Perplexity Ranking Factors: What the Evidence Says

Perplexity documents its crawlers. It does not document how it ranks or selects sources. Treat any list of "Perplexity ranking factors" with that gap in mind.

Here is what Perplexity states, as of October 2026:

• PerplexityBot crawls for search indexing. Perplexity says it is "not used to crawl content for AI foundation models."
• Perplexity-User fetches pages when a user's request triggers it. Perplexity says this fetcher "generally ignores robots.txt rules."
• On the developer side, web search on the Agent API is opt-in. Perplexity announced that the older Sonar Chat Completions endpoint would retire on 27 September 2026, so code that was not migrated can return answers with no web grounding at all.
• Perplexity cites whole sources by an integer index, not text spans.

What it does not publish: any weighting of content, freshness, authority or format. Nobody outside Perplexity can see which pages entered the model's context for a real user's query. So every factor below is graded by the strength of outside evidence, not by a vendor statement.

It helps to separate four stages. Parametric knowledge is what the model knows without the web. Access is whether a crawler can fetch your page. Retrieval is whether your page enters the context for a query. Generation is which retrieved pages get cited. You control access. You influence little else directly.

Access levers work by removing blocks. They don't make a page more likely to be chosen. They decide whether it can be chosen at all.

robots.txt permission for PerplexityBot

robots.txt rules are per bot. Blocking a training crawler such as GPTBot or ClaudeBot says nothing about PerplexityBot, and the reverse is also true. The common real-world failure is a blanket AI block written to stop training that also removes search eligibility. A minimal permission looks like this:

User-agent: PerplexityBot
Allow: /

Remember that robots.txt is necessary but not sufficient. Perplexity-User generally ignores it, so blocking is not a reliable way to stay out of answers either.

Network reachability

A firewall, CDN or bot-management rule can turn away a crawler that robots.txt allows. The usual symptom is a 403 Forbidden or a JavaScript challenge. Nothing in your analytics will flag it. Cloudflare changed its AI crawler defaults on 15 September 2026 for new domains, new sites and existing free-tier customers on pages that show ads, so a site can lose access without anyone editing a rule.

Server-side rendering

Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so they "can't read client-side rendered content." Independent re-testing as of June 2026 found PerplexityBot also does not render JavaScript. If your pricing or product facts only appear after hydration, Perplexity's crawler likely never sees them.

The fix is to put the facts that matter in the initial HTML response. A specific, quotable fact (a price, a number, a one-line definition) on a fetchable page gives the engine a sentence it can use. That is the only positive experimental signal for on-page content in the evidence.

Perplexity cites a small set of sources heavily. A 2025 study of news citations in AI search (arXiv 2507.05301) measured citation concentration with a Gini coefficient of 0.77 for Perplexity, against 0.83 for OpenAI models and 0.69 for Google. The same study found the top 20 news sources accounted for 67.3% of all citations for OpenAI models. Our digest of the news-citation study covers the method.

Read those numbers carefully:

• The data covers news sources only, from March to May 2025, on the AI Search Arena evaluation platform. It is over a year old and the population is self-selected.
• Concentration describes outcomes. It does not show that prominence causes citation, and nothing in the evidence lets you raise prominence on demand.
• Brand size is the dominant uncontrolled confounder in correlational GEO studies. Metrics like Domain Authority are vendor estimates built from backlinks and traffic, so they can't control for it. See correlation vs causation in AI search studies.

What this means in practice: when Perplexity cites other sites for your category, being named accurately on those sites is the route into the answer. That is an association in the evidence, not a promised effect.

Most GEO advice tells you to rewrite pages: add quotes, add statistics, restructure, simplify. The best-designed tests don't support it.

• C-SEO Bench (NeurIPS 2025) found "most current C-SEO methods are not only largely ineffective but also frequently have a negative impact on document ranking."
• A 252,000-trial factorial study found formatting-only edits had little impact. Topical relevance and list position were the biggest drivers of being cited first.
• A critical survey of 45 GEO studies concluded that already-retrieved content can change its citation, but no technique shows a stable, longitudinal, cross-platform effect on discoverability. It also warned that citation-oriented rewrites can impair retrieval.

All of these were measured on documents already inside the model's context. Position mattered most, and no publisher can set position. Topical relevance is the one content factor the studies agree on, and it means genuinely answering the question, not covering keyword variants. Work on predictable prose and citation points the same way: its effect came from whole-page LLM rewrites, not a style you can copy.

The "up to 40% more visibility" claim traces to the original GEO paper (Aggarwal et al., KDD 2024). Its main results came from an engine the authors built, with sources supplied in context.

The paper did run one test on Perplexity. The authors uploaded source text as file attachments, because "perplexity.ai does not allow the user to specify source URLs." That means Perplexity never had to find, crawl or retrieve the pages. The test measured how Perplexity used text it was handed.

Three more limits apply:

• The main metric, Position-Adjusted Word Count, counts words the intervention itself added, so it partly measures its own edits.
• The paper reports a 30 to 40% band across three methods, not the itemised figures often quoted.
• Keyword stuffing decreased the paper's own metric (17.8 against a 19.5 baseline).

So the paper tells you nothing about organic discovery on Perplexity. If you want the wider picture, see why GEO studies disagree.

Start with the gates you control, in this order:

1. Read your robots.txt and confirm PerplexityBot is allowed. Check that a training-bot block didn't catch it.
2. Request a page as PerplexityBot and confirm a 200 response, not a 403 or a challenge page. The AI Crawler Checker requests a page as each AI crawler and shows which ones the firewall turns away.
3. View your key pages with JavaScript disabled. Prices, product names and definitions should be in the raw HTML.
4. Check for accidental noindex, nosnippet or data-nosnippet inherited from a template.
5. Fix contradictions across your own pages, such as a price stated two ways.

Then look outward. Find which third-party sites Perplexity cites for your buyer questions and check whether they describe you correctly. For a B2B angle, see how SaaS companies get cited in ChatGPT and Perplexity. None of this guarantees a citation. It removes the reasons you can't be cited.