LLM Best Practices

Generative engine optimization guide

3 min read

View markdown source

Overview

Generative engine optimization (GEO) is the discipline of structuring content so generative AI engines cite it inside their answers. The generative engines are ChatGPT search, Perplexity, Google AI Overviews, Gemini, and Claude. GEO is the umbrella term; answer-engine-optimization is the off-Google subset and ai-overviews is the Google-SERP subset. This page defines GEO and lists the levers; the pillar llm-seo-best-practices frames where it sits, and discoverable-by-ai-assistants gives the procedure.

GEO optimizes for citation share, not rank position

Classic SEO optimizes for a position in a ranked list of links. GEO optimizes for inclusion in a synthesized answer with an attribution link. The differences that follow.

  • The unit is a quoted claim plus a citation, not a blue link at position N.
  • There is often no scroll and no second result; the engine names one to five sources.
  • The metric is citation share across a prompt set, not average rank across a keyword set.

The levers that move citation share

GEO rewards content a model can lift without rewriting. Four levers do most of the work.

  • Atomic, extractable claims. Self-contained sentences with a named entity, number, or version get quoted verbatim; hedged multi-clause sentences get paraphrased and lose the link. See answer-first-content.
  • Entity grounding. Name the primary entity in the title, H1, and first sentence, and disambiguate it with sameAs markup so the model maps the page to the right concept. See structured-data-for-ai-crawlers.
  • Authority signals. First-hand experience, a named author, and outbound citations to primary sources raise the odds a model trusts the page. See e-e-a-t.
  • Crawler access. A page the engine cannot fetch cannot be cited; allow the answer-engine bots in robots.txt.

GEO and SEO are the same page done well

A page that wins GEO usually ranks too, because entity clarity, fast load, and a direct answer help both systems. Do not build a separate “GEO page.” Harden the page you already rank and add the extractable-claim structure on top.

Measure GEO directly

Rank trackers do not see AI citations. Measure where the citation appears.

  • Run a fixed prompt list weekly through ChatGPT, Perplexity, and Gemini; record which prompts cite the domain.
  • Filter server logs for OAI-SearchBot, PerplexityBot, and Claude-SearchBot; a fetch spike usually precedes a citation by 24 to 72 hours.
  • Track the link, not the impression. A mention with no click-through is worth less than a cited link.

Pitfalls

  • Chasing GEO with FAQ schema on a page that does not answer the questions. The classifier ignores markup that the visible content does not back. See structured-data-for-ai-crawlers.
  • Publishing trend-chasing pages that spike for a week. Evergreen definitions and comparisons accrue citations for months.
  • Ignoring hallucination risk. If the model cannot find a clean claim, it may invent one and attribute it to you; a clear claim is the defense.

Graph View

Backlinks