How Perplexity Picks Citations: What It Documents and Shows

How Perplexity picks citations refers to the way it searches its own web index, scores small passages of each page against the question, writes an answer from the strongest ones and links them as numbered sources. Perplexity publishes that outline but no ranking formula, so everything beyond it is what site owners observe.

• Perplexity offers 2 search modes: Quick Search summarises indexed sources quickly, while Pro Search goes deeper and may ask the user follow-up questions (Perplexity's explanation of how it works).
• Perplexity's Search API returns a ranked list of 10 web results by default and up to 20, each with a snippet, a publication date and a last updated date (Perplexity's Search API reference).
• Perplexity names 2 bots: PerplexityBot surfaces and links websites in its results, and Perplexity-User fetches a page when a user's question needs it (Perplexity's crawler documentation).
• Perplexity says a robots.txt change can take up to 24 hours to take effect, and that Perplexity-User generally ignores robots.txt because a user started the request (Perplexity's crawler documentation).

What does Perplexity say about how it finds sources?

Perplexity says it searches the web in real time and builds each answer from what it finds. It runs its own index rather than borrowing another engine's results.

Perplexity's explanation of how it works describes a live search across articles, websites and journals, with language models that interpret the question and combine the most relevant findings into one answer. It does not say how relevance is measured or how much a source's reputation counts, so claims online about Perplexity's "ranking factors" often go beyond anything the company has written. Treat any precise list of them as a guess until Perplexity publishes one.

More detail comes from Perplexity's announcement of its Search API. Perplexity says the API gives developers access to the same global infrastructure that powers its public answer engine, an index covering hundreds of billions of webpages, with systems that process tens of thousands of index update requests every second. In other words, a page has to be in Perplexity's own index before it can be cited, and the crawler that puts it there is PerplexityBot.

That makes access the first filter, before any question of quality. A page that PerplexityBot cannot reach, because robots.txt disallows it or a firewall refuses it, cannot be scored at all. The one exception is Perplexity-User, which fetches a page when a user's question needs it, for example when someone pastes your address into the conversation, and may then link it in the answer.

How does Perplexity read and score a page?

Perplexity says it splits each document into fine grained units and scores every unit against the query. A page competes passage by passage, not only as a whole.

The Search API announcement describes two steps that happen before any answer is written. First, an AI content understanding module generates parsing logic for messy web pages, so that the useful text can be extracted from pages with untidy layouts. Second, the cleaned document is divided into small units, and each one is surfaced and scored on its own against what was asked.

The API reference shows the result of that process in plain fields. Each result comes back with a title, a URL, a snippet extracted from the page, and, where known, the date it was published and the date it was last updated, in a list that is already ranked. Developers can also set how many tokens of page text come back for each result.

For a business the lesson is concrete. A paragraph that only makes sense after the one before it scores poorly as a unit, because on its own it does not answer anything. A paragraph that names its subject, states the answer first and adds a specific detail is a complete unit, and it is the kind of text a scoring step can match to a question.

How does Perplexity show the sources it cites?

Perplexity shows each source as a numbered footnote that links to the original page. Readers can open each one to check the claim behind it.

Perplexity's own explanation says each answer includes numbered footnotes linking to the original sources, so that users can verify the information or read further. It does not say how a footnote is matched to a particular sentence, so the link between a claim and its source is something readers judge for themselves. In practice, owners see the same source numbered more than once when several sentences rely on it.

The footnote is where the business value lies. A citation puts your domain inside an answer someone is already reading, and a click lands on the exact page that supported the claim. A citation to a clear service page is therefore worth more than one to an old news item that happens to mention you.

Which citation factors are documented and which are only observed?

Documented factors are access, passage relevance to the query, freshness and stored page dates. Directness, specifics and primary sources are what owners observe in cited pages.

Factor	Status	Where it comes from
PerplexityBot can reach the page	Documented	Perplexity's crawler documentation: PerplexityBot surfaces and links websites in results
Small units of a page are scored against the query	Documented	Perplexity's Search API announcement
Freshness is a priority for the index	Documented	Perplexity names information staleness as one of the biggest failure modes for AI agents
Pages carry a publication date and a last updated date	Documented	Perplexity's Search API reference
Pages that answer directly are cited more	Observed	Owners' comparisons of cited and uncited pages
Specific figures, places and named sources help	Observed	Owners' comparisons; consistent with the scoring of small units
Comparison questions cite directories, lists and reviews	Observed	What owners see in Perplexity's numbered sources

The documented rows describe a mechanism; the observed rows describe outcomes people have seen. Observations come from small samples, and answers change from one run to the next, so treat them as working habits to test. If a guide gives you a precise list of Perplexity ranking weights, ask where the numbers come from, because Perplexity has not published any.

Domain reputation is the most common claim in that gap. Perplexity's explanation does not describe a reputation score, and the API's domain filter is a choice developers make rather than evidence of a ranking. Keep your effort on the factors you control and can check.

Why does freshness change which sources Perplexity cites?

Freshness changes citations because Perplexity treats stale information as a failure and records when each page was published and updated. A current, dated page gives it less reason to look elsewhere.

In the Search API announcement, Perplexity calls information staleness one of the biggest failure modes for AI agents and says it optimised indexing to make Perplexity a real time assistant. The Search API reference backs this with filters: developers can limit results to the past hour, day, week, month or year, or to pages published or updated after a given date. Perplexity does not say how its consumer answers apply those dates, but the index clearly stores them.

The useful response is honest dating, not constant redating. Show a "last updated" date on pages whose facts change, update it when the facts change, and keep the facts themselves current: prices, hours, service areas, rules and stock. A hypothetical bakery in Leeds that changes its Christmas order deadline should change the page and the date on the same day, so that a question about the deadline finds a page that is both right and visibly recent.

What can a page do to be the passage Perplexity picks?

A page wins the citation by holding the clearest self-contained answer to the question, backed by a specific fact, a date and a source. Write each important passage so it can stand alone.

Start with the opening of each important page. Put the customer's question in a heading and answer it in 40 to 60 words directly beneath, naming the business, the place and the key fact in the first sentence. A hypothetical plumber in Nairobi answering "how long does a tap replacement take" should give the usual time and what affects it, not a paragraph about the team's values.

Then add the evidence that makes the passage worth trusting. Link the regulator, standard or manufacturer behind a claim, use your own real figures, and keep each fact in plain text rather than inside an image. These are habits drawn from what owners observe, not documented rules, but they match how a scoring step works: a unit that contains the answer and its support is easier to rank than one that only hints at it.

To see how your own pages look as passages, scan your website with the AI Search Visibility Checker. For each page it shows the question the page can win and the passage an AI would quote, and gives an AI Citation Score out of 100, of which 30 points are for an answer to lift and 20 for evidence. It predicts how quotable a page is; it does not measure whether Perplexity cites it today. Access and mentions on other websites, the rest of Perplexity SEO, decide whether a well written passage ever gets the chance to be chosen.

Can you find out why Perplexity cited a competitor instead of you?

You can work it out by comparing the cited passage with your own. Perplexity does not explain individual choices, but the passages usually show the difference.

Ask Perplexity the question your customer asks, open the numbered sources, and find the exact sentences the answer relied on. Then put your best passage beside them. The usual differences are a first sentence that answers outright, a concrete number or place, a newer date, or a mention on a directory or list the answer used.

Repeat the test before deciding anything. Run the same question on several days and note which sources stay and which come and go, since a single answer is a sample of one. If your page never appears, check your server logs for PerplexityBot requests to that page; a page the crawler has never read cannot be cited however good it is.

Questions owners ask about Perplexity's citations

Does Perplexity cite the page that ranks first on Google?

Not necessarily, because Perplexity searches its own index rather than Google's results. Its announcement describes an index of hundreds of billions of pages built by its own crawling, so a page must be reachable by PerplexityBot to be cited.

Why does Perplexity cite a different source each time I ask?

Answers are generated afresh for each question, and the index is updated constantly, so the passages that score best can change between runs. Track the same questions over weeks and judge the pattern, not one reply.

Can Perplexity cite a page that blocks PerplexityBot?

Rarely, because PerplexityBot is the crawler that surfaces and links websites in its results. The exception is Perplexity-User, which may fetch and link a page when a user's question needs it, such as when someone shares the address.

Does the citation link always go to my homepage?

No, it goes to the page that held the passage the answer used. That is why each important question deserves its own page with a clear answer, so that the citation lands where a visitor can act.

Test one page against Perplexity this week

Choose one question your customers ask often and test the page that should answer it. One page tested properly teaches more than a rewrite of the whole website.

Ask Perplexity the question and record the numbered sources and the sentences each one supplied. Check your server logs or the AI Search Visibility Checker to confirm PerplexityBot can reach your page, then rewrite its opening as a 40 to 60 word answer with a real figure, a visible date and a linked source. Ask the same question again over the following weeks and note whether your page enters the sources, then apply what worked to your next most important page.