How Perplexity picks its sources: a beginner's guide

Perplexity SEO is the work of making your web pages easy for Perplexity to find, read and cite when it answers a question. The result you are after is a numbered citation inside the answer, and ideally your business named in the answer text, rather than a position in a list of links. Perplexity describes itself as searching the internet in real time, summarising what it finds into a conversational answer and attaching numbered citations to the original sources, so the whole job comes down to being one of those sources.

The facts below come from Perplexity's own help centre, crawler documentation and research articles, the robots.txt standard, and Google's documentation on its AI features, so you can tell published fact from informed practice.

What Perplexity SEO means, in plain words

Perplexity is an answer engine. You type a question, it searches the web, and it writes a short answer with numbered links to the pages it used. Unlike a classic results list, what the reader sees is a paragraph or two of prose with small numbered markers; a page either contributes and earns a marker, or it is invisible, however well it ranks elsewhere.

So Perplexity SEO has two outcomes worth aiming for: a citation, where your page is one of the numbered sources attached to the answer, and a mention, where your business is named in the answer text itself. The rest of this guide is about how Perplexity decides which pages get either.

How Perplexity turns a question into a cited answer

Perplexity's help centre, in pages last updated in August and September 2026, describes four stages:

1. Understanding the question. A language model works out what you are asking, including the context of earlier questions in the same session.
2. Searching the web. Perplexity says it gathers information from sources such as articles, websites and journals; in Pro Search it runs multiple searches and, depending on the focus the user picks, draws from articles, academic papers, forums, videos and more.
3. Summarising. The model compiles the most relevant material into one answer.
4. Citing. Each answer carries numbered citations linking to the original sources so the reader can check them.

The language model writes the answer, but Perplexity's own search system decides what the model gets to read, and that is the part that decides whether your page is in the running. As of September 2026, Pro users can choose between models including Perplexity's own Sonar, OpenAI's GPT-5, Claude Sonnet 4.6 and Gemini 3.1 Pro.

How deep the search goes depends on the mode. Perplexity describes standard search as fast, surface-level answers from one or two sources. Pro Search runs multiple searches and compiles insights from dozens of sources, with focus options such as Web, Academic, Finance or your own files. Research, which Perplexity also calls Deep Research, runs dozens of searches automatically, reads hundreds of sources and delivers a report in two to four minutes.

Perplexity also works in sessions, which keep the initial question, every follow-up, every response and all the sources used, with the model remembering the context as the conversation goes on. A page not cited on the first question can be cited on the third, once the question is specific enough to need it.

Where Perplexity's sources come from: its own index and live visits

Perplexity runs its own search index. Its research team said in September 2025 that the index tracked more than 200 billion unique URLs and that its production search infrastructure handled 200 million queries a day; those are Perplexity's own figures. Perplexity says the first version of its answer engine used some of the third-party search APIs available in late 2022, and that it built its own because those were costly, too stale or too slow, and, most importantly, not designed to return the passage-level results an AI model needs.

Two different visitors can put a page in front of Perplexity:

• PerplexityBot is the search crawler. Perplexity says it exists to surface and link websites in Perplexity's search results and is not used to crawl content for AI foundation models.
• Perplexity-User acts on a user's request. Perplexity says that when someone asks a question, it may visit a web page to help answer accurately and link to that page, and that this agent neither crawls the web nor gathers content for training.

The index is not refreshed evenly. Perplexity says a machine-learning model predicts whether a URL needs indexing and when, based on its importance and likely update frequency, and that it prioritises documents from authoritative domains and under-covered topics for its fastest storage.

If PerplexityBot cannot reach your page, it is unlikely to be in the pool Perplexity picks from.

How Perplexity picks which pages to cite

Perplexity has not published a list of ranking factors, so any list of factors is informed practice rather than documentation. What it has published, in a September 2025 research article accompanying its Search API, is the shape of its retrieval and ranking pipeline.

1. Retrieval by words and by meaning. Perplexity queries its index lexically (the words in the question) and semantically (what the question means), then merges both into one candidate set, prioritising breadth over precision at this stage.
2. Prefilters. Basic heuristics remove content that is clearly non-responsive or stale.
3. Ranking in stages. Fast scorers based on words and embeddings narrow the candidates, then heavier cross-encoder reranker models make the final cut.
4. Passages as well as pages. Perplexity retrieves and scores results at both the document and the sub-document level, splitting pages into self-contained spans that are ranked individually, so the model is fed the most relevant passages rather than the whole page.

Your page is judged as a whole and section by section, so a single section of it can compete against single sections of every other page in the candidate set. Perplexity's reason is that model performance degrades as the context fills with irrelevant material, so it hands the model the most atomic units it can.

How the page is read matters too. Perplexity's content-understanding module parses each site with rules it adjusts over time, aiming for completeness and for quality, which it defines as capturing information in a way that preserves the original structure and layout.

The ranking models also learn from the millions of user requests Perplexity serves each hour, combined with AI-driven evaluations and human feedback. Nobody outside Perplexity can see those signals, so the advice later in this guide is informed practice built on the published mechanics rather than a checklist of factors.

How Perplexity SEO differs from Google SEO and AI Overviews

Google. Classic results are a ranked list. For AI Overviews and AI Mode, Google says there are no additional technical requirements beyond being indexed and eligible to show with a snippet, and the usual SEO fundamentals apply. Google also describes a "query fan-out" technique for those features, issuing multiple related searches across subtopics to build one response.

Perplexity. It runs a separate index and a separate crawler: PerplexityBot is its own user-agent with its own robots.txt group and published IP ranges. Perplexity's documentation describes no reporting tool for ordinary site owners (its partner publishers' program aside), so for most businesses measurement is manual.

The overlap is in the fundamentals Google lists for its AI features, because both systems must crawl and parse a page before they can use it: crawling allowed in robots.txt and by your CDN or host, important content available as text, internal links, and structured data that matches the visible text if you use it. The difference to remember is that allowing or blocking Googlebot says nothing about PerplexityBot, so a catch-all disallow, or a CDN setting that blocks AI search crawlers as well as training crawlers, keeps PerplexityBot out too unless you say otherwise.

Perplexity is one of several engines that work this way. The umbrella term for the work across Perplexity, ChatGPT, Gemini, Claude and Google's AI features is GEO, which our guide to generative engine optimisation explains, and our beginner's guide to AI SEO covers how each of the major tools chooses its sources.

Letting Perplexity in: robots.txt, firewalls and the crawler debate

What Perplexity asks for

Perplexity's crawler documentation recommends allowing PerplexityBot in robots.txt and permitting requests from its published IP ranges, which it lists in JSON files it says should be treated as the source of truth. If you use a web application firewall such as Cloudflare's or AWS WAF, Perplexity says you may need to explicitly allow its bots, with a rule that matches both the user-agent and the IP range and sits above your blocking rules.

How robots.txt actually works

The format is defined by an internet standard, RFC 9309. Rules are grouped by user-agent, and a crawler follows the group that names it; if there is none, it must follow the "*" group if one exists. The standard also says plainly that its rules are "not a form of access authorization": they are a request to well-behaved crawlers rather than a lock.

Perplexity-User and the crawler debate

Perplexity-User is different. Because a user requested the fetch, Perplexity says this fetcher generally ignores robots.txt rules. Its position, in its August 2025 response to Cloudflare, is that a fetch made on a real person's behalf is not crawling: the content is used immediately to answer that person's question and, it says, is neither stored nor used for training.

That position is contested. In August 2025 Cloudflare published tests in which it said Perplexity obtained content from new domains that blocked both of Perplexity's declared crawlers, attributed three to six million daily requests from an undeclared, browser-like user agent on IP addresses outside Perplexity's published ranges to Perplexity, and de-listed Perplexity as a verified bot. Perplexity disputed the findings, saying Cloudflare had confused user-driven requests with scraping and had misattributed traffic from BrowserBase, a third-party browser service it said it uses for fewer than 45,000 requests a day.

This guide does not take a side. The practical point is narrower: whether to let AI tools in is a business decision, the controls are imperfect, and if you want citations, check that you have not blocked PerplexityBot by mistake.

How to make your pages easy for Perplexity to cite

Each point below follows from how Perplexity says its pipeline works.

1. Answer first, in every section. Perplexity scores passages on their own, so the opening sentences under each heading should answer that heading's question completely, in plain words, with detail and caveats after. Give each distinct question its own heading.
2. Use structure the parser can keep. Perplexity says its parser aims to preserve the original structure and layout and that well-structured lists and tables may suit more formulaic extraction, so short paragraphs, numbered steps and simple tables beat long runs of prose.
3. Keep key facts in HTML text. Prices, specifications, opening hours and service areas belong in the page text rather than only in an image, a PDF or a script-drawn widget. Google's advice for its AI features makes the same point about important content being available in textual form.
4. Keep time-sensitive pages current. Perplexity filters clearly stale content out before ranking, times re-indexing around a site's usual update cadence, and calls information staleness one of the biggest failure modes for AI agents. Pages with dates, prices or availability need a review schedule.
5. Back claims with specifics. Every citation is traceable, so a page that says what you do, where, for whom and at what price gives a model something to use; "the leading provider" does not.
6. Be present where the research happens. Pro Search draws from articles, academic papers, forums and videos depending on the focus a user picks, so reviews, industry roundups, community discussions and video about your business count alongside your own site.

Some businesses do this in-house. Others hand the crawler checks, page restructuring and ongoing prompt testing to a specialist, which is the work our SEO for Perplexity service covers.

Common mistakes that keep a site out of Perplexity's answers

• Blocking PerplexityBot by accident. A blanket disallow for all user-agents, a CDN AI-bot setting that blocks AI search crawlers and not just training crawlers, or a firewall rule that challenges unknown bots will all do it. Check which categories your CDN's AI-bot setting covers.
• Assuming Google rankings carry over. Perplexity uses its own index and crawler, so a page can rank well in Google and be absent from Perplexity because PerplexityBot never reached it.
• Locking facts in images, PDFs and scripts. If the parser cannot read it as text, it cannot cite it. Google has explained how Search indexes and ranks PDF files and why most business documents are better off as web pages.
• Letting time-sensitive pages go stale. Perplexity's prefilters remove clearly stale content before ranking, and an out-of-date price list is worse than none.
• Chasing a secret ranking-factor list. Perplexity has not published one, so treat specific word counts, keyword densities, schema types or backlink targets as hypotheses to test, not rules.

How to tell if Perplexity is citing you

With no reporting tool for site owners in Perplexity's documentation, measurement is a mix of manual checks, server logs and analytics.

1. Run a fixed prompt list. Write down the questions a customer would ask before choosing a business like yours, ask them in Perplexity on a fixed schedule in the same mode each time, and record whether you are cited, whether you are named, and who is cited instead. Ask a follow-up or two as well, because sessions carry context.
2. Check the answer's Source details. On any answer, the Source button shows which model, search mode and source mode were applied, and the three-dot menu lists the sources used. Record the mode with each check so you compare like with like.
3. Check your server logs. Look for the PerplexityBot and Perplexity-User user-agent strings and verify them against Perplexity's published IP lists, since user-agent strings can be faked. Regular PerplexityBot visits confirm you are reachable.
4. Watch referral traffic. Visits from perplexity.ai show up as referrals in your analytics when the referrer is passed through. Tracking gaps happen: a ChatGPT referral drop in GA4 was traced to a missing tracking tag, so check the plumbing before reading anything into a dip.

Frequently asked questions

Can I pay Perplexity to be cited?

Nothing in Perplexity's own help documentation describes a way to buy a citation in an answer; citations come out of the retrieval and ranking pipeline described above.

Does Perplexity use my content to train AI models?

Perplexity says PerplexityBot is not used to crawl content for AI foundation models and that Perplexity-User is not used to collect content for training. Those are Perplexity's statements about its two declared agents.

Will blocking Perplexity hurt my Google rankings?

Not through robots.txt. Google says the rules for Googlebot are the control for how your site is crawled for Google Search, and PerplexityBot has its own group of rules, so what you do with one does not change the other.

Can a small business be cited, or only big sites?

Perplexity scores passages for relevance to the specific question and says it prioritises documents on under-covered topics, as well as from authoritative domains, for its fastest index storage. That is reasoning from its published design rather than a promise, but a focused page that answers a niche question well has a real chance.

Where to start this week

1. Open your robots.txt. Confirm nothing blocks PerplexityBot, by name or through a blanket disallow.
2. Check the CDN and firewall. Look for managed AI-bot blocking or bot challenges, check which AI-bot categories (search, agent, training) your CDN setting blocks, and if you want citations, add an allow rule for PerplexityBot that matches both the user-agent and Perplexity's published IP list.
3. Read the logs. Search the last month for PerplexityBot; if it has never visited, something is still in the way.
4. Write your prompt list. Ten to twenty customer questions, asked in Perplexity, with the results recorded as your baseline.
5. Rewrite your three most important pages answer-first. One question per section, the answer in the first sentences, facts in text, a review date on the page. Re-run the prompt list in a month and compare.

When you want the fuller method, the GEO Playbook covers how a generative engine picks its sources, the crawler access decision and how to measure a channel with no rankings, and includes a prompt-tracker worksheet you can use for step four.