prose. Use TechArticle for reference pages, BreadcrumbList for position, and FAQPage only where the page is genuinely Q&A. See structured-data-for-ai-crawlers.

Step 5: Publish llms.txt and the discoverability files

Ship /llms.txt as the LLM-facing index: every page, one line each, - [Title](url): summary., grouped by category. Add /llms-full.txt for the concatenated corpus and /ai.txt for the usage policy. Generate them from frontmatter so they never drift. See llms-txt and discoverability-files. Allow the answer-engine crawlers in robots.txt, or there is nothing to cite.

Step 6: Eat your own dog food

The site must pass every rule it publishes. Run a self-audit that derives its checklist from the site’s own pages, then grade the site against it. When the playbook is published, the loop closes: any page that violates it becomes a visible bug. Route assistants to a self-description page from site chrome; see for-ai-agents and the /llm-info footer link.

Verification

  • Fetch /llms.txt and confirm every page is listed with an accurate summary.
  • Open three pages in a text browser; confirm the first sentence answers the title’s question without context.
  • Validate JSON-LD with the schema.org validator and Google Rich Results Test.
  • Filter server logs for GPTBot, ClaudeBot, and PerplexityBot; rising fetches are the leading indicator of citations. See ai-search-optimization.

Graph View

Backlinks