How I Earn a Perplexity Citation

I do not try to rank in Perplexity. Perplexity retrieves sources in real time, reranks them, and cites the ones that answer the query with something it can lift. The page is either in that set or it is absent for that query. A Google position does not settle it.

The method for becoming a source Perplexity can use is How to Get Cited by Perplexity: Be a Source It. That page is the deeper method. This essay is how I would use it as a founder: which of the five inputs I would fund first, what I would stop doing, and the measurement that would tell me the work is wasted.

What the system does on one query

Perplexity processes 30 million searches a day and sits on 138.1 million monthly visits of query demand. Every search is a citation decision. The system is retrieval-augmented generation. On a query I care about, four things happen.

The query is decomposed into 3 to 5 sub-queries. "Best CRM for Series A startups" becomes separate retrievals: CRM tools for startups, features for early-stage companies, the stack a Series A team actually buys. Each sub-query hits the web on its own.

Each sub-query retrieves 10 to 30 candidate pages, mixing keyword match with semantic similarity. That is the wide net.

A cross-encoder then scores query and document together. The published accounts of that step name entity relevance, topical match, source authority, freshness, and historical citation quality.

The last layer keeps 3 to 10 sources and cites them inline. The reader can open the source. I can too, which is why a citation I cannot find on the page does not count, in my own audit, as a citation I earned.

Google ranks pages against each other for a query. Perplexity lifts passages and scores them against a sub-question. A page can sit at the top of a Google result and receive nothing here because it was built to be clicked, not lifted.

The five inputs, in the order I would fund them

I reduce the controllable work to five inputs. I would not fund them evenly.

Crawl access comes first, because nothing else runs. If PerplexityBot is blocked in robots.txt or at the firewall, the page is not in the pipeline. I check two user agents, PerplexityBot and Perplexity-User. Bot protection blocks AI crawlers by default often enough that I ask the security owner before I brief a writer. The check takes minutes and it gates the other four. Pages rendered only in the browser, with no server-rendered HTML, are harder for the crawler. I want the HTML.

Passage extractability is the writing standard. Perplexity cites passages. The descriptions I trust put a self-contained block, usually 40 to 100 words, immediately after an H2 that asks the question. Comparison tables with named columns and 4 to 8 rows get lifted nearly verbatim. Ordered steps beat a prose procedure. Definition blocks of 30 to 60 words under a heading get extracted at the highest rate. If our answer sits three paragraphs down, the reranker can prefer a competitor who put the same fact under the heading. I would rewrite the heading and the first sentences before I commissioned a new URL.

Factual density is how a passage survives contact with the others. The reranker prefers specific data points: a number, a date, a named entity. "Many companies are adopting this" loses to a sentence that names the share, the period, and the source. I want at least two named entities and one specific number in a section I hope to see cited. Vague claims fail because the system cannot line them up against the other passages it just retrieved. I do not invent the number to satisfy the rule. A specific claim I cannot source is worse than a shorter section.

Corroboration is the input teams skip. The page is not scored alone. Other sources have to refer to the same claims. Reddit carries disproportionate weight on opinion and comparison queries. Reuters, AP, and major news outlets carry it on factual claims. A Wikipedia page or a .edu page that repeats the claim is corroboration. A press placement is a corroboration signal, which is the argument I make in how earned media shows up in AI search. I would buy the placement that a second source can point at. I would not buy a placement that restates our homepage in a place nothing else cites.

Freshness is a quarterly bill. Perplexity crawls the live web on the query. It is not answering from a frozen training cut. Pages updated within the last 90 days perform better than static pages, on the accounts of this system I use. I update datePublished and dateModified when the page actually changed. In AI, fintech, and software, I treat the half-life of a pillar as roughly 90 days. A "best tools" page last touched in January is losing to a weaker page touched last month. I put the top 20 pages on a 90-day rotation and I treat the rotation as a release, not as a hope.

What I would stop

Habit I see in our own drafts	What it does in this pipeline
A long introduction before the answer	The reranker prefers the passage where the answer is immediate
Thin pages stuffed with the keyword	Low factual density loses to a specific claim
Blocking AI crawlers "for protection"	The page never enters retrieval
Publishing once and leaving it	The 90-day freshness decay applies
No mention of the company anywhere else	The reranker has nothing to corroborate

Each of those habits has been rewarded by a classic ranking system at some point. This retrieval system rewards the opposite shape. I would delete the habit before I added a tool.

How I know it is working

Once a month I run 50 to 200 category queries and record whether we are cited, the position of the citation, and the passage that was lifted. Citations divided by the queries where we should have appeared is the share I manage. The definition and the formula live with the metric. I want the trend on a fixed set, not a new set every month that flatters the last campaign.

I also look at referral analytics for perplexity.ai. Perplexity accounts for 7.23% of AI referral traffic in the cut I use, behind ChatGPT at 74.78%, and it shows up in high-intent professional queries out of proportion to that share. Referral traffic is the visit. The citation is the selection. I do not confuse them. When we are cited, I read the passage that was extracted. The wrong passage means the structure failed. The right passage, attributed to someone else, means the entity signals failed.

How I would brief one page

Take a page we already hope will be cited. I write the buyer question as the H2. Under it I want 40 to 100 words that answer it, with two named entities and one sourced number, before any story about how we got here. If the answer is a comparison, I want a table of 4 to 8 rows with the columns named. If the answer is a definition, I want 30 to 60 words and then the nuance, not the other way around.

Then I split the question the way the system splits it. A query becomes 3 to 5 sub-queries, each retrieving 10 to 30 candidates, before a cross-encoder keeps 3 to 10 sources. If our page answers only the slogan and none of the sub-questions, we are in the wide net and out of the final set. I would rather add the missing sub-answer under its own heading than publish a second article that repeats the first.

The corroboration line on the brief is a list of places the same claim already exists, or the placement we are willing to earn so that it does. Reddit for a comparison the buyer is already arguing. A newsroom for a factual claim. I do not assign someone to "get us on Reddit" as a tactic detached from a claim the page makes. Spam is how that line reverses.

Freshness is a date on the brief. If dateModified is older than 90 days, the page is in the rotation whether or not we feel like editing it. The top 20 pages are the rotation. A pillar in AI, fintech, or software that has sat since January is, on the half-life I use, already losing to a weaker page touched last month.

The monthly read is a ratio on a fixed set. Fifty to 200 category queries. Citations divided by the queries where we should have appeared. If one month is 10 citations on 50 queries, that is 20% on that set, and the only comparison that matters is last month's same 50. I also open the passage Perplexity lifted. Wrong passage, fix the heading. Right passage under someone else's name, fix the entity. Referral traffic from perplexity.ai, 7.23% of AI referral traffic in the cut I use, against ChatGPT's 74.78%, is the visit. I do not let the visit replace the ratio.

Before any of that copy is assigned, I check PerplexityBot and Perplexity-User. A blocked crawler makes the brief a private document. Thirty million searches a day do not include a host that refuses the fetch.

The decision

I would not open a Perplexity program with a rank tracker or a new slogan. I would check the two user agents, rewrite the twenty pages that should already answer buyer questions so the answer sits under the heading with a sourced number, and put a corroborating placement against the claims those pages make. I would refresh that set every 90 days. I would read 50 to 200 queries a month and keep the set fixed.

The scope is Perplexity's retrieval path as the published accounts describe it: decomposition into 3 to 5 sub-queries, 10 to 30 candidates, a cross-encoder, 3 to 10 cited sources. The same five inputs show up on other engines with different weights. I use one standard so a Perplexity project does not become a dialect the other engines ignore. The frame for that is Machine Relations, and the names for citation, corroboration, and freshness are in the glossary. Perplexity at 30 million queries a day is large enough that absence is a choice I can see.

I would change my mind if a fixed query set, run monthly, showed no movement in citation rate after a quarter in which crawl access was open, the target pages had been rewritten to the passage standard, and at least one independent source corroborated the claims. That result would say the five inputs are not the constraint for our category, and I would go look at whether the category's answers are being built from a source class we cannot enter. I would also change the freshness rule if a page older than 90 days kept its citations across a full quarter of the audit. Until then, the 90 days is the rotation, and the passage is the unit of work.

FAQ

How fast can a new page get cited?

Perplexity crawls the live web, so a well-structured page on a current question can appear within days of publishing. The constraint I watch is not the index delay. It is whether the passage is extractable and whether anything else corroborates it. A page that is fast to crawl and empty of both will be fast to ignore.

Does Perplexity cite Reddit?

Yes, with disproportionate weight on opinion, comparison, and recommendation queries. An organic, upvoted presence in the relevant threads feeds the corroboration layer. Spam reverses it. I would not buy comments. I would notice when real users already describe the problem in our language.

How is this different from ranking in Google?

Google orders pages on a results page. Perplexity retrieves passages from the live web, reranks them, and cites a handful inline. A first-place Google ranking with a buried answer can produce zero Perplexity citations. I have stopped using the Google position as a proxy for this pipeline.

How often do I refresh?

Every 90 days on pillar pages, faster in AI, fintech, and software, where I treat the half-life as about 90 days. A page that was the best source six months ago loses to a weaker page updated last week. I keep dateModified honest. A timestamp on an unchanged page is not a refresh.

Does blocking PerplexityBot affect Google?

No. It is a separate crawler. Blocking it removes the site from this pipeline and leaves Google alone. Unless there is a specific legal reason to refuse the crawler, the cost of the block is the entire citation path, and the other four inputs never get a chance to matter.