How Perplexity decides what to cite (not just what to mention)

We've already covered why Perplexity cites almost everything it answers and how to check what it's currently saying about you. Both of those posts stop at the same place: a checklist of what tends to earn a citation — recent, structured, third-party, quotable. True, and worth doing. But it skips the part that actually decides a close call: Perplexity doesn't cite every page it retrieves. For a real buyer question, it typically pulls in a shortlist of candidate pages first, then a separate step picks which of those actually make the numbered list under the answer. This post is that second step, worked through on one concrete example, because a checklist tells you what to build; a worked example tells you why the other page lost.

Retrieval finds candidates. A separate step ranks them.

Ask Perplexity a real buyer question and it doesn't grab the single best-matching page on the internet — it runs a broad first pass that pulls in a set of plausible candidates, then reranks that set before deciding what gets cited. That second step is the one that matters for you, because it's not scoring the same thing a classic Google ranking scores. Independent GEO-research write-ups on this pattern converge on the same directional finding, worth taking seriously even without a hard number attached to it: the reranking step weighs how directly a page answers this specific question and how current it is, more heavily than raw link-based domain authority. A page from a smaller site that states the comparison plainly and was published last month can out-cite a page from a much bigger, better-known domain that answers the question only in passing, buried in older, broader copy.

That's a genuinely different game than SEO ranking. Domain authority still matters elsewhere — it's still part of why a page gets crawled and indexed in the first place — but it stops being the tiebreaker once two pages are both sitting in the candidate set for the same question. At that point, the tiebreaker is: which page actually answers this question, in a form worth quoting, as of right now.

Two pages, one buyer question — a worked example

Here's an illustrative walkthrough, not a real brand or a real query-and-response transcript. Say the buyer question is "best CRM for a five-person sales team that hates data entry." Two pages both plausibly get retrieved:

Page A is a well-established CRM's homepage. High domain authority, years of backlinks, ranks well in classic search. But the page itself is a broad pitch — "trusted by thousands of teams," a features grid, no specific team-size framing, no direct comparison to anything, last meaningfully updated over a year ago.

Page B is a smaller competitor's dedicated comparison page, titled plainly "Best CRM for small sales teams (2026 comparison)," published six weeks ago. It states, in the first two sentences, which tools it's comparing and for what team size, names three real alternatives by name, and includes a short table on setup time and per-seat pricing.

Both get retrieved. Only Page B tends to get cited. Not because Page B's company is bigger or more established — by domain authority alone, Page A wins easily — but because Page B is a direct, current, structurally extractable answer to the exact question asked, and Page A is a generic pitch that happens to be reachable. The reranking step has something clean to lift from Page B: a specific claim, a specific audience, a specific date. From Page A it has only a vague match and no clean sentence to quote.

What this means for the pages you actually control

The practical takeaway isn't "publish more comparison pages" in the abstract — it's this specific reframe: don't compete on your domain's overall authority for a question this narrow. Compete on being the most directly on-point, most current, most extractable answer to the specific way a buyer phrases it. Concretely, that means:

• Match the question's shape, not just its topic. A page about "CRM features" is topically related to "best CRM for a five-person sales team." A page titled and structured around that exact framing is a much stronger candidate for the citation slot, because it's answering the question asked, not a broader one it's adjacent to.
• Keep a visible, real, recent date on comparison content, and actually revisit it — a two-year-old page with an old date on it is telling the reranking step it's stale, even if the content underneath is still accurate.
• Put the specific claim in the first two sentences. We've written about this discipline in more depth — the pattern that wins a citation is the same one that wins any AI-generated recommendation: a plain, checkable claim near the top beats brand-voice copy that takes three paragraphs to say anything specific.
• Check that you're even in the candidate set first. None of this matters if PerplexityBot can't reach the page — reranking only operates on pages that were successfully retrieved in the first place.

See it in your own report, not a hypothetical

The worked example above is illustrative on purpose — a real Perplexity answer for your actual buyer questions is the only version of this that matters to your business, and it changes with question phrasing and over time. Our State of AI Citations report shows the real, live-computed pattern across every domain we've scanned: which third-party source domains Perplexity and our other retrieval-capable engine actually cite most, recomputed on every page load from real scans, never a fixed snapshot. It's the same honest, no-fabricated-number discipline as the rest of this site — worth a look before you assume you know what's winning the citation for your category.

Then run it on yourself: the free scan puts 25 real buyer questions to Perplexity alongside ChatGPT, Claude, and Gemini, and shows you, per question, whether your domain made the citation list — and if it didn't, which page did. No signup, about a minute, and you'll be looking at your own version of "Page A vs Page B," not someone else's example.