LLM Best Practices

Answer engine optimization (AEO)

3 min read

View markdown source

Overview

Answer engine optimization (AEO) is the discipline of getting a page quoted as the answer inside a conversational AI engine: ChatGPT search, Perplexity, Claude, and Gemini. AEO is the citation-side sibling of ai-overviews, which targets the Google SERP feature specifically; this page covers the off-Google answer engines. AEO inherits the shared playbook in ai-search-optimization (atomic claims, entity surfacing, discoverability files) and adds the engine-specific rules below. The unit of success is a citation with a link back, not a ranking position.

Write the answer as the first sentence under a question-shaped heading

Answer engines extract the sentence that most directly answers the user’s prompt. Structure each section so that sentence is unmissable.

  • Use an H2 or H3 that phrases a real question or task: “What is INP?” “How to set a canonical URL.”
  • Open with a single declarative sentence that answers it in under 30 words.
  • Follow with one specific fact per sentence: a number, a version, a named entity.

Hedged, multi-clause sentences get paraphrased and lose the link. Short declarative sentences get lifted verbatim and keep it.

Allow the answer-engine crawlers, or there is nothing to cite

A page that blocks the fetcher cannot be quoted. AEO starts with crawler access in robots.txt.

  • ChatGPT search reads with OAI-SearchBot; user-initiated fetches use ChatGPT-User. Training uses GPTBot.
  • Perplexity indexes with PerplexityBot; user-initiated fetches use Perplexity-User.
  • Claude search uses Claude-SearchBot; user fetches use Claude-User.

Allow the search and user-fetch agents even if you disallow the training agents. See discoverability-files for the full token catalog and ai-search-optimization for the citation-hygiene rules.

Ground entities so the engine resolves the right thing

Answer engines disambiguate using named entities and their identifiers. Surface them so the model maps your page to the correct concept.

  • Name the primary entity in the <title>, the <h1>, and the first sentence.
  • Mark it up with DefinedTerm or Article JSON-LD, with sameAs pointing at the Wikipedia or Wikidata URL.
  • Keep one concept per page; a page that defines five things gets skipped for one that defines one.

This is the same entity work that wins ai-overviews; the markup earns no SERP feature on its own but it grounds the citation.

Measure citations directly, not through rankings

AEO does not show up in rank trackers. Measure it 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 citation with no click-through link is worth less than a ranked blue link.

Graph View

Backlinks