How Perplexity Chooses Its Sources: What Is Documented and What Is Not

What Perplexity actually documents

Perplexity publishes documentation for its crawlers, and that is the most reliable information available about how it finds pages. Its bot documentation lists two agents:

• PerplexityBot is used "to surface and link websites in search results on Perplexity." This is the agent that decides whether your pages are in the pool Perplexity can cite at all.
• Perplexity-User supports "user actions within Perplexity": when someone asks a question, it may visit a page "to help provide an accurate answer." Perplexity says this agent "generally ignores robots.txt rules," because the fetch is initiated by a person.

Two consequences follow. If PerplexityBot is blocked in your robots.txt, you are asking Perplexity not to surface your site. And robots.txt is not a complete off switch, because live fetches for user questions are treated differently. Our AI crawler checker shows what your robots.txt currently says to both.

What the documentation does not describe is how candidate pages are scored, how many are considered, or how the final citations are chosen. Perplexity has not published that.

What has been measured

The most useful public data on what Perplexity cites comes from outside the company. In Profound's analysis of citation patterns, published in June 2025, Reddit accounted for 46.7% of Perplexity's top-10 cited sources.

The same analysis found the engines differ sharply:

Two cautions. This is one company's analysis of top sources at a point in time, and citation patterns move as engines change. But the direction is clear enough to act on: for questions about your category, Perplexity's answer is likely to draw on community discussion alongside, or instead of, your own pages.

What the research says applies to every engine

Perplexity writes answers from documents it retrieves. That design has a well-known research lineage. Lewis et al. (2020) described models that "combine pre-trained parametric and non-parametric memory for language generation," and found they "generate more specific, diverse and factual language" than a model working from its parameters alone. A system built that way can only be as specific as the passages it retrieves.

That is why the evidence from the Princeton GEO study carries over. Across generative engines, the strongest content interventions were adding quotations, adding statistics and citing sources, and citing sources raised the visibility of a site ranked fifth in search results by 115%. Specific, sourced passages give a retrieval system something worth lifting. The full set of interventions is in How to Get Cited in ChatGPT, Perplexity, and Google AI Overviews.

What is not documented, and how to treat it

Search for how Perplexity ranks sources and you will find confident descriptions of multi-stage pipelines, exact numbers of candidate pages, and freshness windows measured in days. None of the ones we checked cited Perplexity for those details.

Some may be reasonable inferences from testing. But a claim without a source is a hypothesis, and the sensible way to use it is to test it on your own prompts rather than reorganise a content programme around it. That is the same standard we apply to any statistic in a blog post, including this one.

What to do

1. Allow PerplexityBot. If you want to be cited, let the agent that surfaces websites reach the pages you want surfaced.
2. Be the complete answer. Pages that define the topic, explain how it works, give examples and address the obvious objections give a retrieval system more to use than a thin page with a strong headline.
3. Source every number. Link each statistic to its primary source on first mention. This is the intervention with the strongest published evidence.
4. Earn presence where Perplexity looks. Reddit is where much of its citation share goes. Take part as practitioners, not as a brand account, and seek coverage in the publications your category reads.
5. Keep time-sensitive pages current. An engine that retrieves when it answers compares your page with whatever else it finds. Update the facts, then update the date.
6. Measure across many runs. Answers vary. SparkToro's research found ChatGPT, Claude and Google's AI returned the same list of brands less than once in a hundred runs. There is no reason to assume Perplexity is steadier, so track how often you appear across a set of prompts over weeks.

Measuring it

A useful baseline is 30 to 50 questions a buyer in your category would ask, run on a schedule, recording which domains appear as sources. The metric that matters is how often your domain appears across the set, not where it sits in any single answer. AI visibility tracking in Veritas runs this across Perplexity and four other engines, and our comparison of AI visibility tools covers dedicated alternatives.

The short version: Perplexity documents how it reaches pages, independent data shows what it tends to cite, and research shows what kind of content gets lifted. Everything else is a guess worth testing.