How does Perplexity choose which pages it cites?
Last updated October 14, 2026. Written by the Third Coast AI team.
Perplexity chooses its citations by searching the web in real time for each question, reading the pages its own index and crawler have made available, and attaching numbered citations to the sentences those pages supported, which means a company is cited when PerplexityBot can reach its pages, the pages answer the question in plain HTML, and the content is current enough and specific enough to be picked over the alternatives for that search. Perplexity publishes two agents with different rules, its crawler PerplexityBot, which honors robots.txt and exists "to surface and link websites in search results on Perplexity," and Perplexity-User, which fetches a page when a person asks for it and "generally ignores robots.txt rules."
This page sets out what Perplexity has published about its crawlers and its answer process, what the independent studies measured about its citations, and what we do differently for Perplexity compared with ChatGPT and Google. For the broader method, see how to show up in ChatGPT and the other engines.
Which Perplexity agents visit your site, and what do they do?
Two, and the distinction matters for anyone considering a block. Perplexity's crawler documentation describes PerplexityBot as "designed to surface and link websites in search results on Perplexity," says it respects robots.txt and that site owners should allow it if they want search visibility, publishes its IP ranges at perplexitybot.json, and states that it is "not used to crawl content for AI foundation models." Perplexity-User is different: it "supports user actions within Perplexity," visiting a page "to help provide an accurate answer" when a person asks a question that needs it, and the documentation says that "since a user requested the fetch, this fetcher generally ignores robots.txt rules." Neither agent is used for model training according to the same page, which separates Perplexity's position from OpenAI's, where GPTBot and OAI-SearchBot split training from search.
PerplexityBot vs Perplexity-User
PerplexityBot is the index crawler: it honors robots.txt, is not used for training, and is the agent that decides whether your pages can appear in Perplexity's search results. Perplexity-User is the on-demand fetcher that follows a person's request in the product, and Perplexity says it generally ignores robots.txt because the person asked for the page. Blocking the first removes you from retrieval; blocking the second does little, since it will still fetch when a user pastes your link.
Does Perplexity search live, or from a stored index?
Both, in the sense that it maintains an index through PerplexityBot and also searches at answer time. Perplexity's help center describes the product as using AI "to search the internet in real-time, gathering insights from top-tier sources," says it gathers "information from authoritative sources like articles, websites, and journals," and explains that "each answer includes numbered citations linking to the original sources." Its Deep Research mode "performs dozens of searches automatically, reads hundreds of sources," which is the same many-searches-behind-one-question mechanism Google describes for AI Mode, run at greater depth. The help page does not publish the ranking criteria beyond the word authoritative, so what follows on selection comes from measurement rather than from Perplexity.
The useful consequence of real-time search is that Perplexity can pick up a new or refreshed page faster than an engine whose answers lean on training data, so a corrected page can appear in a Perplexity answer before the other engines pick it up. The cost is that it also drops pages faster when something fresher appears, which the drift studies below bear out.
Can Perplexity read JavaScript-rendered pages?
No. Vercel's December 2024 crawler study examined request logs across its network and concluded that "none of the major AI crawlers currently render JavaScript," listing PerplexityBot alongside OpenAI's and Anthropic's agents, and it recorded PerplexityBot at 24.4 million fetches in the sampled month, a small fraction of GPTBot's 569 million and Googlebot's 4.5 billion. The volume number is worth a moment: Perplexity's crawler visits far fewer pages than Google's, so a site that relies on Perplexity discovering a deep page on its own may wait a long time, and clean internal linking from pages it already knows matters more than it does for Google. The JavaScript rule is the same as for ChatGPT, and the rendering page covers how to test and fix it.
How stable are Perplexity's citations?
More stable than ChatGPT's and less stable than AI Overviews. Profound's July 2025 volatility study compared identical prompts a month apart on roughly 80,000 prompts per platform and found 40.5% of Perplexity's cited domains had changed, the lowest drift of the four platforms it measured, against 59.3% for Google AI Overviews, 54.1% for ChatGPT and 53.4% for Copilot. SparkToro's January 2026 study of 2,961 prompt runs across ChatGPT, Claude and Google's AI found the same brand list returned less than 1 time in 100 for a repeated question, and while Perplexity was not in that sample, our own repeated runs show the same pattern of a stable core of sources with the remainder rotating. The Princeton researchers (Aggarwal and colleagues, KDD 2024) also replicated their quotation, statistics and cited-source findings on Perplexity, with smaller but still positive effects, so the page-level craft transfers.
Perplexity is the engine that notices first. When a client fixes a wrong fact or publishes the comparison page nobody had written, Perplexity usually cites it before ChatGPT or Google do, and it also moves on first when someone publishes something better.
Third Coast AI
Should you treat Perplexity differently from ChatGPT?
In three ways, and otherwise the same. The common ground is crawler access, raw HTML, answer-first passages with numbers and named sources, and a strong third-party record, since Ahrefs' 75,000-brand study found mentions on other sites correlate with AI citations three times as strongly as backlinks and that signal appears to carry in every engine we run. The differences are about freshness, breadth and discovery.
- Freshness carries more weight. Refresh the pages that earn Perplexity citations on a schedule, and change the content rather than the date, since a header-only refresh is detectable. The refresh guide covers the routine.
- Breadth of sources pays. Perplexity attaches numbered citations to many sentences, so a page that answers several sub-questions can earn several citations in one answer rather than competing for a single slot.
- Discovery is slower. PerplexityBot fetches a fraction of what Googlebot fetches, so link new pages from pages Perplexity already cites and keep a current sitemap.
- Allow PerplexityBot explicitly in robots.txt and check your CDN against the published IP ranges, and do not bother blocking Perplexity-User, which will fetch on request regardless.
- Measure with repeated runs and report a rate; with Profound's 40.5% monthly drift, a single answer tells you little about next month.
Our method does not change by engine, only the emphasis does. We build the prompt bank the way buyers ask, run it with retrieval on in Perplexity alongside the others, capture the hidden searches and decision factors, audit the client's pages for proof of each factor, read the pages Perplexity already cites for the topic, and write briefs modeled on them, then re-run to verify. The audit is where that starts.
Frequently asked questions
Perplexity's crawler documentation says neither PerplexityBot nor Perplexity-User is used to collect content for training AI foundation models. PerplexityBot exists to surface and link sites in Perplexity's search results.
Generally not. Perplexity says Perplexity-User fetches pages when a person asks for them and generally ignores robots.txt for that reason. PerplexityBot, the index crawler, does honor robots.txt.
No. Vercel's December 2024 log study found none of the major AI crawlers, PerplexityBot included, render JavaScript, so content loaded by scripts is invisible to it.
Profound measured 40.5% of Perplexity's cited domains changing over one month on identical prompts, the lowest drift of the four platforms in its study, which still means a single run is not a reliable measure.
Yes. Aggarwal and colleagues replicated their findings on Perplexity, where quotations, statistics and cited sources still lifted visibility, with smaller effects than on their test engine.