Free Growth Assessment
SEO & AEO

Perplexity SEO: How Perplexity Picks the Sources It Cites

Perplexity documents its crawlers, its labels and its search stages. It doesn’t publish ranking weights. Here’s which parts of Perplexity SEO are documented and which are guesses.

By Holly Mack Last updated: November 22, 2026 11 min read

Perplexity SEO begins with access. PerplexityBot has to reach your pages first. After that, Perplexity searches, ranks pages and passages, and cites sources, but it describes those stages without publishing the weights behind them.

A hand holding a blank paper badge up to a brass card reader beside a glass door
The first question is whether the door opens.

Perplexity numbers its sources. Every answer arrives with small numbered links, and the only question an owner really cares about is whether one of them is theirs.

So you go looking for the steps. Fair enough. Page after page promises them. Publish here, restructure there, win the first hour after you post.

Some of that is documented. Some of it is a guess repeated by enough vendors that it starts to sound like a rule, and plenty of advice with very little source behind it is how a guess becomes a checklist. Perplexity itself documents only a slice: who gets in, how an answer gets built, and what its labels mean.

That makes this an access problem and an evidence problem before it’s a content problem. The content work that follows is what our generative engine optimization services cover.

Key Takeaways

The Short Version

Perplexity documents two crawlers, PerplexityBot for search and Perplexity-User for fetches a person asked for, and the second one generally ignores robots.txt. It describes a retrieval and ranking pipeline that scores passages as well as pages, labels some whole domains as Government, Academic, or Trusted, and publishes no ranking weights. The citation studies are all from companies with something to sell. The steps built for MSPs are in how MSPs get cited by Perplexity.

How Perplexity Picks Sources: What Its Documents Show

A Perplexity source is a page the answer links to with a numbered citation. Perplexity’s Help Center says it interprets the question, searches the internet in real time, gathers material from sources such as articles, websites, and journals, and compiles an answer with numbered citations.

That’s the whole public description of the answer process. Short.

The detail sits in a research article on its Search API, published September 25, 2025, and it describes the machinery in real depth, from how URLs get indexed to how rerankers sort the final candidates, and still stops short of the weights.

What Perplexity Documents, and What It Doesn’t

TopicWhat Perplexity saysWhere
The answer processIt interprets the question, searches the internet in real time, gathers material from sources such as articles, websites, and journals, and attaches numbered citations.Help Center
The indexIt tracks over 200 billion unique URLs. It can’t refresh every one often, so a model predicts which URLs need indexing.Search API research article
Retrieval and rankingLexical and semantic retrieval are merged, then several ranking stages end in cross-encoder rerankers. Results are scored at the document and the sub-document level.Same article
DatesEach result carries a publication date and, when known, a last-updated date. The snippet is pulled for the query, so one page can return different text for different queries.How Search Works
Not in any page I readRanking weights, a score for source quality, a rule for how a page earns a citation.Checked November 22, 2026

Read the scope of that research article carefully. It’s narrower than it looks. It describes the Search API, which returns ranked pages and snippets and writes no answer. Perplexity says its ranking stack is built together with its products, so I treat the article as a good picture of the machinery and not as the answer app’s rulebook.

A list of ranking factors has to come from somewhere other than Perplexity. Ask where.

Perplexity SEO Starts With Access: PerplexityBot and Perplexity-User

Perplexity names two agents. They do different jobs and follow different rules, and it’s easy to talk about them as one “Perplexity crawler.”

The Two Agents Perplexity Documents

AgentIts job, in Perplexity’s wordsrobots.txtTraining use
PerplexityBotDesigned to surface and link websites in search results on Perplexity.Perplexity recommends allowing it in robots.txt and permitting its published IP ranges.Not used to crawl content for AI foundation models.
Perplexity-UserSupports user actions. When someone asks a question, it may visit a page and link it in the answer.Because a user requested the fetch, it generally ignores robots.txt rules.Not used for web crawling or to collect content for training AI foundation models.

Source: Perplexity’s crawler documentation. Each bot has its own published IP list.

Two details do the work. Perplexity says each setting works independently and a change can take up to 24 hours to show. And for PerplexityBot it asks you to allow both the robots.txt line and its published IP ranges, since a firewall rule can stop a request that robots.txt would have let through.

If your office ran like this, you’d repaint the lobby before checking whether the front door opens. Content nobody can fetch doesn’t compound. It just sits there. Locked door. Nice lobby.

The same page includes firewall guidance for Cloudflare and AWS. It also carries a warning that Cloudflare’s Bot Fight Mode can’t be skipped by a custom WAF rule, so if your site runs that setting, a rule you wrote to let Perplexity in may not do what you think.

A Dispute You Can Test in Your Own Logs

On August 4, 2025, Cloudflare published a post saying it saw Perplexity use an undeclared, browser-style user agent and other IP addresses after its declared crawler had been blocked on its test domains. Perplexity answered the same day. It said Cloudflare had misattributed traffic from a third-party browser service, and that agents acting for a person differ from crawlers.

Cloudflare sells bot management. Perplexity runs the agents at issue. Neither side is neutral. Each has a stake in how the argument ends, and I can’t settle it from here.

What you can do is read your own records. Perplexity’s page says neither agent is used to collect content for training foundation models, and I can’t verify that from the outside either. My working rule is to check the door before you touch a page: pull recent server or firewall logs, search for both agent names, and compare the source IPs with Perplexity’s published lists. If neither agent shows up, ask whoever runs your firewall what it does with them.

What Perplexity Labels, and What It Doesn’t

A hand pressing a brass stamp onto one of a row of blank white cards on a slate desk
A label rates the whole site, not the page.

Some citations carry a small shield. That domain has been rated. Perplexity’s source review process gave it a Government, Academic, or Trusted label.

The rating covers the whole website, not the page. The questions are objective ones, such as whether the site corrects its mistakes, says who wrote each piece, and keeps news apart from advertising and opinion.

Perplexity says partnerships, payments, and other business arrangements don’t affect a label. It also says the majority of domains have no label, and that no label isn’t a negative judgment. It means the domain hasn’t been rated yet, or there’s no clear rating for it.

What the page doesn’t say is whether a label changes what gets cited. I found nothing that says it does, so I wouldn’t build a plan around getting one. Perplexity’s source labels page has the details. The three questions are still a free audit of any MSP site. Do you post corrections? Does every article name its author? Is advice kept apart from the sales pitch?

Then there’s the Publishers’ Program. Perplexity announced it on July 30, 2024 for media organizations and online creators, with TIME, Der Spiegel, Fortune, Entrepreneur, The Texas Tribune, and WordPress.com as its first partners. Partners were promised a share of advertising revenue when their content is referenced, plus free API access. The announcement is more than two years old and describes no route in for a company blog.

What the Citation Data Says About Perplexity SEO

Two studies give Perplexity numbers. Both come from companies with a product to sell, and neither is built for an IT buyer. Keep that in mind.

What Each Study Measured

Study, dates and platformsWhat it found about PerplexityLimits
Semrush, published November 10, 2025. Weekly snapshots, July 14 to October 12, 2025. 230K prompts and over 100M citations across ChatGPT search, Google AI Mode, and Perplexity combined. Country not stated.Top sources were Reddit, LinkedIn, NIH, Microsoft, and Google. Wikipedia held near 0.8% of responses. Reddit slipped slightly in September.The vendor sells AI visibility tools. All topics. No definition of a citation.
MarTech, published February 27, 2026. Run period not stated. 1,000+ tool-written prompts, 29 technology brands, in English across ChatGPT, Perplexity, Grok, and Google AI Mode or Gemini, blended.Perplexity isn’t broken out. Brands with a top-10 cited domain averaged 25% visibility, against 7.6% for the rest. More than twice as many owned sites were cited as earned media sites.The author runs a technology PR firm. MarTech is owned by Semrush.

Two honest readings. In Semrush’s all-topics sample, the top five on Perplexity include LinkedIn, plus NIH and Microsoft, which say more about the mix of prompts than about Perplexity. And the technology sample blends four engines, so you can’t pull Perplexity out of it.

Neither study tells you which sources Perplexity reaches for when an IT buyer asks about managed services, because neither one was built around that buyer, that question, or that engine alone. Only asking can. For how the same kind of numbers swing on another assistant, see what the ChatGPT citation studies show.

What Perplexity SEO Looks Like for an MSP or IT Firm

Here’s what I’d do with the documented parts. These are my working rules, and each one rests on a fact above, not on a ranking factor.

  • Let PerplexityBot in, in robots.txt and at the firewall. Check the logs a week later.
  • Make each section stand on its own. Perplexity’s research article says its Search API scores passages as well as pages, so a paragraph that only makes sense after the three before it is harder to lift. That’s my inference, because Perplexity hasn’t said how passages get picked.
  • Date the page, and change the date only when the page changes. The API returns a publication date and a last-updated date when it knows them.
  • Be where the studies keep finding Perplexity. LinkedIn sits in Semrush’s top five for Perplexity, and in the technology sample more than twice as many owned sites were cited as earned media sites. Your own pages, then, and your own LinkedIn presence.
  • Ask Perplexity your buyers’ questions, write down who gets the numbered links, and do it again next quarter.

The steps built for MSPs are in how MSPs get cited by Perplexity. For what each assistant documents, there’s our LLM optimization guide, and for LinkedIn’s part, how LinkedIn posts end up in Perplexity answers. If you’d like a quick first look, try the free AI visibility self-check.

How to Read the Next Claim About Ranking on Perplexity

Another guide will land next week. Run it through six questions before you repeat anything from it.

  • Does it point to a Perplexity page? “Perplexity says” with a link is a document. “Perplexity prefers” without one is a guess.
  • Where does the percentage come from? Several guides repeat citation-rate figures with no source I could find.
  • Who sells what? Tools, services, firewalls. It doesn’t void the claim. It tells you whose advice sits at the bottom.
  • Which Perplexity? The app, Pro Search, Deep Research, and the Search API are different things, and the research article above is about the last one.
  • When was it written, and has Perplexity published anything newer? The research article above is dated September 2025.
  • Can you test it on your own site in an afternoon? A robots.txt line, yes. A ranking factor, no.

My working rule, and not a published standard, is to trust a claim more when it cites a Perplexity page, a date, and a sample, and less when it hands you a factor with none of them. Documented beats confident.

What to Do This Week

Three small jobs. Nothing fancy. Read your robots.txt and your firewall rules for the two agent names. Ask Perplexity ten of your buyers’ real questions and note who gets the numbered links. Then hold off on rewriting pages around a ranking factor nobody has published, because you’d be spending a week on a guess when the afternoon version of the work costs nothing.

That’s an afternoon, not a project. Cheap. If the door turns out to be locked, you’ve found the quickest fix in the whole plan.

The better question isn’t how to rank on Perplexity. It’s whether Perplexity can reach you, and whether a buyer’s question has a passage of yours that answers it. If you’d like a second set of eyes on that, book a growth strategy session.

What Owners Ask About Perplexity SEO

Is Perplexity just Google underneath?

Perplexity describes its own index and its own crawler, and the pages I read don’t say whether it also draws on another engine’s results. Its research article says the index tracks over 200 billion unique URLs.

Treat flat answers as guesses.

If I block PerplexityBot, do I drop out of Perplexity?

Blocking PerplexityBot works against the reason it exists. Perplexity says the bot is designed to surface and link sites in its search results, and it recommends allowing it if you want to appear.

Perplexity-User is a separate agent that fetches pages people ask about and generally ignores robots.txt. In Cloudflare’s August 2025 test, answers got less specific once its block held. One test, from a company with a stake in the outcome.

Does a Trusted label make a site more likely to be cited?

Nothing I read says it does. Perplexity’s Help Center describes labels as a way for readers to see what kind of source they’re looking at, and it says a label isn’t an endorsement of what a site publishes.

It also says business arrangements don’t affect a label, so you can’t buy one. No shortcut.

Can my company join the Publishers’ Program?

The announcement doesn’t describe a way in for a company blog. Perplexity introduced the program on July 30, 2024 for media organizations and online creators, and its six launch partners were publishers or a publishing platform.

If that has changed since, Perplexity’s own site is where it would say so.

Does Perplexity train on my pages?

Perplexity says no, for both of its named agents. Its crawler page says neither is used to crawl content for AI foundation models.

I can’t verify that from outside, and Cloudflare disputes how Perplexity behaves when it’s blocked. If it matters to you, set the robots.txt rules, watch your logs, and decide on what you see. Trust, then verify.

How long before a robots.txt change reaches Perplexity?

Up to 24 hours, according to Perplexity’s crawler page.

Each setting works independently, so change the line for the agent you mean. Then check your logs the next day.

Perplexity SEO AI Citations GEO SEO & AEO
Share
READY TO ACT?

Find out which sources Perplexity cites for your category

Talk to us about checking that Perplexity can reach your site and testing the questions your buyers actually ask, so you know who gets the numbered links.

Book a Growth Strategy Call