LLM Best Practices

LLM SEO best practices

3 min read

View markdown source

Overview

LLM SEO is the practice of structuring a site so large language models retrieve, quote, and cite its pages, not just so a search engine ranks them. It shares the technical base of classic SEO but optimizes for a different unit of success: a citation with a link back inside an AI answer. This is the pillar for the LLM/AI-search cluster. It links down to the atomic playbooks: generative-engine-optimization for the definition and engine landscape, discoverable-by-ai-assistants for the step-by-step method, answer-engine-optimization for off-Google answer engines, and ai-overviews for the Google SERP feature. For optimizing for specific AI search engines (Perplexity, Google AI Overviews, ChatGPT Search, Bing Copilot), see ai-search-optimization.

Build topical clusters, one pillar per topic

LLMs and search engines both reward topical authority, which a single long page cannot signal. Structure the topic as a cluster.

  • One pillar page that frames the topic and links down to its children.
  • Atomic single-concept notes that each answer one question in 400 to 700 words.
  • Children link up to the pillar and across to siblings, so every page is one hop from its neighbors.

This is the same shape as content-clusters. A cluster of 12 atomic pages out-cites one 5,000-word guide because each page is a clean retrieval target.

Target the literal query in the title and H1

Make the title the exact phrasing a user or model would search. “LLM SEO best practices” beats “Thoughts on optimizing for the new search.” Capture variant phrasings as frontmatter aliases so the page is reachable under more than one query form, without diluting the canonical title. See keyword-research.

Open answer-first so the model can quote the chunk

LLMs extract discrete chunks and quote them. Lead each page with a direct, self-contained answer to its core question that names its subject and does not lean on the title for context. Follow with one specific fact per sentence: a number, a version, or a named entity. See answer-first-content for the pattern in full.

Ship structured metadata and JSON-LD

Crawlers parse structure faster than prose. Give every page complete frontmatter (title; a description carrying the query plus a one-line answer; tags; date) and JSON-LD: TechArticle for reference pages and FAQPage only where the page is genuinely question-and-answer. See structured-data and structured-data-for-ai-crawlers.

Treat llms.txt as the LLM-facing sitemap

Ship /llms.txt listing every page with an accurate one-line summary, and /llms-full.txt concatenating the corpus. Models read these to decide what to fetch without crawling the HTML tree. Keep them generated from frontmatter so they cannot drift. See llms-txt.

Eat your own dog food

The strongest LLM-SEO signal is a site that visibly follows the rules it publishes. Pass every rule on this page on this page: complete frontmatter, answer-first opening, dense internal links, JSON-LD, and an entry in /llms.txt. A site that contradicts its own advice loses authority with both crawlers and readers.

Pitfalls

  • Blocking AI crawlers in robots.txt while wanting AI citations. Allow GPTBot, ClaudeBot, and PerplexityBot first.
  • Keyword-stuffed pages with no answer. The classifier skips them; one useful page beats five thin ones.
  • Treating LLM SEO as separate from SEO. The same entity work and citation hygiene win both. See e-e-a-t.

Graph View

Backlinks