People notice something odd about Perplexity. They search for a topic, read the answer, and see five or six sources cited. Then they look at the same query in Google and realize Perplexity cited pages ranking in positions 3, 7, and 12, but skipped the page in position 1. Why?
It is not random. Perplexity runs a retrieval pipeline, and at each stage, pages get filtered out. Some never get retrieved. Some get retrieved but are too hard to read. Some get read but do not contain the specific information Perplexity needs to build its answer.
If you understand the pipeline, you can figure out why your pages are or are not getting cited, and what to actually change.
Perplexity is a retrieval-augmented search engine
Perplexity does not answer questions from memory alone. It runs a live web search for each query, pulls relevant pages into a working memory, reads them, and writes a synthesized answer with numbered citations linking back to the sources.
This architecture is called retrieval-augmented generation, or RAG. It is the same fundamental approach used by ChatGPT Search, Google AI Overviews, and most other answer engines. The specifics differ, but the pipeline is similar: find pages, read pages, write an answer, attach citations.
The important thing to understand is that citation is the last step in a chain. If a page fails at any earlier step, it never reaches the citation stage. Most pages that do not get cited fail at retrieval or extraction, not at the final synthesis stage.
Here is how each step works.
Step 1: Retrieval, or how Perplexity finds candidate pages
When you ask Perplexity a question, it needs to find pages that might contain the answer. It does this through a combination of its own web crawl and third-party search indexes.
Perplexity's crawler is called PerplexityBot. It discovers and fetches web pages, building Perplexity's own index of the web. But Perplexity does not crawl everything. No crawler does. It crawls a subset of pages, prioritizing pages it has discovered through links and pages that are already in search indexes it has access to.
This means two things have to be true for your page to get retrieved.
First, PerplexityBot needs to be able to crawl your site. If your robots.txt blocks PerplexityBot, or if a CDN, firewall, or security plugin blocks it before it reaches your server, Perplexity cannot build its own copy of your content. You can check this by running a curl test from your terminal:
curl -A "PerplexityBot" -I https://yourdomain.com/A 200 response means the crawler can reach you. A 403 means something is blocking it. Many sites block PerplexityBot accidentally through Cloudflare Bot Fight Mode, WordPress security plugins, or generic "block all AI" configurations that do not distinguish between training crawlers and search crawlers.
Second, your page needs to be discoverable in the indexes Perplexity searches. Perplexity supplements its own crawl with results from traditional search indexes. If your page does not rank well enough for the query to appear in those indexes, Perplexity will not retrieve it.
This is why pages ranking in the top organic results get cited more often. They are more likely to be retrieved in the first place. Pages buried on page three or four of traditional search rarely make it into Perplexity's candidate pool.
Pages that fail at retrieval are invisible to Perplexity. They never make it to the next step. No amount of content quality matters if the retrieval system never pulls your page.
Step 2: Extraction, or how Perplexity reads and filters pages
Once Perplexity retrieves a set of candidate pages, it reads each one. This is where a lot of pages get filtered out, even though they were successfully retrieved.
Perplexity's language model reads the content of each page and looks for information that directly answers the question. It is trying to extract specific, usable answers from the text. Pages that make this easy survive. Pages that make it hard get dropped in favor of pages that are easier to read.
A few things determine whether your page survives extraction.
Does the page state the answer clearly?
Perplexity's model looks for content that directly addresses the query. If your page has a heading that matches the question and a concise answer in the first paragraph, the model can extract that answer with high confidence.
If your page buries the answer under three paragraphs of context, company history, or preamble, the model has to work harder. It might still find the answer. But if another retrieved page states the same answer more clearly and earlier, Perplexity will probably cite that page instead.
The pages that get cited most often follow a simple pattern: a heading that mirrors the question, a direct answer in the first 40 to 60 words, and supporting detail below. A human reader can find the answer in under five seconds. If a human can, Perplexity can too.
Does the page contain specific information?
When Perplexity synthesizes an answer from multiple sources, it prefers sources that add specific, non-obvious information. If five retrieved pages all say "structured data helps with AI search," Perplexity cites one of them. If your page says "pages with FAQPage schema see higher citation rates in our audit data," that is the page that gets cited because it contributes something the other sources do not.
Numbers, dates, percentages, named sources, and specific examples all make your page more valuable to Perplexity's synthesis engine. The model is trying to build the most useful answer possible. It pulls from the sources that give it the richest material.
Can the model parse the HTML?
Perplexity reads the rendered HTML of your page. If the HTML structure is clean and semantic, extraction is reliable. If the structure is messy, extraction gets harder.
The problems that hurt extraction are the same ones that hurt traditional SEO and accessibility. Multiple H1 tags on a single page. Heading levels that skip (H1 followed by H3 with no H2). Content hidden behind JavaScript that does not render in the initial HTML. Important text embedded in images with no alt text.
Pages with one clear H1, a logical heading hierarchy, and content in semantic HTML elements like article and section are easier for Perplexity to parse. The model can identify which sections are headings, which are body text, and how the content is organized.
Does the page have structured data?
JSON-LD structured data helps Perplexity understand what your page is about and what type of content it contains. The schema types that matter most for citation:
- Article or BlogPosting for editorial content
- FAQPage for question-and-answer content
- HowTo for step-by-step guides
- Organization for your homepage
Structured data is not a requirement for citation. Pages with no schema get cited when the content is well-written and easy to extract. But schema increases your odds, especially for competitive queries where Perplexity retrieves many good pages and has to choose between them.
FAQPage schema deserves a specific mention. When you mark up Q&A pairs with FAQPage schema, you give Perplexity pre-formatted question and answer pairs it can pull directly. The model does not have to parse your paragraphs and guess at intent. The questions and answers are already structured for extraction. Every major AI search engine reads FAQPage schema, including Perplexity, ChatGPT Search, Google AI Overviews, and Claude.
Step 3: Synthesis and citation
This is the final step. Perplexity has retrieved a set of pages, extracted information from each, and now needs to write an answer and decide which sources to cite.
The model synthesizes an answer by combining information from the extracted sources. It writes a response that directly answers the query, drawing on the specific details, data points, and claims it found across the retrieved pages.
Citations are assigned to the specific claims in the answer. When Perplexity writes a sentence that draws on information from a particular source, it attaches that source's citation number to the sentence. Sources that contributed specific, unique information get cited. Sources that only repeated what other sources already said might not.
This is why specific data matters so much. If your page is the only retrieved source that contains a particular statistic, date, or example, Perplexity has to cite you to include that information in its answer. If your page says the same thing as five other pages, Perplexity picks one and cites it. Being the source of unique information is the most reliable way to get cited.
Why Perplexity might skip your page even when it ranks well
The most common scenario: your page ranks in the top organic results for a query, but Perplexity does not cite it. This happens at the extraction stage.
Your page was retrieved. Perplexity read it. But when the model looked for information to build its answer, your page did not contribute anything that the other retrieved pages did not already provide, or the information was too hard to extract.
Common causes:
The answer is buried. Your page contains the answer, but it is three paragraphs into a long section with no heading. Another page puts the answer in the first sentence under a clear heading. Perplexity picks the easier extraction.
The page is too general. Your page covers the topic broadly but does not contain specific claims, data, or examples. Another page has the exact statistic Perplexity needs. That page gets cited.
The content is hard to parse. Your page uses non-semantic HTML, has inconsistent heading structure, or serves content through JavaScript that does not render in the initial HTML response. The extraction model struggles and moves on to a cleaner page.
The page is a duplicate of other sources. If ten retrieved pages all say the same thing, Perplexity cites one or two. Being generic is the same as being invisible at this stage.
How to check whether Perplexity is finding and citing you
Go to perplexity.ai and ask questions your customers would ask. Use the exact phrasing a person might use, not your internal terminology. Note whether your site appears as a cited source.
Then check your server access logs for PerplexityBot activity. Filter your logs for the PerplexityBot user-agent string:
grep "PerplexityBot" access.log | awk '{print $7}' | sort | uniq -c | sort -rnThis shows every URL PerplexityBot requested and how many times. If you see zero hits over a period of weeks, either Perplexity is not discovering your site or it is being blocked. The curl test above tells you which.
If PerplexityBot is crawling your pages but Perplexity is still not citing you, the problem is at the extraction stage. Your content is being retrieved but not surviving the read. That is a content structure problem, not a crawlability problem.
How this fits the IAB visibility framework
The IAB published "Measuring Visibility in the AI Era" on August 3, 2026. It defines four dimensions of AI visibility: Presence, Prominence, Portrayal, and Persuasion.
Perplexity's citation pipeline maps cleanly to the first two. Retrieval determines Presence. If Perplexity cannot retrieve your page, your Presence metrics are zero. Extraction determines Prominence. If your page is retrieved but not cited because the content is hard to extract, you are technically present in the candidate pool but not prominent in the answer.
This distinction matters for diagnostics. If your AI visibility tool reports that you are not appearing in Perplexity answers, you need to know whether you are failing at retrieval or extraction. They require different fixes. Retrieval failures are crawlability and ranking problems. Extraction failures are content structure and formatting problems. A Parceit audit checks both layers and tells you which gate you are failing.
Frequently asked questions
How does Perplexity decide which sources to cite?
Perplexity retrieves candidate pages from its web crawl and search indexes, reads each page to extract relevant information, and synthesizes an answer citing the sources that contributed specific, usable content. Pages that directly answer the question, contain specific data, and use clean semantic HTML are more likely to be cited than pages that bury answers or speak in generalities.
Does Perplexity cite pages that do not rank in the top Google results?
Yes, but less often. Perplexity supplements its own crawl with third-party search indexes, so pages that rank well in traditional search are more likely to be retrieved. Pages outside the top organic results are sometimes cited, but the odds are lower because they are less likely to enter Perplexity's candidate pool in the first place.
Why does Perplexity cite a lower-ranking page instead of the page in position one?
This happens at the extraction stage. The page in position one may have been retrieved, but Perplexity's model found it easier to extract a clear, specific answer from a different page. If the top-ranking page buries the answer under introductory content while a lower-ranking page states it directly under a matching heading, the lower-ranking page gets cited.
Does FAQPage schema help with Perplexity citations?
Yes. FAQPage schema gives Perplexity pre-formatted question and answer pairs it can extract directly from your HTML. Instead of parsing paragraphs and guessing at intent, the model reads the structured Q&A pairs. Every major AI search engine reads FAQPage schema, including Perplexity.
How can I tell if Perplexity is crawling my site?
Check your server access logs for the PerplexityBot user-agent. If you see requests, Perplexity is crawling your pages. If you see zero requests over several weeks, either your site is not being discovered or PerplexityBot is being blocked. You can test for blocks by running curl with the PerplexityBot user-agent string against your domain.
What is the difference between being retrieved and being cited by Perplexity?
Retrieval means Perplexity pulled your page into its candidate pool for a query. Citation means your page was one of the sources Perplexity actually referenced in its answer. Many pages get retrieved but never cited because they did not survive the extraction step. They were too hard to read, too generic, or did not contain specific information the model needed.
Want to know exactly where your site stands on the signals that determine Perplexity citations? Run a free audit at parceit.com. You see your full score and every issue in under 60 seconds.