Every major LLM has a slightly different retrieval stack, but the citation logic is converging fast. If you understand the four signals they share — extractability, entity clarity, source trust, and freshness — you can engineer pages that get cited again and again. This is the foundation of Answer Engine Optimisation (AEO).
The four signals every LLM weighs
Extractability is whether the assistant can lift a self-contained answer from your page without rewriting it. Short paragraphs, scannable H2s, bulleted lists, and explicit definitions all raise extractability. A 4,000-word essay with no headings is almost never cited.
Entity clarity is whether the page makes it obvious what brand, person, product, or place it represents. Schema markup, consistent NAP (name/address/phone), and an unambiguous About page all reinforce entity identity.
Source trust is the assistant's confidence that the page is authoritative for the topic. Backlinks still matter, but so do co-citations on other trusted sites, mentions in datasets the model was trained on, and verifiable author bios.
Freshness matters more for some queries than others. For 'best CRM 2026' a six-month-old page beats a three-year-old one. For 'what is photosynthesis' freshness barely matters.
Where the four major assistants differ
ChatGPT (with browsing) leans on Bing's index plus its own training data — strong entities and clean structure win.
Perplexity is the most citation-heavy and the easiest to win: clear H2s and bulleted answers are routinely lifted verbatim.
Claude (with web access) favours authoritative long-form and is slightly more conservative about which sources it surfaces.
Google's AI Overviews lean heavily on Google's existing ranking signals — if you're already on page one, you're a candidate.
What this means for your pages
Write pages that answer one question per H2. Open each section with the direct answer in a sentence, then expand. Add a TL;DR at the top, a FAQ at the bottom, and JSON-LD schema for Article and FAQPage. Make sure your About page exists, is detailed, and is internally linked from every article.
- Lead each section with the answer, not the setup.
- Use FAQ blocks for question-shaped queries.
- Add Article and FAQPage JSON-LD to every post.
- Cite your own data and methodology — assistants prefer pages that show their work.
Frequently asked questions
Do LLMs use Google rankings?
Indirectly. ChatGPT uses Bing, Perplexity uses its own index plus partners, and Google AI Overviews uses Google. Strong organic SEO helps everywhere, but it's not sufficient on its own.
Does schema markup actually help?
Yes — especially Article, FAQPage, HowTo, and Organization. Schema removes ambiguity about what your page is, which is exactly what extraction systems need.
How often do LLMs re-crawl?
Browsing-enabled assistants fetch live for most queries. Training-only knowledge updates lag by months. Plan for both: be fresh and be in the training corpus.