HOW PERPLEXİTY PİCKS SOURCES AND HOW TO GET CİTED

Perplexity is an answer engine that responds to a question with a live web search and visible citations, rather than listing links the way a classic search engine does. Being visible there takes two things: a technical door left open, and content written in a way that can be quoted. Coverage of the topic usually collapses into generic AI-visibility advice, yet Perplexity has two separate crawlers, its own published IP lists, and a way of working that differs structurally from ChatGPT and Google. Below we describe the mechanics from official documentation and label the evidence level behind each practice.

How Perplexity Works

When Perplexity receives a query it does not rank results from a pre-built index alone; it runs a live web search, reads the pages it finds, writes the answer with a language model, and places numbered source links beside the sentences. Nor does it settle for a single query: it searches variations of the question and merges the results. With the Search API announced in September 2025, the company described its own search infrastructure as an index covering hundreds of billions of web pages, which indicates the system is not a wrapper around third-party search results. On scale, the only verifiable company statement is this: founder Aravind Srinivas said in June 2025 that the platform had received 780 million queries the previous month and was growing more than 20 percent monthly. The 2026 user and query figures in circulation come from aggregator sites, contradict one another and rest on no audited company statement, so they do not appear here.

What Separates It From a Classic Search Engine

The difference concentrates in three places. First, output format: Google gives you ranked links, Perplexity gives you a written answer with links as footnotes. Second, click economics: because the user gets the answer in place, click-through falls. The return on visibility is brand recall and citation, not traffic. Third and most important, selection logic: Perplexity does not simply copy Google's top result.

In Ahrefs' August 2025 study across fifteen thousand long-tail prompts, only 12 percent of AI citations appeared in Google's top 10; by platform, Perplexity had the highest overlap at 28.6 percent while ChatGPT and Gemini fell into single digits. The same study attributes the gap to query fan-out: engines try multiple query variations and merge the results, choosing the page that ranks consistently across variations rather than the one ranked first for the original query. The practical consequence is that ranking first for one keyword guarantees nothing in Perplexity, while being consistently strong across a topic matters more.

PerplexityBot and Perplexity-User: Two Separate Crawlers

This is the most confused part of the technical side, and Perplexity's official crawler documentation draws the line clearly. PerplexityBot is the conventional crawler that fetches pages to build the search index; it obeys robots.txt and, in the documentation's words, is not used to crawl content for AI foundation models. Perplexity-User is a different thing: it activates when a user clicks a link or asks for a page to be fetched, and in the official wording, since a user requested the fetch, it generally ignores robots.txt rules.

The distinction matters in practice: blocking PerplexityBot in robots.txt keeps you out of the index, but Perplexity-User may still arrive if a user requests your page directly. Full user-agent strings for both are published in the official document, and IP lists for verification are served at separate JSON endpoints; the perplexitybot.json file contains the crawler's official IP prefixes. Prefixes for user-triggered fetches sit in perplexity-user.json. To protect against bots impersonating them, match on IP as well as user-agent.

Here is a detail almost nobody spells out: the two lists are independent and share no addresses. A check in August 2026 found 8 IPv4 prefixes in perplexitybot.json and 4 in perplexity-user.json. Each file carries its own creationTime stamp, and the stamps show different dates. Three consequences follow:

• Do not trust one list. Allowlisting only the perplexitybot.json prefixes leaves user-triggered fetches outside the gate.
• Do not hard-code the prefixes. The lists get updated, so pull the JSON on a schedule and feed your WAF rule from it.
• Log user-agent and IP together. Spoofed user-agents are common, and verification is impossible without an IP match.

Opening Access Correctly: robots.txt, WAF and Log Checks

The problem most often seen on sites is not robots.txt but the security layer. Bot-protection defaults in services such as Cloudflare, or security plugins from hosting providers, can block PerplexityBot without anyone noticing.

The correct order is this: first check your server access logs to see whether PerplexityBot is actually arriving, then edit robots.txt. Perplexity notes that robots.txt changes can take up to 24 hours to take effect and provides Cloudflare and AWS WAF configuration examples for those running a WAF. If you want to allow access, write an explicit rule for PerplexityBot; if you want to block it, do that in the same place, because writing no rule creates ambiguity. We covered general crawling and indexing logic in our technical SEO guide; the difference here is that you must consciously manage AI crawlers alongside classic search bots. Whatever you decide, decide it deliberately: accidental blocking is the quietest way to become invisible.

So can Perplexity-User be blocked? Not through robots.txt, because the documentation says that fetcher runs on a user request and generally ignores robots rules. The only route is rejecting the perplexity-user.json prefixes at your server or WAF layer. Weigh one thing first: when a reader asks for your page by hand, the request usually comes from a real person. Ten minutes today is enough to search your access logs for "PerplexityBot", and you should not touch robots.txt before you see that result.

The Cloudflare Dispute: Does Blocking Actually Work?

There is a contested side to this that deserves an honest account. In an analysis published in August 2025, Cloudflare accused Perplexity of using stealth, undeclared crawlers to evade no-crawl directives. The method was this: new test domains were created that prohibited all automated access in robots.txt, and Perplexity still summarised the content on those domains accurately. Cloudflare shared the traffic it measured, removed Perplexity from its verified bot program and added detection signatures to its managed rules. Perplexity rejected the claims.

Reading the dispute, most summaries skip one distinction: a site owner faces not a single Perplexity traffic stream but three separate layers. PerplexityBot obeys robots.txt, Perplexity-User is documented not to, and the undeclared fetches Cloudflare alleges sit outside both. A single robots.txt line binds only the first layer; the second needs an IP rule and the third needs behaviour-based detection. Using one tool for three jobs explains much of why blocking decisions fail to deliver what site owners expect.

For a site owner the takeaway runs both ways: first, if you decide to block, know that it may not produce a technically airtight result; second, both blocking and allowing are strategic choices rather than technical reflexes. If you want your content to appear in answer engines, blocking works against your goal; if unauthorised use troubles you, factor in that a single robots.txt line will not fully protect you.

Citation Mechanics: How Perplexity Picks Sources

Perplexity matches sources at the sentence level, quoting the passage that answers the question rather than the page as a whole. The most practical consequence is that having a good page overall is not enough; individual sections must be understandable on their own. Cited content typically contains clear definitional sentences, numeric data, dated information and step lists, because the engine prefers self-contained fragments that need no surrounding context.

Freshness is a separate criterion, since the system searches live and pages carrying dates signal recency. Source diversity is respected, so no single domain is expected to fill an entire answer; in niche topics with few quality sources, the visibility threshold drops. That is exactly where the opportunity lies for Turkish content: because the Turkish source pool is thinner than the English one, a well-structured Turkish page has an easier path to being cited.

Structural Differences Between Perplexity, ChatGPT and AI Overviews

All three produce answers, but their relationship with sources differs. Perplexity searches live on every answer by design and numbers its sources visibly; searching is its default behaviour. In ChatGPT, searching is conditional: the model goes to the web when it judges it necessary or when the user asks, and otherwise answers from training data; we covered how to show up in ChatGPT in a separate guide. Google AI Overviews sits on top of classic search infrastructure, so Google's own index and ranking systems are decisive, and no special markup is required.

The Ahrefs data quantifies the difference: overlap with Google's top 10 is 28.6 percent for Perplexity while other platforms fall into single digits. The strategic conclusion is that rather than seeking separate tricks for each, you should do what all three expect in common: clear structure, verifiable information and a strong entity identity. We covered the general framework in our AEO guide, the shared logic of generative engines in our GEO guide, and the trust side in our E-E-A-T article.

Pro Search, Deep Research and the Mode Difference

Perplexity has no single answering behaviour; the mode in use changes how many sources are scanned and how deeply. In standard search the engine consults a limited set of sources and answers quickly. Pro Search tries more query variations and scans a broader source set. Deep Research runs multi-step investigation, reading dozens of sources and returning a report-style output.

For visibility this means long-tail and niche sources have a markedly better chance of being cited in the deeper modes, because the engine does not settle for the best-known sites. Take that into account when testing your own content: not appearing in standard search does not mean you will not appear in Deep Research. Note which mode you used, otherwise the results you compare will be inconsistent with one another.

The Publishers Program and Revenue Sharing

Perplexity runs a program that shares revenue with the publishers it cites. The Comet Plus subscription is positioned at 5 dollars a month, is included for existing Pro and Max subscribers, and 80 percent of subscription revenue is distributed to publishers with 20 percent retained by the company. Figures circulating about the number of publisher partners and the size of the pool have not been confirmed by the company, so they are not repeated here. The program's significance may be limited for small publishers, since participation conditions and scale thresholds matter; still, it is directionally important because it proposes a model for how answer engines will relate to content producers. If you publish from Turkey, verify participation terms directly with Perplexity and disregard campaign claims on intermediary sites.

Evidence-Based Practices for Getting Cited

Because you are working with a system that does not rank, it helps to separate tactics by evidence level. The safe ones, derived directly from the mechanics:

• Access. The page is open to PerplexityBot and the WAF rules are clean.
• Self-contained sections. The text under each heading still reads when lifted out of context.
• A clear answer sentence. The question is answered in the first line.
• Numbers and dates. Quotable data, carried with its source.
• Freshness. A visible date with a real revision behind it.

For structural clarity, question-form headings and short definitional paragraphs work; we detailed content writing for AEO in a separate guide. On structured data, honesty is required: schema markup is valuable for search engines, but there is no official statement that Perplexity uses schema in ranking, so selling it as a Perplexity tactic would be wrong. The same applies to llms.txt: Perplexity publishes one for its own documentation, yet nowhere states that it uses third-party sites' llms.txt as a visibility signal. Entity clarity is an indirect but powerful lever, because once the engine recognises the brand, the chance of being mentioned in the right context rises.

Measuring Visibility: Referral Traffic and Manual Testing

There is no ready-made console for Perplexity visibility, so measurement rests on two legs. The first is analytics: visits from Perplexity appear as referrals, and you can track them separately by defining a channel group by domain in your analytics tool. Do not let low traffic mislead you, because part of the value in answer engines materialises as brand recall rather than clicks.

The second leg is manual testing: ask your target questions in Perplexity at regular intervals and record whether you are cited and, if so, in what position among the sources. Run the test in the same mode and, where possible, the same language, because changing mode or language changes the result set. Always record the date, since the system searches live and results can shift within a day. Tying measurement to a monthly rhythm gives a far more meaningful trend than one-off checks. Where manual testing stops scaling, AEO tools take over; even so, a first month of testing by hand is the cheapest way to learn what a tool actually measures. Pick five target questions and write them into a table with the date.

Practical Notes for Turkish Content

A warning comes first: there is no reliable published data on Perplexity's user numbers, query volume or market share in Turkey, so treat content quoting such figures with suspicion. The concrete advantage lies elsewhere: because the Turkish source pool is thinner than the English one, the threshold for being cited on niche Turkish topics is lower. The engine answers Turkish queries in Turkish and cites Turkish sources, so well-structured Turkish content finds a direct response. One practical tactic is keeping the English equivalent alongside the Turkish name of a concept, which raises the chance of matching when the engine runs sub-queries in English. On the technical side, the highest-return task is checking whether your security layer blocks PerplexityBot, because that is the most common cause of invisibility. Finally, do not repeat unverified claims about free Perplexity campaigns through carriers or banks that circulate on intermediary sites.

Where do you start? In this order: check the access log for PerplexityBot, make the robots.txt and WAF call deliberately, rewrite the first sentence under every heading on your three most valuable pages so it stands alone as an answer, then test five target questions monthly. Four steps, one weekend. You cannot improve visibility you never measure, so take the first reading this week.