How Does Perplexity Choose the Sources It Cites?

Perplexity is the easiest of the major assistants to reason about, because it shows its sources as numbered links on every answer. That makes it the best place to start measuring whether your business is cited at all. It also means the question in the title has a documented half and an undocumented half, and this article keeps them apart.

What Perplexity documents about how it answers

Perplexity’s help center article on how the product works is short, and three sentences in it carry the weight. Perplexity “uses advanced AI to search the internet in real-time, gathering insights from top-tier sources.” It then “compiles the most relevant insights into a coherent, easy-to-understand answer.” And “each answer includes numbered citations linking to the original sources, allowing you to easily verify the information or explore further.”

Read as an engineer, that describes a retrieval pipeline: a search runs at question time, a set of pages is selected, the relevant passages are compiled into prose, and the pages that contributed are numbered and linked. Everything that follows about getting cited is about being in that selected set and being easy to compile from.

What the article does not say matters as much. It does not list ranking signals, name a freshness window, or describe how “top-tier” is decided. We cover the terminology of AI citations separately; here, treat the undocumented part as exactly that.

The two agents: PerplexityBot and Perplexity-User

Perplexity’s bot documentation is more concrete, and it splits the work into two user agents with different rules.

• PerplexityBot is “designed to surface and link websites in search results on Perplexity” and “is not used to crawl content for AI foundation models.” This is the indexing agent. Perplexity recommends allowing it in robots.txt if you want to appear, and publishes its IP ranges at perplexity.com/perplexitybot.json.
• Perplexity-User handles the moment a person asks. When a user poses a question, it “might visit a web page to help provide an accurate answer” and may “include a link to the page in its response.” Because a human requested the fetch, it “generally ignores robots.txt rules.” Its IP ranges are published at perplexity.com/perplexity-user.json.

The distinction decides where a citation can fail. If PerplexityBot cannot crawl you, you are unlikely to be in the candidate set at all. If Perplexity-User is blocked by a firewall when it tries to read your page for a specific question, the answer for that user is built without you. Both have to work. The wider cast of agents from OpenAI, Anthropic and Google is in our guide to which AI crawlers visit your site.

Firewalls: Perplexity’s own advice

Because a user-agent string can be typed by anyone, Perplexity tells site owners configuring Cloudflare or AWS WAF rules to “combine both User-Agent string matching and IP address verification,” using its published JSON lists as the source of truth and refreshing them periodically. The same advice protects you in the other direction: if you intend to allow Perplexity, verify that your bot-management layer is not challenging it anyway. A managed challenge page returned to Perplexity-User is a citation that never happens. The ten-minute version of that check is in is your CDN blocking AI crawlers.

The five levers that follow from the documentation

None of these is a published ranking factor, because Perplexity publishes none. Each is the practical consequence of something the documentation does state.

Access first

The first two levers are binary and cheap to fix, which is why they come first. A page PerplexityBot cannot index is not a candidate. A page Perplexity-User cannot open is not a citation. Most businesses that are “never on Perplexity” fail one of these without knowing it, usually because of a bot rule written for a different purpose.

Then quotability

Perplexity compiles insights and links to where it found them. The page that states the answer plainly in its first paragraph, and backs it with a table or a list, gives the compiler something to lift and a reason to attach your number to it. A page that circles the question for 600 words before answering is harder to use, for a machine and for a person.

Then consistency and credibility

The help center’s words are “top-tier” and “authoritative.” Perplexity does not define them, so we do not pretend to know the weighting. We do know what a careful human editor means by them: facts that agree with other sources, a named author, a date, and references. The same work that earns a citation from a journalist is the work that makes a page safe for an answer engine to cite.

How to test it yourself in an hour

Perplexity’s numbered citations make it the cleanest assistant to measure, so build your baseline here before anywhere else.

1. Write ten real customer questions. Not keywords. The sentences a prospect actually types: “who does bathroom remodels in Roseville with a fixed price,” “is a headless CMS worth it for a ten-page site.” Pull them from sales calls and inbound emails.
2. Ask each one in a fresh session and record every domain that appears as a numbered citation, plus whether your business is named in the prose at all.
3. Note the competitors that appear and open the pages Perplexity cited for them. Compare their first paragraph, their tables and their bylines to yours. That comparison is usually more instructive than any guide, including this one.
4. Repeat monthly with the identical list. Answers vary between runs, so report the share of questions where you were cited across the month, not one result.

The full measurement setup, including server logs and analytics referrals for every assistant, is in how to measure AI search visibility. Our earlier guide to Perplexity business visibility covers positioning for local and service businesses, and the whole sequence sits inside the AI Search Playbook.

What we deliberately did not claim

You will find articles stating that Perplexity weights domain authority, recency, backlinks or a specific schema type. Perplexity publishes none of that. Its documentation covers what its agents are for, how robots.txt applies, where its IP lists live, and that answers are compiled from live search with numbered citations. We built the levers above from those statements and labeled the rest as inference. When a vendor documents more, we will update this page and say so.

Frequently asked questions

About the work

Webvello is an AI search optimization and index engineering company founded in 2024, based in Roseville, California. We’ve delivered 50+ projects across 37+ cities, with average client ROI in the 3–5x range on properties where the technical foundation was the binding constraint. Founder George Shvaya came to search from cybersecurity and systems engineering, which is why this article separates what a vendor documents from what people assume about it. Our generative engine optimization work starts with that distinction.

Sources: Perplexity’s help center article on how Perplexity works and its bot documentation, both linked above and re-read on October 16, 2026. Perplexity changes its product and its agents; check the linked pages before relying on a statement here.