LLMs cite what search engines surface
Perplexity and ChatGPT browse mode use live web search (Bing for ChatGPT, a hybrid index for Perplexity) to ground responses. Claude and Gemini use trained-in knowledge plus, in some configurations, web search. In all cases, the candidate set of sources the LLM chooses from is heavily weighted toward what the underlying search engine ranks for the query.
The implication: you cannot meaningfully optimise for AI citation without also ranking in conventional search. Sites that appear consistently in AI responses almost always rank in the top 5–10 organic results for the underlying queries. Fix the search ranking first; the citation follows.
Make the answer extractable in the first paragraph
LLMs preferentially cite passages that directly answer the query in a self-contained way. A page that buries the answer in paragraph eight, after extensive brand storytelling, is structurally worse for citation than a page that opens with a clear definitional answer.
The pattern that consistently earns citations: question-as-heading, two-sentence direct answer in the first paragraph, then expansion. This mirrors the structure that wins Featured Snippets in Google — the same content engineering pays off in both surfaces.
Author markup, expertise signals, and brand mentions
LLMs weight source credibility heavily, and the credibility signals they read are largely the same E-E-A-T signals Google uses: named authors with verifiable credentials, organisation-level expertise pages, dated content with revision history, and citations from authoritative third-party sources.
Brand mention frequency across the open web is a particularly strong AI-citation signal. Sites that are mentioned often (with or without a link) tend to be cited more frequently. This is one reason digital PR and earned coverage are now AI-search levers, not just link-building tactics.