Question
How Perplexity Decides Which Sources to Cite in Its Answers
Chat assistants 4 min read
Perplexity’s public documentation explains how it reaches your site, not how it ranks what it finds. Its crawler documentation states that “PerplexityBot is designed to surface and link websites in search results on Perplexity” and that “it is not used to crawl content for AI foundation models.” Beyond access, that documentation does not describe how Perplexity selects among the pages it finds. What can be said about how Perplexity chooses sources comes from two kinds of outside evidence: controlled experiments run on Perplexity, and audits of the answers it gives.
How Perplexity reaches your site
Perplexity documents two agents, and they follow different rules.
- PerplexityBot crawls the web to surface and link sites in Perplexity’s search results. Perplexity recommends allowing it in robots.txt, and permitting its published IP ranges, for sites that want to appear in results.
- Perplexity-User fetches pages when a user’s request prompts it. Perplexity says: “Since a user requested the fetch, this fetcher generally ignores robots.txt rules.”
Perplexity says each setting works independently and that changes may take up to 24 hours to take effect. Its documentation also notes that sites behind a web application firewall may need to allow its bots explicitly, which is worth checking if your pages never appear in Perplexity despite an open robots.txt.
What controlled experiments on Perplexity show
Direct evidence about what helps a page get used in Perplexity’s answers comes from the GEO paper, presented at KDD 2024 by researchers from IIT Delhi and Princeton. Alongside tests on their own generative engine, the authors ran their methods on Perplexity itself and measured how much of a source’s text appeared in the answer and how prominently.
On Perplexity, the authors report that Quotation Addition, adding relevant quotations from credible sources, performed best on their word-count visibility measure, improving it by 22% over the baseline. Methods that did well on their own engine, such as Cite Sources and Statistics Addition, showed improvements of up to 9% and 37% on the two measures. Keyword stuffing, a traditional SEO tactic, performed 10% worse than the baseline.
Two limits apply. The experiments measure visibility of a source’s text in the answer, not whether Perplexity links to it. And they were run on the version of Perplexity available to the researchers in 2023 and 2024; the system has changed since. The direction of the findings, that sourced, quotable and specific content does better than keyword repetition, is the useful part.
What audits of Perplexity’s answers show
Independent audits have looked at how accurately Perplexity, among other tools, cites sources.
The Tow Center for Digital Journalism tested eight AI search tools in early 2025 by giving them excerpts from news articles and asking them to identify the source. Collectively, the tools answered more than 60% of queries incorrectly. Perplexity answered 37% incorrectly, while Grok 3 answered 94% incorrectly.
The same audit examined publishers that block AI crawlers. Tow reports that Perplexity Pro correctly identified nearly a third of 90 excerpts from articles it should not have had access to, and that Perplexity’s free version correctly identified all ten excerpts from paywalled National Geographic articles, even though that publisher disallows Perplexity’s crawlers. The researchers note that there are other means through which a chatbot could obtain information about restricted content, such as references to the work in publicly available sources.
A broader study coordinated by the European Broadcasting Union and the BBC, published as a News Integrity toolkit in October 2025, evaluated more than 3,000 responses from ChatGPT, Copilot, Perplexity and Gemini across 22 public broadcasters. It found at least one significant issue in 45% of the responses overall.
For a site owner, the audits mean two things. Being cited does not guarantee being represented accurately. And robots.txt is not the only route by which information about your content can reach an answer.
What you can do
- Allow PerplexityBot if you want to appear in Perplexity, and check that your firewall does not block its published IP ranges.
- Understand Perplexity-User. Requests made on a user’s behalf generally ignore robots.txt, according to Perplexity. Decide whether any server-level controls matter to you.
- Write content that is easy to use accurately. The GEO results on Perplexity favor quotations from credible sources, specific statistics and cited claims over keyword repetition. Clear, sourced statements can also be harder to misrepresent.
- Check how Perplexity describes you. Ask it questions about your brand and your key topics, repeat them over time, and note both whether you are cited and whether the citation is accurate.
- Do not expect stability. A preprint by Sielinski, posted in July 2026, examined rank stability across Gemini, SearchGPT and Perplexity and concluded that no fixed number of samples can be justified across contexts for measuring visibility. Measure with repeated prompts, not single checks.
Perplexity is a search engine with its own crawler and its own answer layer, so plan for it separately. Open access to PerplexityBot, publish sourced and quotable content, and check both whether you are cited and whether you are cited correctly, since the published audits show those are separate questions.