In late 2024 Perplexity quietly shipped Pages — a feature that lets anyone publish a research article assembled from cited sources. It looked like a B2C tool. It turned out to be a teaching tape on how Perplexity decides who gets cited.

We spent a couple of weeks generating Pages on different topics — SaaS comparisons, technical SEO, AI-search definitions — and watching which underlying sources Perplexity surfaced. The patterns that emerged weren't mysterious. They were almost embarrassingly aligned with the SEO fundamentals everyone already knew about.

Who Perplexity cited (and who it didn't)

Across roughly 40 Pages we generated, the top three citation slots overwhelmingly went to:

  • Pages with a clear, direct first paragraph. Not a hook. Not a context-setting introduction. The first paragraph answered the question. If a page opened with "Imagine for a moment that…" it got skipped in favor of a competitor that opened with "X is a Y for Z that does W."
  • Pages with a short quotable sentence early. Often the second or third sentence of the body. A factual claim with a subject, a verb, and a concrete number or named entity. Perplexity lifted these almost verbatim.
  • Pages already ranking in the top 10 organically. Perplexity is not running a parallel index. It is fetching from search results — usually Bing — and re-ranking. If you were not on page 1 already, you were not in the candidate pool.

Who didn't get cited

  • Sites with heavy interstitials, cookie banners, or paywalls. We noticed several cases where a clearly relevant article was skipped in favor of a slightly less relevant one with a cleaner crawl.
  • Pages where the first 300 words were headers, navigation, and marketing language. The substantive content was further down, and Perplexity seemed to read shallowly.
  • Brand-new content (published in the past few days). Pages still seemed to need a couple of weeks of index time before Perplexity would consider them.

What this implies for how to write

If you want to be cited by AI engines — not just Perplexity, the same patterns showed up across Pages, ChatGPT-Search, and Bing Chat at the time — there are three things you can do today:

  • Open with the answer. Move your hook, your framing, your context-setting paragraph below the first answer. Or cut it entirely.
  • Plant quotable sentences early.Sentence two or three of your body copy should be a standalone factual claim worth being lifted. "X is the only [category] that does Y for Z customers." "Y costs $Z and takes W weeks." Concrete, declarative.
  • Earn page-1 rank first. AI visibility runs on top of classic SEO. If you have to choose between AI-search optimization work and bread-and-butter SEO, do the SEO first.

Six months later

Perplexity has since updated Pages and changed parts of how citations are assembled. The specifics shift; the underlying logic does not. AI engines re-rank what classic search retrieves, and they re-rank on excerptability and trust signals. If you write in a way that reads well to a human skimming for an answer, you write in a way that reads well to a model deciding what to lift.

The pages we noticed Perplexity citing in late 2024 are, by and large, still the pages Perplexity cites in mid-2025. Hygiene compounds.