Perplexity answers a question by searching the web in real time, reading a small set of retrieved pages and writing an answer with numbered citations that link back to those pages. To be visible there, a brand has to clear three hurdles: its pages must be retrievable, they must be selected among the candidates, and the brand must actually be named in the generated text. Perplexity does not publish a ranking formula, so no tactic guarantees a citation, but each hurdle can be influenced and measured.
Key takeaways
- Perplexity is a retrieval-first answer engine: what it cites depends heavily on what its search step finds for that specific question, not only on what the model "knows".
- Being cited (your URL appears as a source) and being mentioned (your brand is named in the answer) are different outcomes and should be tracked separately.
- Technical access comes first: Perplexity's own documentation says sites that want to appear in its results should allow PerplexityBot in robots.txt.
- Pages with clear, self-contained, evidence-rich passages are easier to quote; the GEO study (Aggarwal et al., KDD 2024) measured gains of up to 37% on Perplexity.ai for such optimizations.
- Answers vary between runs, so visibility is a rate measured over a fixed question set, not a single screenshot.
How does Perplexity find and choose its sources?
Unlike a chatbot answering from memory alone, Perplexity is built around retrieval-augmented generation: every answer starts with a search. The exact internals are proprietary, but the publicly visible behavior and the developer documentation allow a reasonable model of the pipeline:
- Interpreting the question. The question is turned into one or more search queries. For complex questions, and especially in the deeper research modes, the system may run several searches covering different sub-questions.
- Retrieving candidates. Results come from Perplexity's own index, which is built by its crawler, PerplexityBot. A page that has never been crawled cannot be retrieved.
- Selecting and reading. Only a limited number of results are read in detail. Passages that answer the sub-question directly have an obvious advantage over pages where the answer is buried.
- Writing and citing. The model composes the answer and attaches numbered citations to the claims that came from specific sources.
The developer API makes this structure visible. A Sonar API response returns the answer together with a citations list and search_results objects that carry the title, URL, date and snippet of each source. Developers can also narrow retrieval with parameters such as search_domain_filter (up to 20 domains to include or exclude) and search_recency_filter. These are API features rather than a description of consumer ranking, but they show the dimensions retrieval works with.
Why search ranking alone does not decide the outcome
Strong organic rankings help because they correlate with relevance and authority, yet the final selection happens at passage level. A page ranking fifth can be cited ahead of the first result if it contains the cleanest answer to one sub-question, for example a comparison table or a clearly stated price.
Citation or mention: two kinds of visibility
Perplexity's interface makes sources very prominent, which tempts teams to count only citations. That leaves half of the picture out. Our article on which sources AI cites explores this in depth; the short version:
| Outcome | What it means | Typical cause |
|---|---|---|
| Cited and mentioned | Your page is a source and your brand is named | Strong own content on the exact question |
| Mentioned, not cited | Your brand is named, based on third-party pages | Reviews, comparisons or news about you |
| Cited, not mentioned | Your page informs the answer but your brand is not named | Generic educational content |
| Neither | You are absent | Not retrieved or not selected |
For a buying question such as "best project management tool for a 10-person agency" (a hypothetical example), the mention is often what matters commercially, and it frequently comes from third-party comparison pages rather than your own site. An AI citation is valuable, but it is not the same thing as being recommended.
What plausibly influences selection
Perplexity has not published its ranking factors, so the following list combines documented behavior with research evidence and should be read as informed hypotheses, not rules.
Relevance at the level of a passage
Retrieval works on chunks of text. Passages that state a definition, a number or a recommendation in one or two sentences, with the subject named explicitly instead of "it" or "our solution", survive extraction better. Our guide on how to write content that AI quotes covers this structure in detail.
Evidence density
The GEO paper by Aggarwal et al. (arXiv:2311.09735, KDD 2024) tested rewriting strategies on a benchmark of queries and on Perplexity.ai itself. Adding quotations, statistics and citations to credible sources produced the largest visibility gains, while classic keyword stuffing performed poorly. The improvements reported on Perplexity.ai reached up to 37%. These are research averages, not a promise for any single page.
Freshness
Because retrieval happens at question time, recently updated pages can be picked up quickly, and the API's recency filter shows that publication date is a usable signal. Visible update dates matter most for fast-changing topics such as prices and versions.
Third-party corroboration
When several independent sources say the same thing about a brand, an answer engine has more material to support naming it. Reviews, industry lists, analyst pages and forum discussions often supply the evidence for "mentioned, not cited" outcomes.
Technical prerequisites: let the right crawler in
According to Perplexity's crawler documentation, there are two user agents. PerplexityBot is designed to surface and link websites in Perplexity's search results, and Perplexity states it is not used to train foundation models. Perplexity-User fetches pages when a user's question requires it; because a person triggered the request, Perplexity says it generally ignores robots.txt. A minimal robots.txt rule that allows indexing looks like this:
User-agent: PerplexityBot
Allow: /
Three practical notes. First, firewalls and bot-protection services often block unknown crawlers by default; Perplexity publishes its IP ranges as JSON files so you can verify and allow genuine requests. Second, robots.txt changes can take up to 24 hours to take effect, according to the same documentation. Third, content that only appears after heavy client-side JavaScript, behind logins or inside images is harder for any crawler to read reliably.
How can a brand stand out in Perplexity?
A workable program looks like this:
- Map the real questions. List the questions customers actually ask in each market and language.
- Publish answer-first pages. Put the direct answer at the top, then the evidence: specifications, prices, methodology, limitations. Use tables for comparisons.
- Make claims verifiable. Cite primary sources, show dates and name authors with relevant expertise. Unsupported superlatives give a model nothing to quote.
- Earn independent coverage. Accurate listings on review platforms, inclusion in credible comparisons and expert commentary in trade media feed the "mentioned" outcome.
- Keep facts consistent. Your name, product names, prices and positioning should match across your site and third-party profiles so that retrieved snippets agree.
- Measure, change one thing, measure again. Without a baseline, you cannot tell whether an edit helped or whether answers simply varied.
How to measure Perplexity visibility honestly
A single answer proves little: the same question can produce different sources a day later. A sound approach fixes a question set, runs it repeatedly in the target market and language, stores every answer with its sources, and reports a presence rate: answers that mention the brand divided by valid answers, with failed answers excluded from the denominator. Citation share and competitor comparisons should come from the same sample. Our article on how to measure AI visibility explains sampling in detail, and GeoDog's methodology describes how it applies these rules when it queries Perplexity alongside ChatGPT, Gemini and Claude.
FAQ
Does blocking PerplexityBot remove my site from Perplexity?
Blocking PerplexityBot reduces the chance that your pages are indexed and surfaced in results. However, Perplexity documents that its user-triggered fetcher, Perplexity-User, generally ignores robots.txt, so a page may still be fetched when a user explicitly asks about it.
Is ranking well on Google enough to be cited by Perplexity?
It helps but is not sufficient. Perplexity uses its own index and selects passages per sub-question, so a well-structured page with clear evidence can be cited even without a top Google position, and a top page can be skipped.
How often should Perplexity visibility be measured?
For most brands, a monthly run of a fixed question set is a good baseline, with extra runs after major content changes or launches.