How to rank in Perplexity in 2026

The mistake is treating Perplexity like a results page with positions to improve. You check an answer, see a competitor named near the top, then search for a ranking signal that explains why your page sits below it. The cost is wasted work. You optimise for a position Perplexity does not publish, and miss the useful questions: did it retrieve your page, cite it, or name your product?

There is no stable ranking to claim. Perplexity retrieves pages, chooses sources to support an answer, and decides which products or companies to mention. These events overlap, but they are not interchangeable. Your work should improve the chance that relevant pages enter the retrieved set and give the model clear facts it can use.

How do you rank in Perplexity in 2026?

Start by separating the language people use from the thing you can measure. “Ranking” suggests a fixed order, a position and a route from position eight to position three. Perplexity answers are assembled from retrieved sources, and the visible answer can name a product without making that product’s own page the source. A competitor may be named because a roundup compares it, but the competitor’s site can be absent from the citations.

Three terms matter when you review an answer. Retrieval means a page was selected as a possible source. Citation means the answer linked to that source. Naming means the model mentioned the product or company in its prose. A page can be retrieved and cited without its product being named. A product can be named because another domain described it. Keep these steps separate in your reporting, because each points to a different action.

The available scan data shows why this matters. Perplexity answers in the sample cited 5.3 sources per answer on average, and the scanned product was named in 8.2% of answers. Gemini cited 4 sources and named the scanned product in 6.8% of answers. ChatGPT family answers cited 3.4 sources and named the scanned product in 4.4%, but Claude cited 4.8 sources and named it in 6%.

Those naming rates are observations across samples, not positions. CiteGraph sampled 1,787 Perplexity answers, 1,784 Gemini answers, 1,781 ChatGPT family answers and 1,467 Claude answers, according to how CiteGraph measures. The sample tells you how often the scanned product was named in those answers. It does not show that a page holds a permanent place in Perplexity.

What does Perplexity choose when it answers your question?

Perplexity chooses sources for a question, then produces an answer from the material it has found. The source that earns a citation is not automatically the product the answer recommends. A buyer asking for software options may receive a roundup as a citation, see a product named in the text, and never visit the product’s own site. Reporting only citations hides the business outcome. Reporting only mentions hides the page that supplied the evidence.

There is no published ledger of official factors that produces a public rank. You can still inspect conditions that affect inclusion: whether the page is accessible, whether it answers the buyer’s wording directly, whether other pages describe the product consistently, and whether the page fits the source types Perplexity already uses. These are working hypotheses, not a published scoring formula.

Perplexity’s own documentation makes access relevant. It says, “PerplexityBot is designed to surface and link websites in search results on Perplexity.” The same page adds, “It is not used to crawl content for AI foundation models.” That is a narrower claim than a promise that every page is stored and ranked. It supports checking whether the retrieval path is open, not assuming inclusion.

Use the retrieval, citation and naming split from above when a page appears in a citation but the product does not appear in the answer. The source entered the answer’s context, but another source or product won the naming decision. Fix access when the page is absent, improve its usefulness when it is cited but ignored, and investigate third party descriptions when the product is named from elsewhere.

Which domains should you study to understand Perplexity retrieval?

The broad source mix points away from a site-only strategy. In CiteGraph’s recorded citations, listicles and roundups account for 67.5%, with 30,585 citations. Competitor product pages account for 24.4%, with 11,052 citations. Together, those page types take 91.9% of citations. Every other page type shares what remains.

The domain view gives you a sharper starting point. These are the five most-cited domains in the available series:

Otterly.ai was cited 285 times across 1 category, Medium 267 times across 31, TikTok 230 times across 5, G2 203 times across 30 and YouTube 201 times across 31. The ten most-cited domains together account for 9.1% of all citations, so even the leading domains do not own the whole system.

CiteGraph recorded 885 domains with at least ten citations and 84 with at least fifty. That is a long tail, not a small club with a registration form. Use the weekly research on who owns AI answers to find patterns in your category, then inspect the exact pages instead of treating domain totals as a universal ranking.

Which page types should you build to enter the retrieved set?

Build pages around the questions that cause a product to be selected, compared or rejected. A product page can state what the product does, but a comparison page can explain where it fits and where it does not. A use case page can connect the product to a buyer’s situation. An alternatives page can give the model language for distinguishing your product from another one.

Make the relevant claim easy to find and safe to repeat. Put decisive facts in normal prose. Keep names, capabilities, limits and audiences consistent. If a page needs a reader to infer the answer from slogans, navigation labels or a table rendered only in a script, it gives retrieval less useful material to work with.

Your off-site plan needs the same discipline. Review how roundups and competitor pages describe you, then look for factual gaps or contradictions. A third party page may be the route by which Perplexity names your product. That route can matter even when your own page receives no citation.

Community threads account for 3.4% of recorded citations, and the scanned product’s own site accounts for 1.5%. Documentation accounts for 1.1%. Social posts and review platforms each account for 0.8%, YouTube accounts for 0.4%, news accounts for 0.1%, and Wikipedia accounts for 0% in the recorded data, despite 21 citations. Your own site still needs clear product facts. The distribution shows that third party context carries more of the observed citation load.

How do you make your pages available to Perplexity?

Check robots.txt and server logs before rewriting the page. Perplexity’s documentation recommends allowing PerplexityBot in robots.txt and permitting requests from its published IP ranges so that a site can appear in search results. The same page publishes the full user agent string as Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; PerplexityBot/1.0; +https://perplexity.ai/perplexitybot). Use the current details from the Perplexity bot documentation, not a copied rule from an old audit.

The two bots have different jobs. PerplexityBot crawls and indexes for retrieval, and respects robots.txt. Perplexity-User is a separate, on-demand fetcher triggered by a live question, and it generally ignores robots.txt because a person asked for that specific page. A blocked PerplexityBot is therefore a meaningful problem to investigate. Allowing it makes retrieval possible, but it does not guarantee that Perplexity-User or the indexed system will retrieve the page for every question.

Make the page stable and specific once access is sound. Give each page a clear job. Put the answer near the language a buyer uses. Keep the product name, category and key capabilities explicit. Avoid making the model assemble a basic fact from several unrelated pages when one page could state it plainly.

Then test more than your own domain. Compare a sample of answers for product naming, cited URLs and source type. Record whether your page was cited, whether another page supplied the product description, and whether the answer used your preferred positioning. That record separates a retrieval issue from a naming issue, which is more useful than a single manual search.

How do you measure inclusion instead of rankings?

Set up a query set that reflects the decisions you want to influence. Include category questions, comparison questions, alternatives and use cases. Run the same set over time and save the answer, citations and named products. Track repeated inclusion in the retrieved set and repeated naming in relevant answers, not a position that can change between answers.

Read the cited pages as carefully as the answers. If listicles and roundups dominate your category, track which pages appear and how they frame the choice. If a competitor page is cited but your own site is not, inspect whether the competitor page contains a clearer comparison, stronger evidence or the wording the query calls for. If your site is cited but your product is absent, the retrieval problem may already be solved.

A domain leaderboard can show who receives citations, but it cannot explain every naming decision. The concentration numbers from earlier, a thin top ten and a long tail of 885 domains, mean a leaderboard is a map, not a scoreboard. Use the leaderboards as a map of observed sources, then inspect the category and page behind the total.

Our data has a hard boundary. We measure samples and publish their counts; we do not know what we have not measured. The figures describe CiteGraph scans, not every Perplexity answer and not a promise about your next query. Treat them as evidence for where to look, then validate your own pages and queries.

The opening mistake was searching for a position. Replace it with a short audit: can Perplexity retrieve the page, does it cite the page, and does the model name the product? In the recorded Perplexity sample, the scanned product was named in 8.2% of answers and answers cited 5.3 sources on average. What has changed is the question you ask of that number. It records inclusion and naming, not a rank you can climb.