Perplexity SEO: how it retrieves and cites sources, and how to get cited.

1. How Perplexity retrieves and cites sources.

Perplexity is an answer engine. Its help pages describe four steps: it works out what the question means, searches the internet in real time, summarises what it found, and attaches numbered footnotes that link to the sources. Its CEO, Aravind Srinivas, has described the design rule as the model not being supposed to say anything it did not retrieve. That is the main difference from ChatGPT, which answers most prompts from memory and searches on roughly a third of them, as we cover in how to rank in ChatGPT. In Perplexity, search is the default, so nearly every answer is a chance to be cited.

Perplexity AI SEO, or SEO for Perplexity, is the work of becoming one of those numbered sources. In September 2025 Perplexity's engineers published a description of the search system behind the product. It runs in four stages.

1. It crawls and indexes the web itself. A crawler called PerplexityBot feeds an index that tracks more than 200 billion URLs. The write-up says the system processes tens of thousands of index updates every second, and schedules crawling by how important a page is and how often it is likely to change.
2. It splits pages into passages. Documents are divided into small units, and each unit is scored on its own against the question. Your page is not competing as a whole. Each passage is.
3. It retrieves and ranks. Retrieval combines keyword matching with matching by meaning, and the candidates then go through several ranking stages.
4. It writes the answer and cites. The model writes from the passages it was handed and numbers the sources it used.

Because Perplexity ranks from its own index, its results sit closer to Google's than any other assistant's do. Ahrefs ran 15,000 long-tail queries in 2025 and checked where each assistant's cited pages ranked. For Perplexity, 28.6% of cited pages were in Google's top 10 for the same query, and 16.6% were in Bing's top 10. ChatGPT, Gemini and Copilot were each under 9% for Google. Read that both ways: ordinary SEO does more for you here than in ChatGPT, and about seven in ten cited pages still come from outside Google's first page.

The audience is real. Srinivas said Perplexity handled 780 million queries in May 2025, and the engineering write-up puts the load on its search system at 200 million queries a day. The AI search engines we compared shows how it sits next to the others as a product, and LLM SEO is the wider discipline across all of them.

One caution before the advice. Nobody outside Perplexity has the ranking weights. In August 2025 a researcher, Metehan Yesilyurt, published what he said were internal ranking parameters found by inspecting the product, and by his own account Perplexity did not reply. Guides that tell you freshness is 40% of the ranking are quoting a number with no source. This guide sticks to what Perplexity has published and what independent studies have measured.

2. PerplexityBot and Perplexity-User: get access right.

Perplexity documents two agents. They do different jobs and follow different rules.

The documentation is direct about the first one: “To ensure your site appears in search results, we recommend allowing PerplexityBot in your site's robots.txt file and permitting requests from our published IP ranges listed below.” It also says PerplexityBot “is not used to crawl content for AI foundation models.” That matters for the decision. ChatGPT and Claude each run a separate training crawler you can block while staying in search. Perplexity documents no training crawler at all, so blocking PerplexityBot buys you no protection. It only takes you out of the answers. A robots.txt that keeps you eligible looks like this:

# Perplexity search index: required to be cited
User-agent: PerplexityBot
Allow: /

# Pages fetched for a person's question
User-agent: Perplexity-User
Allow: /

Sitemap: https://www.example.com/sitemap.xml

Three more checks sit outside the file.

• Your CDN and firewall. A block at the edge means the bot never reaches robots.txt. Cloudflare has blocked known AI crawlers by default on newly onboarded domains since July 2025. Perplexity's documentation publishes IP ranges for both agents and recommends matching on the user agent and the IP address together. Then check your logs: PerplexityBot requests should return 200, not 403.
• The 24-hour lag. The documentation says changes can take up to 24 hours to take effect. Do not test five minutes after you deploy.
• Text in the HTML. Keep answers, prices and specs in the server-rendered HTML. Content that only appears after JavaScript runs, or that sits in an image or a PDF, is a risk you do not need to take.

One thing robots.txt will not do is keep a page private. The documentation says Perplexity-User “generally ignores robots.txt rules” because a person asked for the page. And in August 2025 Cloudflare published a report saying Perplexity also used undeclared crawlers, which it measured at 3 to 6 million requests a day, to reach sites that had blocked the declared ones. Cloudflare removed Perplexity from its verified bot list. Perplexity called the report a publicity stunt and said the traffic belonged to a third-party browser service. You do not need to pick a side to act on it: if a page must stay out of AI answers, put it behind a login.

Claude runs a similar set of bots with its own rules, and our ClaudeBot guide has the matching robots.txt. To audit crawl access properly, work through the crawl checks in our technical SEO audit checklist, because the faults that keep Googlebot out (blocked paths, broken rendering, server errors) usually keep PerplexityBot out too.

3. Freshness: what the data says, and what it does not.

Most Perplexity guides tell you to update every page every few days. The evidence does not support that. It supports something narrower.

• The index is built to be current. Tens of thousands of index updates a second, with crawling scheduled by how often a page is likely to change. A page that changes can be re-read quickly, and a new page on an active site does not wait long.
• AI assistants do lean newer than Google. Ahrefs analyzed about 17 million citations in July 2025. Pages cited by AI assistants were 1,064 days old on average, against 1,432 days for Google's organic results. That is 25.7% fresher.
• Perplexity is not the most recency-hungry of them. In the same study the pages Perplexity cited averaged 1,166 days old, roughly three years. That was older than the pages cited by ChatGPT, Copilot and Gemini. The average time since those pages were last updated was 993 days.
• Newer sources tend to come first. Ahrefs found a weak tendency for Perplexity to list newer sources earlier in an answer.

So a three-year-old page that is still right gets cited all the time. What loses is a page whose facts have expired: last year's price, a retired plan, a statistic with no year on it. When the question is time-sensitive (a price, a “best of 2026” list, a product that changed last month), the current page wins. Four habits cover it.

• Show a real updated date. Put it where a reader can see it, and change it only when the content changes. Ahrefs warns against bumping a date on an unchanged page, and so do we.
• Date the facts that expire. Write “$449, checked October 2026”. A passage quoted alone should still say when it was true.
• Refresh on a schedule. List the pages that carry prices, versions, screenshots and statistics, and review them on a calendar.
• Publish new pages for new questions. Buyers ask about this year's models and this quarter's changes. A site that publishes regularly always has a recent page to offer.

How fast the ground is moving under all of this is in our AI SEO statistics, and the publishing rhythm that keeps a blog current is covered in the blog SEO guide.

4. Structure: write passages Perplexity can lift.

Stage two of the pipeline decides this section. Perplexity scores passages, so a page earns a citation when one passage, read alone, is the best available answer to the question. A strong page with the answer spread across six paragraphs loses to an average page with the answer in one.

• Make the heading the question. Write it the way a buyer would ask it. “Is the D2 stable at full height?” matches a real question. “Built to last” matches nothing.
• Answer in the first two sentences under it. Give the answer, then the detail. That is the core of answer engine optimization, and it is the cheapest edit on this list.
• One question per section. A section that answers three questions gives the scorer three weak passages. Split it.
• Back claims with numbers and named sources. The GEO benchmark presented at the KDD conference in 2024 tested content edits on Perplexity itself. Adding quotations improved a source's visibility in the answer by 22% on one measure, and adding statistics improved it by 37% on another. Keyword stuffing did 10% worse than leaving the page alone. The generative engine optimization guide walks through that study.
• Use tables and lists for what they suit. A comparison belongs in a table and a procedure in a numbered list. This is our reasoning, not a measured result: a table row or a list item carries its own context, which is what a passage scorer needs.

None of this is a trick. It describes a good reference page. The hard part is volume: a section like that for every question your buyers ask, with a source behind every number.

5. Entity clarity: say who and what the page is about.

An entity is a specific thing with a name: a company, a product, a person, a category. Entity clarity means a reader, or a model, can tell which one a passage is about without reading anything else. It matters more in Perplexity than on a results page because the passage is lifted out of the page that gave it context. “It costs $449” is a fact about nothing once it leaves your site.

• Name the thing in the answer. Write “The Varnley D2 standing desk costs $449”. Pronouns and phrases like “our platform” or “this model” mean nothing outside your page.
• Say what it is and who it is for. One plain sentence: the name, the category, the audience. A model that is deciding whether your product belongs in an answer about a category needs to be told the category.
• Use one name, spelled one way. Your site, your listings, your profiles and your reviews should all use the same brand and product names. Three spellings read as three things.
• Publish the official facts on your own domain. Price, specifications, limits, integrations, and what the product does not do, each on a page whose title says what it is. If Perplexity cannot find your price on your site, it will quote the one in an old review.
• Disambiguate a shared name. If your brand shares its name with something else, add the qualifier every time it could be misread.

This is the point where the work stops looking like classic keyword SEO. Our piece on how GEO differs from SEO draws that line, and the AI search optimization page covers the same levers across Perplexity, ChatGPT, Gemini and Google's AI features.

6. Where Perplexity citations go: Reddit, reviews and you.

Everything so far is about your own pages. A large share of Perplexity's citations do not go to anyone's own pages.

• Across all topics. Profound analyzed 680 million citations between August 2024 and June 2025. Reddit was Perplexity's most cited source, at 6.6% of all its citations. YouTube followed at 2.0%. The rest of the top ten were Gartner, Yelp, LinkedIn, Forbes, NerdWallet, TripAdvisor, G2 and PCMag, each at 1% or less.
• For buying questions. Tinuiti's first-quarter 2026 report tracked commercial prompts in nine product categories across seven AI platforms. On Perplexity in January 2026, 31% of citations went to social media, and Reddit alone was 24%. YouTube was 3%. The report says Perplexity stood out from every other platform it studied on this measure.

So when someone asks Perplexity for the best product in your category, or whether yours is worth it, the sources are often a Reddit thread and a review site. Your product page may never be opened. Two jobs follow.

• Find the pages first. Ask Perplexity the ten questions a buyer in your category would ask. Open the sources under each answer and write down the domains and the exact threads. That list is your off-site plan.
• Earn a real place on them. Answer in the threads as yourself, say who you work for, and bring facts a reader can check. Get the product in front of the review sites and roundups that show up. Planted praise gets removed, and it is the kind of thing a thread remembers.

These shares move, so recheck them. If you want software to do the watching, the GEO tools we tested cover the whole market, and the AI visibility tools roundup narrows it to the ones that track citations in assistants such as Perplexity.

7. How to rank on Perplexity: the checklist, and how to measure it.

Fourteen checks, in the order to do them. Access comes first because nothing else counts until the crawler can read you. For the classic search basics underneath, use our SEO checklist.

Measuring it takes three signals, and you already have two of them.

• Referral traffic. Filter your analytics for sessions whose source is perplexity.ai. Most tools file it under referrals, so it is easy to miss until you look for it.
• Server logs. Count requests from PerplexityBot, which tell you the index is reading you, and from Perplexity-User, which tell you a person's question led Perplexity to open your page. Perplexity-User hits per URL are the closest thing you have to an impression count.
• A fixed question set. Write 20 questions your buyers would really ask. Run them every month and record two things: whether your page is in the numbered sources, and whether your brand is named in the answer. Answers vary from run to run, so follow the trend. The AEO tools roundup lists software that runs a set like this for you.

Google's own AI answers work from a different pipeline and need their own checks, which we cover in the Google AI Mode guide.

A worked example: 30 days from absent to cited.

Take a fictional company, Varnley, that sells standing desks online. The company, its site (on the reserved .example domain) and every number below are invented to show the method. The mechanics match the documentation and studies above.

Varnley has two problems. Ask Perplexity for the best standing desk under $500 and the answer cites two Reddit threads, a review site and two competitors. Ask whether the Varnley D2 is stable at full height and it quotes a year-old comment about the previous model.

The fixes ran in the same order as the checklist.

1. Day 1, access. Allowed PerplexityBot in robots.txt, exempted its published IP ranges from the firewall rule, and confirmed 200 responses in the logs two days later.
2. Days 2 to 10, structure and names. Rebuilt the buying guide as nine question sections, each opening with a two-sentence answer that names the product in full. One spelling, Varnley D2, everywhere.
3. Days 10 to 20, freshness. Dated every price and specification, added a visible updated date, and published four new pages for questions buyers ask: stability at full height, weight capacity, assembly time, and D2 against D1.
4. All month, off-site. A Varnley engineer replied in both threads, said who they were, and posted the wobble-test figures for the new frame. Varnley also sent a D2 to the review site Perplexity cited for the category.

Here is one section of the buying guide before and after.

The second version answers the heading in one word, names the product, gives numbers with dates, and says what the desk does not include. Every sentence still works if it is quoted alone.

The questions about Varnley itself moved first, because Varnley's own page became the best source on them within days of the rewrite. The category question moved slowest. Being named among the best standing desks depends on what the threads and the review site say, and by day 30 the review had not been published. The rewrites and the new pages took most of the month, and Varnley's list of buyer questions has fifty more on it.

Common mistakes in Perplexity SEO.

1. Blocking PerplexityBot to stop AI training. It feels like the cautious choice. But Perplexity documents PerplexityBot as a search crawler that is not used for training, so the block removes you from answers and protects nothing. Check the CDN as well as the file.
2. Treating robots.txt as a privacy wall. People assume a disallow rule keeps a page out of AI answers. Perplexity-User generally ignores it when a person asks for the page. Private pages belong behind a login.
3. Faking freshness. Changing the date on an unchanged page is quick, which is why it gets recommended. The pages Perplexity cites average about three years old, so the date alone is not what gets a page picked. Update the facts, then the date.
4. Writing in pronouns. “Our platform” and “it” read fine on your site, where the logo is in the corner. A lifted passage has no logo. Name the product in the sentence that carries the fact.
5. Working only on your own site. It is the part you control, so it gets all the effort. For buying questions, about a quarter of Perplexity's citations went to Reddit in the Tinuiti data. Read the threads and the review pages it cites, then earn your place on them.
6. Shipping an llms.txt and calling it done. It is a one-hour task that looks like progress. Perplexity has not said its crawlers read the file, as our llms.txt guide lays out. Access and passages are what move citations.