What earns a citation in Perplexity

Perplexity searches the web for the question, reads a set of pages, writes one answer and attaches numbered citations to the pages it used (Perplexity). Because it searches for nearly every query, the sources change as the web changes.

That makes Perplexity the most transparent AI engine to work with. You can see exactly which pages it used, and you can see the effect of your changes faster than in engines that lean on model memory. It also means there is no shortcut: to be cited, your page has to be found, read and judged useful for that specific answer.

Complex questions are often split into several smaller searches. A question like "best accounting software for a 10-person agency in Germany" may lead to searches on agency accounting needs, German tax compliance and pricing. Each search can surface a different page, which is why one answer often cites five or more sources. If you want the background on this splitting, Google calls its version "query fan-out" and our GEO guide explains why it matters.

Check this first, because nothing else matters if the answer is no. Perplexity uses PerplexityBot to build its search index and Perplexity-User to fetch pages a user asks about (Perplexity).

Three checks cover most problems:

1. robots.txt. Make sure neither bot is disallowed. Remember that a bot obeys only the most specific group that names it, so a named group needs its own rules (RFC 9309).
2. CDN and firewall. Bot protection can block AI crawlers before robots.txt is read. Look for 403 responses in your bot logs.
3. Rendering. Put the main text, prices and answers in the server-rendered HTML. Content that appears only after JavaScript runs may not be read.

Our AI crawler list has every user agent and a robots.txt template.

Cited pages tend to answer one specific question early, in plain language, with facts that can be lifted into a sentence. Look at the sources Perplexity shows for your own prompts and you will see the same traits again and again.

A direct answer near the top. The cited passage is usually the first sentence under a heading that matches the question. Pages that build up slowly to the point are cited less often than pages that state it and then explain.

Specific facts. Numbers with a unit, named standards, dates, prices, integrations and limits. "Supports SEPA and SWIFT payments in 34 currencies" is citable. "Flexible global payments" is not. The original GEO study found that adding statistics, quotations and citations to a page made it more visible in generative answers in its benchmark (Aggarwal et al.).

A clear scope. One page per topic beats one page that covers everything. If a buyer asks "X vs Y", a page titled "X vs Y" with a comparison table is a natural source.

Freshness where it matters. For questions about prices, versions, laws or "best of this year", recent pages have an edge. A visible "last updated" date with a real change behind it helps. Changing the date without changing the content does not.

Recognizable sources. Review platforms, established publications, documentation and community threads are cited often because they are already trusted for the topic.

Usually because their page answers the sub-question more directly, or because a third-party page that lists them, and not you, is the one being cited. Find out which case you are in before you act.

Run the prompt, open every cited source and sort them into two groups:

• Competitor-owned pages. Compare their page with yours on the same question. Is the answer higher up? Is it more specific? Does it have a table you lack?
• Third-party pages. Reviews, listicles, directories and forums. Check whether you are on them at all, and whether what they say about you is current.

The second group is often larger than teams expect. For many category questions, Perplexity cites "best X" articles and review sites more than vendor pages. Getting onto those pages accurately can matter more than rewriting your own.

Rewrite for the question, not for a keyword. Take a prompt you lose, write down the sub-questions Perplexity seems to search for, and make sure one clear passage on your page answers each one.

A practical checklist:

• Use the buyer's question as an H2, and answer it in the first sentence below.
• Put the key facts in text, not only in images or PDFs.
• Add a short table for comparisons, plans or specs.
• Name your product, category and audience plainly in the first paragraph.
• Link to your sources when you cite a number.
• Keep each page to one main topic, and link related pages to each other.
• Update the page when the facts change, and say what changed.

None of this is special formatting for AI. It is the same writing that helps a busy human reader find the answer, which is exactly what Perplexity is doing on that reader's behalf.

Work through the third-party pages Perplexity already cites for your prompts, starting with those that appear in the most answers. You are not trying to be everywhere, only on the pages the engine trusts for your topic.

• Review platforms. Claim your profile, complete it and ask real customers for reviews.
• "Best of" articles. Many are updated yearly. Send the editor a short, factual note on what you do and who you serve.
• Communities. Answer real questions as a named employee. Don't post fake recommendations; communities spot them, and they can backfire.
• Partners and integrations. Make sure partner directories list you with a current description.

Tricks aimed at the engine rather than the reader rarely work, and some backfire. Skip these:

• Keyword stuffing. Repeating the question ten times doesn't make a passage more useful. Perplexity reads the page; it isn't matching strings.
• Fake freshness. Changing the date without changing the content. Readers notice, and so do engines that compare versions.
• Hidden text or text only for bots. Content that users can't see is a trust problem in every search system.
• Thin pages for every variation. Fifty near-identical pages for "best X in [city]" spread your authority thin. One strong page with a clear section per market works better.
• Self-written reviews and astroturfed threads. They can get removed, and a removed thread that once cited you is a negative signal waiting to be found.

The same goes for gated content. A whitepaper behind a form can't be read, so it can't be cited. If a report is your best evidence, publish a summary page with the key findings in plain HTML and keep the full PDF behind the form if you need the leads.

In short: write for the reader who asked, and you're writing for Perplexity too.

Track a fixed set of prompts, record every cited source each day, and count two things: how often your own pages are cited, and how often cited pages mention your brand. Both move independently.

Your own citation rate tells you whether your pages are winning. The mention rate inside cited pages tells you whether your third-party work is paying off. Ossvy records both in citation tracking, and the Perplexity tracker shows your visibility there next to competitors.

Perplexity cites what it can crawl and what answers the sub-question best. Allow PerplexityBot and Perplexity-User, put direct, specific answers high on focused pages, keep time-sensitive pages current, and get accurately listed on the third-party pages it already trusts. Then measure your citations every week. For a full plan, see how to improve AI visibility in 30 days.