How do AI engines decide which sites to cite?

No AI engine publishes its citation formula, so the honest answer has two layers: the qualifying conditions the engines document — your site must be crawlable by their bots and, for Google, indexed and snippet-eligible in ordinary Search — and the selection tendencies visible in their guidance and in observed citations: answer-first structure, datable freshness, and identifiable source authority.

The qualifier: anyone quoting a precise recipe — "pages with FAQ schema get cited 3x more" — is quoting a number no engine has published. What follows is only what the engines state in their own documentation, verified live on August 18, 2026, plus clearly-labeled observation.

What the engines actually publish

• Google (AI Overviews / AI Mode): "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary… The best practices for SEO remain relevant for AI features." The one hard condition: "a page must be indexed and eligible to be shown in Google Search with a snippet." No special schema, no AI text file required. (Google Search Central, "AI features and your website," verified 2026-08-18.)
• OpenAI (ChatGPT search): "OAI-SearchBot is used to surface websites in search results in ChatGPT's search features" — a site that disallows it opts out of being surfaced; GPTBot, the training crawler, is documented separately. ChatGPT-User fetches pages when a user asks. (OpenAI "Bots" documentation, verified 2026-08-18.)
• Perplexity: PerplexityBot "is designed to surface and link websites in search results" and webmasters should allow it "to ensure their sites appear in search results"; Perplexity-User fetches pages for user questions. Both are documented as not used for model training. (Perplexity crawler documentation, verified 2026-08-18.)

The selection layer — labeled as observation

Above the qualifying bar, four properties consistently distinguish cited pages, and each has a mechanism rather than a mystery. Answer-first structure: a generated answer is assembled from quotable passages, so a page whose first sentence completely answers a question can be used verbatim, while a page that buries the answer under windup requires reconstruction. Freshness you can date: retrieval systems handling time-sensitive topics need a date to trust; visible review dates and a schema dateModified supply it. Source authority: Google's long-published quality framework (E-E-A-T) and the engines' preference for identifiable sources reward named authorship, credentials, and claims tied to primary sources. Retrievable speed and cleanliness: a crawler budget spent waiting on an 8-second page is a crawl that ends early — one reason we treat the 99+ mobile PageSpeed bar as part of AEO, not separate from it.

Our first-hand evidence sits underneath this page's own advice: the two original studies we published — PageSpeed on 1,080 agency sites and a tracking-pixel scan of 1,286 agency sites — exist partly because engines cite data they cannot get anywhere else. Original, datable, citable numbers are the strongest single citation magnet we know of, and that claim is itself an observation from running the experiment, not a published formula.

Start with the gate you control: make sure the crawlers can get in, then make every page quotable. Which questions are worth building pages around depends on what you sell — the vertical guides cover Medicare and health, life, and commercial lines.