LLM Best Practices

How to write LLM-friendly content for a public site

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

Agents prefer a short summary over five paragraphs, raw Markdown over hydrated HTML, and named rules over loose advice. This guide retrofits an existing site with those properties. /llms.txt and /ai.txt are community proposals, and Google states that AI Overviews and AI Mode need no special files or markup, so treat the agent files as a convenience, not a ranking lever. The checklist below links to the page that covers each step.

Prerequisites

  • Stable URLs; agents cite them.
  • Frontmatter with at least title, summary, and category. The summary is the routing hint.

Steps

1. Write a one-sentence summary per page

Put it in frontmatter and reuse it for the meta description, the llms.txt note, and JSON-LD description, so the three never disagree.

2. Ship the agent index

Generate /llms.txt, /llms-full.txt, /llms-context.txt, and per-category /llms-full/<category>.txt from frontmatter in the build; see ship-llms-txt. If you publish /ai.txt, state real preferences and keep them consistent with robots.txt; the format is in ai-txt. Crawler allow and block rules still live in robots.txt.

3. Add JSON-LD

One entity type per page kind (usually Article), generated from frontmatter; see add-jsonld-to-static-site. HowTo and FAQPage no longer earn Google rich results, so ship them only if the markup is accurate.

4. Serve raw Markdown next to every page

Emit <slug>.md beside each <slug>.html at build time, and link it from llms.txt. If the repo is public, raw.githubusercontent.com URLs also work. Markdown costs agents fewer tokens than HTML and keeps code blocks intact.

5. Name the rules

Agents cite a named, numbered rule more accurately than a paragraph. State the rule first, then the rationale.

## Rule: five wikilinks per page
Every page links to at least five neighbors, so an agent reaches related pages in one hop.

6. Keep pages atomic

One topic per page; when a section outgrows the page, split it and link. Agents follow links more cheaply than they re-read long pages.

Verify it worked

curl -sI https://yourdomain.com/llms.txt | head -1                    # HTTP/2 200
curl -s https://yourdomain.com/foo | grep -o 'application/ld+json'    # present in the HTML
curl -sI https://yourdomain.com/foo.md | head -1                      # HTTP/2 200

Model check: give a model the page URL and ask for a two-sentence summary and the key rule. If the summary matches your frontmatter and the rule comes back verbatim, the page is agent-ready.

Common errors

  • /llms.txt lists a page twice. Give each page one category.
  • Summaries run past one sentence. Cut them; the summary routes, it does not teach.
  • No Markdown source is reachable, so agents parse HTML and lose structure.
  • ai.txt contradicts robots.txt. Decide the policy once and mirror it.

FAQ

Do I need llms.txt to appear in Google AI Overviews?

No; Google states that AI Overviews and AI Mode need no special files or markup, so treat /llms.txt and /ai.txt as a convenience for agents, not a ranking lever.

Should I serve Markdown versions of my pages for AI agents?

Yes; emit <slug>.md beside each <slug>.html and link it from llms.txt, because Markdown costs agents fewer tokens than HTML and keeps code blocks intact.

Is FAQPage or HowTo schema still worth adding?

Only when the markup is accurate; HowTo and FAQPage no longer earn Google rich results.

How do I test whether a page is agent-ready?

Give a model the page URL and ask for a two-sentence summary and the key rule; if the summary matches your frontmatter and the rule comes back verbatim, the page is ready.

Backlinks