How does Perplexity choose which sources to cite?

Most AI assistants can answer from training memory alone. Perplexity is built the other way round. Every query triggers a live search, a shortlist of pages comes back, and the model composes its answer from what those pages say, attaching numbered citations to the claims it borrows. Citations are built into the product rather than bolted on afterwards. If your page is never retrieved, it cannot be cited. If it is retrieved but awkward to quote, it usually loses out to a page that is not.

This design matters commercially. In our July 2026 test of 30 brands, 27 were described with at least one materially wrong claim when the AI answered without live web search (full study). Retrieval-first engines like Perplexity largely sidestep that stale-memory problem, but only for brands whose pages are easy to find, easy to parse and worth quoting. So there are two gates to pass: make the retrieved shortlist, then be the source the model actually leans on. Everything below is about passing both.

First, the honest caveat. Perplexity does not publish its ranking factors, and nobody outside the company can tell you the weights it applies. What anyone can do is run the same queries week after week and watch which sources get cited. Four patterns hold up consistently.

Answer-shaped pages win. A page that opens with a direct two-sentence answer to the question, then expands, gets cited over a page that winds slowly towards its point. The engine is assembling an answer, and pages already shaped like answers make its job easy.

Freshness is visible and rewarded. For anything time-sensitive, such as pricing, comparisons and "best X in 2026" queries, recently updated pages with visible dates appear far more often than undated evergreen content, even when the older page is more thorough.

Domain trust does heavy lifting. Established publications, well-known review platforms, official documentation and active community threads recur constantly in citation lists. A brand-new domain with a perfect page faces a steeper climb than a mediocre page on a domain Perplexity already cites daily.

Clean, extractable statements get quoted. Sentences that state one verifiable fact (a price, a feature, a comparison) are liftable. Marketing copy ("a revolutionary all-in-one solution") gives the model nothing it can responsibly attribute, so it quotes someone else's plainer description of you instead.

Write answer-first. Under every heading, make the first paragraph a direct, self-contained answer to the question that heading implies, and treat everything after it as supporting detail. Use question-style headings that match how buyers actually phrase queries, because retrieval matches queries against your text, not your intentions.

Keep one claim per sentence. "Plan X costs £29 a month and includes unlimited seats" is citable. A 60-word sentence that mixes pricing, positioning and a caveat is not. Put a visible "last updated" date on the page and mean it: refresh the content when you refresh the date, because a stale page wearing a fresh date is exactly the kind of thing repeated retrieval exposes.

Then check the plumbing. Make sure your robots.txt does not block PerplexityBot, and that your key pages deliver their content in server-rendered HTML rather than only after JavaScript runs. A quick way to test both at once: paste the page's URL into Perplexity and ask it to summarise what is there. If the fetch fails or the summary garbles your content, you have found a retrieval problem that no amount of copywriting will fix. None of it is exotic. You are simply removing reasons for the retrieval step to skip you.

Run your ten most commercially important queries in Perplexity and write down every domain it cites. You will usually find the same handful recurring: a couple of review platforms, one or two industry publications, a comparison article or two, and often a Reddit thread. That list is your outreach plan, in priority order.

Being cited is frequently second-hand: Perplexity quotes what a trusted third party says about you, not what you say about yourself. So make sure those pages describe you accurately. Correct stale directory listings, pitch genuine inclusion in the comparison round-ups it keeps citing, and contribute honestly (and with disclosure) in the communities that keep appearing. Where a round-up describes you inaccurately, with old pricing or a dropped feature, email the author with the correction and a source; comparison pages get refreshed far more often than most brands assume, and the fix propagates into future answers. One accurate mention on a domain Perplexity already trusts can outperform ten new pages on your own site.

Do not guess. Pick ten real buyer queries, run them in Perplexity today, and record which sources are cited and what they say about your brand. Repeat monthly: citation sets shift as pages get updated and new sources appear, and the changes tell you which of your fixes are landing.

To baseline how AI engines describe your brand more broadly, run our free AI visibility checker. It currently tests Claude, with other engines rolling out. For the Perplexity-specific signals to watch as you work through the steps above, see our Perplexity visibility tracker page. The engines will not tell you their rules, but repeated testing will show you their behaviour, and that is enough to act on.