How Perplexity picks the pages it cites
Perplexity describes itself as an answer engine: it searches the web in real time for each question, summarises what it finds and links the sources with numbered citations. It does not publish a ranking formula, but its engineers have described the system. Its own index tracks over 200 billion URLs. Pages are parsed and split into self-contained passages, each retrieved and ranked on its own; retrieval mixes keyword and semantic search, and a final model reranks the candidates. Pages from authoritative domains are kept in its fastest storage.
The index is built by its crawler, PerplexityBot. A second agent, Perplexity-User, may visit a page when a user asks about it. We tested what that means in practice on 13 October 2026, with a request logger on a server of ours with nothing in front of it. Asked in a new thread to open three URLs that could not exist in any index, Perplexity sent no request for any of them and answered within a second that the fetch had failed. PerplexityBot did arrive about ten seconds after the first question, from an address in Perplexity’s published range, and fetched only robots.txt and the homepage. ChatGPT, asked the same, fetched the page within about five seconds.
Perplexity SEO starts before the question is asked: the page has to be in Perplexity’s index already, and the passage that answers has to be the most usable one it finds there. Asking Perplexity to read your page tells you nothing about whether it can; your server log does.
What Perplexity actually cites
On 22 October 2026 we put 60 questions that buyers of gambling, crypto and AI-search marketing ask (who are the best agencies, how do I choose one, what does it cost, who can launch us in a market) to Perplexity through its API. Every answer cited ten pages. We sorted all 600 by what kind of page they are.
| Page type | Citations | Share |
|---|---|---|
| Roundup or directory (“best X agencies”, listings) | 206 | 34% |
| Service page | 175 | 29% |
| Article or guide | 48 | 8% |
| LinkedIn, X and other posts | 43 | 7% |
| Press release or news | 40 | 7% |
| Homepage | 34 | 6% |
| Official docs, reference or PDF | 10 | 2% |
| Other | 44 | 7% |
Method: Perplexity sonar through the API, US, one answer per question, 22 October 2026; 600 citations, 478 distinct URLs on 312 domains. Page type by host and URL path, checked by eye on a sample. On crypto questions roundups were 94 of 200 citations; on gambling questions service pages led with 77 of 200.
The most cited domain was one marketing agency’s site, mostly its own “best agencies” roundups: 32 citations on 23 questions. LinkedIn came next with 32, most of them articles listing the best agencies, then a handful of agencies’ own sites with 12 to 14 citations each. Perplexity also cited content Google’s spam policies would call third-party pages on a borrowed domain: for “What should I ask an SEO agency before hiring them for a gambling site?”, one source was a casino-agency buyer’s guide uploaded to a US university’s WordPress folder.
When the question is “who should I hire”, Perplexity builds the answer from roundups and from the sellers’ own pages. Being listed, fairly, on the roundups it already cites matters as much as your own page.
Users’ questions look the same. For 36 questions crypto users ask before choosing an exchange, wallet or protocol, 69% of the pages Perplexity cited were ranked lists or comparisons, and 6% were the own site of a product it recommended. More in our crypto SEO guide.
Ranking in Google is not the same as being cited
On 23 October 2026 we asked Perplexity five questions about Perplexity SEO and pulled Google’s US top 10 for the five matching searches the same day. The two lists barely meet.
| Question put to Perplexity | Pages cited | In Google’s top 10 | On a top-10 domain |
|---|---|---|---|
| What is Perplexity SEO? | 10 | 1 | 1 |
| How does Perplexity choose which sources to cite in its answers? | 10 | 1 | 3 |
| How do I get my website cited by Perplexity? | 10 | 0 | 1 |
| Does ranking in Google help a page appear in Perplexity answers? | 10 | 0 | 1 |
| How do I check whether Perplexity cites my website? | 10 | 0 | 0 |
| All five | 50 | 2 | 6 |
Google searches: perplexity seo, how to rank in perplexity, perplexity ai seo, perplexity optimization, how to get cited by perplexity; US desktop top 10, 34 distinct URLs. Perplexity sonar through the API, US. One read of each: a direction, not a rate.
Across the 60 buyer questions the overlap was wider but still small: 17% of the cited URLs were in Google’s US top 10 for at least one of the category’s 44 head terms, and 51% were on a domain that held one of those slots. A domain Google trusts is more likely to be read; the page Perplexity cites is usually a different, question-shaped page on it.
Larger studies point the same way with different numbers. Ahrefs found that 28.6% of the URLs Perplexity cited for 15,000 long-tail prompts were in Google’s top 10 for the same query (August 2025), against 6–9% for ChatGPT, Gemini and Copilot. The 60% overlap figure still quoted on ranking pages comes from a 2024 press release with no published method.
Make sure Perplexity can read you
- robots.txt. Check that PerplexityBot is not disallowed, by name or by a blanket rule for unknown bots. Perplexity says a blocked site’s text is not indexed, though its domain, headline and a short summary may be, and that a change takes up to 24 hours to apply. Perplexity-User, which acts on a user’s request, generally ignores robots.txt.
- The firewall and the CDN. Bot protection can challenge a crawler that robots.txt allows. Perplexity publishes the IP ranges of both agents and advises allow rules that match the user agent and the IP range together, since anyone can send the user agent string. In August 2025 Cloudflare reported Perplexity reaching sites that blocked it through an undeclared crawler; Perplexity disputed the attribution.
- Do not judge access by a test request. A request from your laptop with Perplexity’s user agent tells you what your firewall does to that string, not what it does to Perplexity. Server logs showing PerplexityBot requests, and your pages appearing in answers, are the evidence.
- Content in the HTML. Text that appears only after scripts run, a login or a consent wall may not be in what a crawler stores. Keep the answer in the page’s HTML.
How to write a page Perplexity cites
- Answer the question in the first lines. A self-contained first paragraph that a reader, or a model, can quote without the rest of the page, then sections that each answer on their own. Perplexity’s engineers describe ranking passages rather than whole pages; that is a description of the system, not a published ranking rule, but it is why the section is the unit to write.
- One question per page. In the asker’s words, in the title and the main heading. Most of the pages Perplexity cited in our run were roundups and service pages, each built around one list or one service.
- Facts nobody else has. Your own measurements, prices, rules read at the source, with the method and the date next to them. A page that repeats what ten others say has no reason to be the one cited.
- Name your sources and date your page. Link the primary source for every claim and show when the page was last checked. Both let a reader verify the answer, which is the point of a citation.
- Tables for comparisons. Rows a model can lift one at a time: the option, the fact, the date.
- Behave like a source it labels. Perplexity marks some cited sites Government, Academic or Trusted, rating the whole site on whether it corrects its mistakes, says who wrote each piece and keeps news apart from advertising. It says payments do not affect the labels, and it does not say the labels change ranking; the same habits are what make a page worth citing.
- Markup is a help, not a lever. Article and breadcrumb markup make the page easy to parse. There is no published evidence that markup changes which pages Perplexity cites, and Google stopped showing FAQ rich results on 7 May 2026, so FAQ markup is no longer a display feature in Google either.
Be on the pages Perplexity already cites
Most citations in our run were not on the domain of the brand the answer named. For a recommendation question, the work off your own site is as large as the work on it.
- Find the pages. Run the questions your buyers ask, collect every cited URL, and count which domains recur. Those are the roundups, directories, posts and articles that decide the answer.
- Earn a fair place on them. Correct facts on directories, a real entry in roundups that accept submissions, coverage in the outlets that keep appearing, a byline on the professional network your buyers read.
- Keep paid placements labelled. Sponsored is fine when it says so. Fake reviews, planted comments and undisclosed paid mentions break platform rules and, in the US and the UK, consumer law.
How to check whether Perplexity cites you
- A fixed set of questions. The questions your buyers ask, in their words, a few dozen at most. The same set every time, so a change means something.
- Record the cited URLs, not just the names. Whether you are named and whether your page is a source are two different results. Write down the prompt, the date, the answer’s brands and every cited URL.
- Repeat on a schedule. Answers change from run to run and with the wording. One run is a direction; a monthly series is a measurement.
- Know which Perplexity you measured. We measure through Perplexity’s API, which returns at most ten sources an answer. The app a user sees can differ by model and mode; say which one a number comes from.
- Watch the logs and the referrals. PerplexityBot requests in your server logs show it is reading you; visits referred from perplexity.ai in analytics show its answers sending people.
What does not work
- Different content for AI crawlers. Showing bots something other than what users see is cloaking under Google’s spam policies, and it puts the same pages at risk in Google.
- Pages for every variation of a question. Google’s guide to its AI features says pages built for every fan-out variation of a query, mainly to manipulate AI answers, fall under its scaled content abuse policy.
- Files and tricks instead of content. Google says that for its own AI features you can ignore “chunking” content and AI text files such as llms.txt. Whatever Perplexity does with them, no file replaces a page worth citing.