How Perplexity Picks Its Sources, and Why It Matters More Than Its Size

Perplexity answers every question with numbered citations, which makes it the one major engine where you can watch retrieval happen in the open. For anyone doing AEO work, that transparency is worth more than the engine's market share: patterns you verify on Perplexity usually predict behavior you cannot observe directly inside ChatGPT or Gemini.

The mechanics

Perplexity runs its own crawler (PerplexityBot) and maintains its own index, supplemented by search partnerships. When a question arrives, it retrieves a candidate set, ranks passages for relevance and trustworthiness, and generates an answer where every sentence is grounded in a numbered source. If a claim cannot be grounded, it tends not to survive into the answer.

Watching thousands of retrievals across categories, consistent patterns emerge:

• It loves structured, sectioned pages. Clear headings, short answer-first paragraphs, tables, and lists get cited disproportionately. Long unstructured essays get skipped even when they contain the answer.
• It weights freshness visibly. For commercial questions, recently published or updated sources appear far more than archival content, and its answers shift week to week as new sources enter the index.
• It cross-checks. Claims that appear in multiple retrieved sources get stated as fact. Claims from one source get attributed ("according to..."). Brands corroborated across several sources get described confidently, which reads as a stronger recommendation.
• It leans on communities and review platforms for opinions. For "best X" and "is X good" questions, Reddit threads, review aggregators, and comparison sites dominate the citation slots. Vendor sites get cited for facts about themselves, almost never for claims of being the best.

What this means for your program

First, Perplexity is your test bench. Before investing a quarter of effort on a content or citation strategy, check whether the pages you are trying to influence actually appear in Perplexity's citations for your buyer prompts. It is the fastest feedback loop in AI search.

Second, the vendor-site ceiling is real. Your own site can win the factual citations (pricing, features, who you serve), and it should. But the recommendation slots belong to third parties, which is why citation and mention campaigns are half of any serious AEO program.

Third, structure is not cosmetic. The same content, restructured into direct answers under clear headings, measurably gains citations. That is the cheapest win available to most brands.

The leading-indicator argument

Every engine is converging on retrieval plus citation: ChatGPT cites more than it did, Google's Overviews cite by design, and enterprises increasingly demand grounded answers. Perplexity simply arrived at that architecture first and shows its work. Optimize for the engine you can observe, and you are, in effect, optimizing for where the others are heading. The community-source pattern in particular is covered further in why Reddit shows up in AI answers.