Perplexity SEO: the technical guide to Perplexity citations

How Perplexity crawls, indexes and ranks#

Perplexity is unusually open about its search stack. Its architecture write-up, published with the Search API in September 2025, explains that it started on third-party search APIs and left them over cost, staleness, latency and “document-level granularity.” The replacement — built on Vespa — “tracks over 200 billion unique URLs” and runs “tens of thousands of indexing operations per second.” Every stage below is from that document.

1. 1

   A model decides when to crawl each URL

   An ML model predicts whether a URL needs indexing and when, calibrated to “the importance and likely update frequency of the specific URL.” Documents from “authoritative domains” and “undercovered topics” are kept hot.

   Your lever: Earn links from authoritative sites, update pages substantively, and cover the gaps others don’t.

2. 2

   Pages are parsed into self-contained spans

   A content-understanding module splits each document into spans that are “individually retrieved and ranked at query time.” List- and table-heavy sites get “more formulaic parsing.”

   Your lever: Write sections that make sense on their own, and put specs and comparisons in real HTML tables.

3. 3

   Keyword and semantic retrieval run together

   Lexical and embedding retrieval run in parallel and merge into one candidate set, so both exact terms and meaning count.

   Your lever: Use the words buyers use — product names, units, category terms — alongside natural explanation.

4. 4

   Stale and non-responsive content is filtered out

   Prefilters remove “clearly non-responsive or stale content” before scoring. The index stores publish and last-updated dates per page — the Sonar API even filters on them.

   Your lever: Show a real published and updated date, and keep them honest.

5. 5

   Cross-encoders rerank passages

   Fast lexical and embedding scorers narrow the set; cross-encoder rerankers make the final cut, scoring at document and sub-document level. Rankers keep training on signals from “millions of user requests… each hour.”

   Your lever: Answer the question in the first sentence of the section that covers it.

6. 6

   The answer cites its best passages

   A standard answer cites 4–8 sources, shown as numbered cards above the text. When the index isn’t enough, Perplexity-User fetches pages live. Research mode “performs dozens of searches, reads hundreds of sources.”

PerplexityBot, Perplexity-User and robots.txt#

Perplexity documents two bots, with IP ranges published for each. Changes to robots.txt “may take up to 24 hours” to apply.

# Eligible for Perplexity search and live fetches
User-agent: PerplexityBot
Allow: /
Disallow: /account/
Disallow: /cart/
 
User-agent: Perplexity-User
Allow: /
 
Sitemap: https://example.com/sitemap.xml

It’s worth knowing the history. In August 2025 Cloudflare reported that when its declared bots were blocked, Perplexity fetched pages with an undeclared Chrome user agent from IPs outside its published ranges, and removed it from its verified-bot list. Perplexity responded that user-requested fetches aren’t crawling and that Cloudflare had misattributed traffic from a third-party browser service. Practically: if you want Perplexity’s traffic, don’t rely on Cloudflare’s default AI-bot rules to get the policy right — set it explicitly.

Technical requirements#

Perplexity has no webmaster console, doesn’t take URL submissions and isn’t an IndexNow participant. Everything rests on its crawler finding, fetching and parsing your pages cleanly — which puts the weight on rendering, speed and structure.

Server-rendered HTMLRequired

PerplexityBot didn’t render JavaScript in Vercel’s study, and Glenn Gabe found Perplexity “failed at finding the content for every url I tested” on client-rendered pages.

Bots allowed at the WAFRequired

Allow both published IP ranges through bot-fight modes, AI-bot managed rules and rate limits. Refresh the IP lists automatically.

Fast, stable responsesHelps

Perplexity-User fetches in real time. Timeouts, 403s or 429s mean the answer is built from someone else’s page.

Accurate datesHelps

A visible published/updated date plus datePublished and dateModified. The index stores dates and prefilters stale content — never fake a refresh.

HTML tables & listsHelps

Perplexity parses list- and table-heavy content “formulaically” — structure is extracted, not guessed.

Canonicals on syndicated copiesHelps

Columbia’s Tow Center caught Perplexity citing republished copies instead of originals. Require partners to canonicalize back to you.

SitemapsUnconfirmed

Not documented. Keep one with honest lastmod values, referenced from robots.txt, for general discovery.

HTML versions of PDFsUnconfirmed

Perplexity cites PDFs, but how it handles open-web PDFs isn’t documented. Publish key reports as HTML too, and never as scans.

IndexNowNo effect

Perplexity isn’t on IndexNow’s list of participating engines.

llms.txtNo effect

Perplexity publishes its own for developers, but has never said PerplexityBot reads other sites’ files or uses them to rank.

curl -s -A "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; PerplexityBot/1.0; +https://perplexity.ai/perplexitybot)" \
  https://yoursite.com/pricing | grep -c "Plans start at"
 
# 0 = client-rendered text, or a WAF challenge

What Perplexity cites#

Of all the assistants, Perplexity’s citations overlap most with classic search — so strong SEO carries over here better than anywhere. On top of that, its passage-level design rewards sections that stand alone, and its freshness filtering rewards pages that stay current.

• Passage quality

  Official

  Spans are retrieved and reranked individually. A section that answers in its first sentence — with names, numbers and units — competes on its own.

• Freshness

  Official

  Stale content is prefiltered, recrawls follow update frequency, and citations are ordered newest-first with an average age well below Google’s results.

• Authoritative domains

  Official

  Documents from authoritative domains get crawl and storage priority — they’re kept hot in the index.

• Classic rankings

  Observed

  28.6% of Perplexity’s citations rank in Google’s top 10, the highest of any assistant. BrightEdge’s older data put domain overlap near 60%.

• Community and video

  Observed

  Reddit is 46.7% of citations among Perplexity’s top-10 domains — but 6.6% of all citations. YouTube follows. Useful as a second path, not a strategy.

• The "59 ranking factors" leak

  Our read

  Parameter names found by inspecting Perplexity’s web app — rerankers, time decay, curated domains — are unconfirmed and several look like Discover feed settings. Don’t build tactics on them.

Comet Plus, publishers and ads#

Perplexity is the only major answer engine that pays publishers per use. Comet Plus costs $5 a month on its own and is included in Pro and Max; Perplexity says it distributes the revenue “minus a small portion for compute,” reported as 80% to publishers from an initial $42.5M pool. It pays on three kinds of traffic: human visits, search citations and agent actions. Partners include Condé Nast, Fortune, Le Monde, the LA Times and The Washington Post; publishers can apply via publishers@perplexity.ai.

Ads are gone. Sponsored follow-up questions ran from late 2024, new advertisers stopped around October 2025, and in February 2026 executives said Perplexity isn’t pursuing ads. There is no paid placement in answers.

Tracking Perplexity traffic#

Clicks arrive with a perplexity.ai referrer and no UTM parameters. GA4’s new AI Assistant channel doesn’t name Perplexity — Google’s documentation lists ChatGPT, Gemini, DeepSeek, Copilot and Grok — and practitioners report its traffic still lands in Referral. Give it a custom channel, placed above Referral:

# Session source — Perplexity
(^|\.)perplexity\.ai$
 
# All major AI assistants
perplexity\.ai|chatgpt\.com|openai\.com|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai

In your logs, Perplexity-User hits on a URL are the best evidence that the page is being pulled into live answers. Verify the source IP against both JSON files — the user agents are trivial to spoof:

grep -oE "PerplexityBot|Perplexity-User" access.log | sort | uniq -c
 
# Pages Perplexity fetched live for users
grep "Perplexity-User" access.log | awk '{print $7}' | sort | uniq -c | sort -rn | head -20

Myths worth dropping#

The action checklist#

Everything above, in the order we’d do it. Tick items off as you go — your progress is saved in this browser.

Perplexity SEO: frequently asked questions#