Perplexity: how it finds sources and how to be one of them

Perplexity is an answer engine: you ask a question, it searches the web, reads a set of pages and writes an answer with numbered citations linking to them. Unlike assistants where sources are optional or hidden, citations are central to how Perplexity presents itself. For site owners that has two consequences. Being cited is visible and checkable, and the traffic it sends tends to come from people who want to verify or act on something specific.

How Perplexity gets its sources

Perplexity documents two agents:

Agent	What it does	robots.txt
PerplexityBot	Crawls the web to build Perplexity's search index, from which answers draw sources	Perplexity says it respects robots.txt for this crawler
Perplexity-User	Fetches a page when a user's question requires reading it, for example a link the user provided	Perplexity says these user-initiated fetches generally do not follow robots.txt

Some publishers have questioned whether all of Perplexity's fetching follows their stated rules. If strict control matters to you, enforce it at the firewall or CDN level, not only in robots.txt. For most businesses that want to be cited, the opposite question matters: make sure PerplexityBot can reach you. The AI bot checker shows whether your robots.txt allows it, and the AI crawlers guide explains how the rules combine.

How it differs from a search results page

On a traditional results page, ten links compete for one click, and position decides most of it. In an answer engine, the searcher reads a synthesised answer first, and the sources appear as supporting references. Three things follow. First, being cited for one specific claim inside an answer is a realistic goal even for a small site, because an answer may draw on six or eight sources. Second, the click, when it comes, is usually a verification or next-step click: checking a specification, reading a full procedure, buying. Third, the brand mention inside the answer text can matter as much as the link, because many readers never click at all but remember the name.

What makes a page a likely citation

Perplexity answers by assembling claims from several sources, and each claim tends to be attributed to a specific page. Pages get cited when they contain claims worth attributing:

• Specific, verifiable facts: figures, dates, prices, specifications, steps, definitions. "Most standing desks range from 70 to 120 cm in height" is citable; "standing desks come in many sizes" is not.
• Clear structure: a heading that matches the question and a direct answer below it. Tables for comparisons.
• Original information: your own tests, data, documentation or first-hand experience. If ten pages say the same thing, any of them can be cited; if only yours has the measurement, yours is.
• Freshness: answers to "current" questions favour recently updated pages with visible dates.
• Accessible HTML: content in the page source, not loaded later by JavaScript or hidden behind interactions.

The guide to content AI assistants quote goes into more detail on writing for this.

What kinds of questions favour smaller sites

Broad questions ("what is the best CRM") tend to be answered from large, well-known sources. Specific questions open the field: a particular use case, a niche product, a local issue, a technical detail. A small business that documents its niche thoroughly, such as a specialist bike shop explaining compatibility between specific drivetrain components, can be cited for exactly those questions. That makes long-tail documentation, which often earns little traditional search traffic, more valuable than it used to be.

Checking whether you are cited

Perplexity's visible citations make manual checks straightforward:

1. Write questions your customers ask, without your brand in them.
2. Ask them in Perplexity and record which sources are cited, whether your domain appears, and whether your brand is named in the answer text.
3. Repeat periodically. Answers vary between runs and change as the index updates.
4. Note competitors that are cited where you are not, and look at what their cited page offers.

Also ask a few questions about your business by name, to check that the facts Perplexity states about you (location, prices, products, opening hours) are correct, and note which sources it relies on for them. Treat all results as a sample, not a ranking. The AI visibility guide explains how to set up a question set and avoid common measurement errors.

Measuring the traffic

Visits from Perplexity citations usually carry perplexity.ai as the referrer and can be grouped in analytics as an AI referral channel; the AI referral traffic guide shows how. In server logs, PerplexityBot requests show what is being indexed and how often, and Perplexity-User requests show pages being read in response to user questions, which is evidence of use even when nobody clicks.

Common reasons a site is never cited

• Blocked by robots.txt through a broad "block all AI" list that included PerplexityBot.
• Blocked by bot protection at the CDN, which returns challenges to crawlers.
• Content rendered client-side, so the crawler sees an empty shell.
• Thin or generic pages with nothing specific to attribute.
• Information only in PDFs or images, which are harder to parse and quote.
• Outdated pages on topics where freshness matters.

Pages worth creating for answer engines

Some page types are cited far more often than others, because they contain exactly the kind of statements answers are built from:

• Specifications and compatibility pages, with full details in text and tables.
• How-to procedures with numbered steps, required tools and common errors.
• Comparison pages that state differences plainly, with a table and a recommendation by use case.
• Pricing and policy pages that are clear and current: what is included, what costs extra, how returns work.
• Original data with the method explained, which answer engines can attribute to you specifically.
• FAQ sections on real customer questions, each answered in two or three sentences.

Many sites already have this information in sales decks, PDFs or support macros. Turning it into crawlable pages is often the quickest win.

Accuracy and corrections

Sometimes Perplexity cites your page for something it does not say, or summarises it wrongly. The source of the error is usually ambiguity: a page that discusses several options without clearly stating which applies, a table without units, an old page and a new page on your site that disagree. Clarifying the page, removing or redirecting outdated duplicates, and stating key facts in unambiguous sentences reduces misquotation across all AI tools, not just this one. Perplexity also offers feedback controls on answers, which are worth using for clear factual errors about your business.

A worked example

A hypothetical maker of replacement parts for espresso machines notices it is never cited when people ask which gasket fits a particular popular machine, although it sells exactly that part. Its product pages show compatibility only in an image of a chart, and the part names use internal codes. It rewrites the pages to state compatibility in plain text ("fits group heads of models A, B and C made after 2015; for older models use part X"), adds a compatibility table in HTML, and publishes a short guide on how to identify which gasket a machine needs, with measured diameters. Within a few weeks of PerplexityBot recrawling the pages, the guide and product pages begin appearing among cited sources for those specific questions.

The lesson generalises: information that exists only in images, PDFs or internal codes is invisible to answer engines, however useful it is to the customers who find it.

Summary checklist

1. PerplexityBot allowed in robots.txt and not blocked at the CDN.
2. Main content present in the HTML.
3. Key facts stated plainly in text and tables, with units and dates.
4. Specific pages for specific questions your customers ask.
5. Regular citation checks with a fixed question set.
6. AI referral channel in analytics; crawler activity reviewed in logs.