# How do I get cited by Perplexity: two named crawlers and one API deprecation on 27 September > Perplexity documents two named crawlers with different robots.txt behaviour, and its Sonar Chat Completions endpoint is deprecated with support ending on 27 September 2026 in favour of the Agent API. - Canonical page: https://lantad.co/blog/how-do-i-get-cited-by-perplexity-two-crawlers - This file: https://lantad.co/blog/how-do-i-get-cited-by-perplexity-two-crawlers.md - Last substantive update: 2026-09-03 ## Key facts - **Published:** 2026-09-03 - **Category:** Guides - **Author:** Lantad - **Length:** 3110 words - **Takeaway 1:** Perplexity's bots documentation, read on 3 September 2026 at docs.perplexity.ai/guides/bots, names two user agents: PerplexityBot, which the guide states respects robots.txt rules, and Perplexity-User, which the guide states generally does not apply robots.txt disallow rules because it acts on a direct user request. - **Takeaway 2:** How do I get cited by Perplexity begins at retrieval and ends at a ranking Perplexity has not published: the Search API reference at docs.perplexity.ai/api-reference/search-post lists five documented fields returned per result, being title, url, snippet, date and last_updated, with no ranking factor or authority signal disclosed. - **Takeaway 3:** The Sonar Chat Completions endpoint carries a deprecation notice, read on 3 September 2026 at docs.perplexity.ai/api-reference/chat-completions-post, stating that Sonar Chat Completions is now Agent API and that Sonar Chat Completions will be supported until 27 September 2026, leaving 24 calendar days. - **Takeaway 4:** Perplexity publishes IP ranges for both agents at perplexity.com/perplexitybot.json and perplexity.com/perplexity-user.json, so an operator can verify a fetch by IP range rather than trusting the user agent header alone. - **Takeaway 5:** Read out of this repository on 3 September 2026, Lantad's answer engine registry names 8 engines, of which Perplexity is one, and the paid Perplexity row is Sonar based, so a downstream tool holding a Sonar Chat Completions integration must have migrated to the Agent API before 27 September 2026 or its citation counts stop returning. ## Summary How do I get cited by Perplexity is one of the searches that arrives at this scanner from Google, and the honest answer sits inside three of Perplexity's own documentation pages rather than inside any measurement this site has taken. Perplexity is an [answer engine](https://lantad.co/glossary/aeo): it responds to a query with a synthesised answer and a numbered list of source links, and a page can be one of those sources only if Perplexity can reach it, fetch it, and read the words on it. Two named crawlers do the reaching, one API family does the answering, and none of what sits between them is a ranking factor that Perplexity publishes. Two things about the middle of that pipeline changed during 2026 and are the reason to write about it now. Perplexity's Sonar Chat Completions endpoint, the API a developer called to get a grounded answer with citations, is documented as deprecated with support ending on 27 September 2026, replaced by the Agent API. And Perplexity documents both of its crawlers in one place at [its bots guide](https://docs.perplexity.ai/guides/bots), with published user agent strings and published IP ranges, so an operator can decide what to allow and can verify what arrived rather than trust the header. This post walks through what those documents say, where Lantad can measure something adjacent, and where the honest answer is that Perplexity has not published the mechanism. What it does not do is promise a citation. The site's own [how-to-get-cited reference page for Perplexity](https://lantad.co/how-to-get-cited/perplexity) is the standing guide next to which this post sits. ## How do I get cited by Perplexity How do I get cited by Perplexity starts with a mechanical prerequisite and ends with a behaviour Perplexity does not fully describe. A page enters the pool of candidate sources only if a Perplexity fetcher can reach it, receive a 200 response, parse the HTML that comes back, and find the words on it. That set of prerequisites is the shape [Lantad measures](https://lantad.co/methodology) directly, because it is the same shape every other AI answer engine imposes: [ChatGPT](https://lantad.co/how-to-get-cited/chatgpt), Claude and Google AI Overviews each begin from a fetch, and where the retrieval fails no downstream ranking rescues the page. Once a page is in that pool, what Perplexity does with it is documented at the surface and not documented at the mechanism. Perplexity's [Search API reference](https://docs.perplexity.ai/api-reference/search-post) describes exactly five fields returned per result: title, url, snippet, date and last_updated. That is a shape the site can be presented in, not a formula for winning a slot inside it. The documentation makes no statement about which of those five fields is weighted, whether a source's authority label affects ranking, or how the snippet is chosen from a page. There is nothing here on link graphs, on domain age, on freshness. Where a person searching how do I get cited by Perplexity is expecting a recipe, Perplexity is instead publishing a contract. The rest of this post reads what Perplexity has published, sets it against what Lantad can and cannot measure from outside, and marks where the honest answer is not yet available. That marking matters: a scanner that sells measurement has to name the parts it does not measure, which is why the [LantadBot page](https://lantad.co/bot) publishes what our own crawler does before anything about what other crawlers do. ## PerplexityBot follows robots.txt, Perplexity-User generally does not The difference between the two agents is not a technicality and it is the single line of the guide that operators most often miss. PerplexityBot is a search crawler, so its behaviour fits the ordinary [AI crawler](https://lantad.co/glossary/ai-crawler) mental model: a Disallow rule against it in the file at that authority will stop it from fetching a URL under that authority. Perplexity-User is the fetcher invoked when a person is using Perplexity and the answer needs a page Perplexity has not seen. Perplexity's guide states that agent generally does not apply a Disallow rule, because the fetch is on the direct request of an actual user. Both statements sit in the same paragraph of the guide, so neither is an inference this post is drawing. Two consequences follow. The first is that a robots.txt line disallowing PerplexityBot is not a block on being consulted; it is a block on being surfaced in Perplexity's own search index, and the same page can still be pulled in on demand under Perplexity-User. The second is that a per-token count in a log tells a smaller story than most dashboards present it as: a request labelled Perplexity-User is one the operator probably intended to receive, and a request labelled PerplexityBot is one they can decline. Both are subject to the general point that [a user agent is a claim, not an identity](https://lantad.co/blog/a-user-agent-is-a-claim-not-an-identity), which Perplexity's own IP publication is what makes verifiable at all. Perplexity is not alone in the split. OpenAI [documents GPTBot and ChatGPT-User separately](https://platform.openai.com/docs/bots), and Anthropic [separates its training crawler from its on-demand fetcher](https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler). The general lesson is that treating AI crawler as one thing loses the information that decides whether a rule you write applies to the request you meant it to. Lantad's own registry ships fifteen tokens across the nine vendors that publish them, one row per token, and each token is scored separately for that reason. ## The Sonar family, and what turns a fetched page into a cited source The crawler is not the surface that produces a citation. That surface is the Sonar family of APIs, which is where a call sends a query and receives back an answer with sources, and where a page's status as fetched becomes its status as cited. Perplexity's documentation names four Sonar models: Sonar, described as a lightweight search model with grounding; Sonar Pro, an advanced search model with grounding supporting complex queries and follow-ups; Sonar Reasoning Pro, a precise reasoning model with Chain of Thought; and Sonar Deep Research, an expert-level research model conducting exhaustive searches. Grounding is Perplexity's own word for the retrieval step in which documents are pulled in and cited alongside the answer. Underneath those models sit two API shapes that a developer chooses between. The [Search API](https://docs.perplexity.ai/api-reference/search-post) returns raw ranked web results with five fields per result and no synthesised prose. The Agent API, which Perplexity's documentation says is now the recommended way to build agents that need grounded citations, returns an answer together with the sources used to compose it. Neither API's public reference discloses ranking factors, and the Search API reference is explicit that its five returned fields are the ones named above. There is no field for authority label, no field for a rank score, and no explanation of the ordering: the response says what came back, not why. That is not a criticism of Perplexity. It is the state of the information available to a publisher, and it decides which of the two answer engine optimisation tactics is worth spending time on. The set of things a page can control end to end is small: whether it fetches cleanly, whether its [prose parity](https://lantad.co/glossary/prose-parity) is high enough that the crawler reads the words a person sees, whether its structured data labels the entity correctly, whether the page is fresh. The set of things a page cannot control from outside is what the model then does with a hundred candidate documents. The honest posts about that gap sit under the [research page](https://lantad.co/research), and the general observation is that [three of the eight engines this scanner watches cannot search the web at all](https://lantad.co/blog/ai-visibility-tracking-three-of-eight-engines-cannot-search), so the citation surface differs sharply by which engine a person is asking. ## Sonar Chat Completions retires on 27 September 2026 The Sonar Chat Completions endpoint was the interface a developer historically called to pass a chat-shaped request to a Sonar model and receive an answer back with cited sources. On 3 September 2026 the deprecation notice on that endpoint's documentation states that Sonar Chat Completions is now Agent API, and that Sonar Chat Completions will be supported until 27 September 2026. That leaves 24 calendar days between this post's date and the deprecation date, and the migration document Perplexity links from the notice is where the mechanical details of moving a caller to the Agent API live. The change is a surface change, not a mechanism change: the crawlers are the same, the models are the same, and the pool of pages a source can be picked from is the same. It matters to a publisher for two reasons. The first is that a third-party integration or plugin that was quietly calling the old endpoint, without a maintenance schedule set against it, is going to stop returning citations sometime around 27 September without any signal to the plugin's users. The second is that a dashboard or reporting tool built on Sonar Chat Completions, of which there are [several sold as AI visibility tools](https://lantad.co/blog/best-ai-visibility-tools-two-machines-one-category), will need its provider to have shipped an Agent API migration before that date. There is no way for a customer of such a tool to check the state of that migration from outside. The observation applies to this site too. Lantad ships an [engine registry](https://lantad.co/glossary/ai-visibility) held as constants in this repository, and the paid Perplexity row it wires is Sonar based. The registry is not immune to a vendor deprecation, and the honest position under such a change is the one the site already has: state which API is called on which date, and update the row when the vendor moves. Adjacent to the crawlers themselves, a body of licensing arrangements has grown up between Perplexity and publishers, [as recorded on 1 September 2026](https://lantad.co/blog/ai-publisher-licensing-deals-10-2-citations-against-6-9). Those deals do not change the crawler or the API. What they change is the shape of the pool that Sonar's grounding step reaches for, and that shape has been moving through 2026. Perplexity's own bots guide remains the single-page reference for the fetcher end of the pipeline, and a publisher's first move is what it was in July: keep the pages fetchable, allow the crawler you intend to allow, and check the log. ## What to check on your own site before then None of the above is a citation recipe. Perplexity does not publish one, and the honest version of what to check is a set of things a publisher can verify from outside using what the vendor has actually written down. The list below is worth walking through on the same day this post is read, because the checks are cheap and the state of the answers is a moving target. Start with the fetch. Every page a publisher wants cited has to answer both PerplexityBot and Perplexity-User, and answer them with the same words a human visitor would see. That is the same prose parity requirement every other AI crawler imposes, and the [what GPTBot sees tool](https://lantad.co/tools/what-gptbot-sees) exists to demonstrate the shape of failure when real content only appears after JavaScript runs. Framework stacks with a client-side render path have their own guides on the same problem: the [Next.js fix page](https://lantad.co/fix/nextjs) covers the specific misconfigurations that leave a page's body empty in the raw HTML a crawler receives, and the React fix page next to it covers the same pattern for that stack. Then the robots. Confirm the [robots.txt tester](https://lantad.co/tools/robots-txt-tester) reports the outcome you want for PerplexityBot on the exact pages that matter, remembering that the file is scoped per authority and that a rule at the apex does not govern a subdomain. Then set the log-side alert: for a subdomain you did not know about, a Perplexity-User request is often the first sign that a person asked the assistant about a page you host, and its absence is the first sign no person has. Finally the entity signal. Perplexity's Search API returns a title and a snippet, and both are extracted from the page as served, so the shape a page presents itself in matters. Structured data is not documented as a ranking factor. It is documented nowhere in the Search API response, and Perplexity's public references say nothing about it. What it does help with is entity attribution, which is a different job from getting the page picked in the first place. The [GEO vs SEO post](https://lantad.co/blog/geo-vs-seo-same-crawl-four-differences) sets out the four operational differences after the shared crawl, and the honest answer to what changes for citation is smaller than most vendor posts claim. ## Questions and answers **How do I get cited by Perplexity?** Perplexity does not publish a ranking recipe. What its documentation supports is a set of prerequisites: the page must return 200 to a Perplexity fetcher, its words must appear in the HTML the crawler receives, and the crawler you want to allow must not be blocked in robots.txt. PerplexityBot follows robots.txt; Perplexity-User, its documentation says, generally does not apply Disallow rules because it is acting on a direct user request. Beyond those prerequisites, the Search API returns five fields per result and discloses no ranking factor. **What are PerplexityBot and Perplexity-User, and how do they differ?** They are the two user agent tokens Perplexity publishes at docs.perplexity.ai/guides/bots. PerplexityBot is the search crawler and respects robots.txt rules. Perplexity-User is invoked when a person is using Perplexity and the answer needs a page not already in the index, and Perplexity's documentation states this agent generally ignores robots.txt Disallow rules. Both carry published IP ranges at perplexity.com/perplexitybot.json and perplexity.com/perplexity-user.json, which lets an operator verify a fetch by range rather than trusting the user agent header. **When does Sonar Chat Completions stop working?** On 27 September 2026, per the deprecation notice on the Perplexity Sonar Chat Completions API reference read on 3 September 2026. The notice states that Sonar Chat Completions is now Agent API and that Sonar Chat Completions will be supported until that date. A migration guide accompanies the notice. Any product calling the old endpoint has to move to the Agent API before that date to continue receiving grounded answers with citations. **Does adding schema get me cited by Perplexity?** There is no statement from Perplexity that schema is a ranking factor. Its Search API reference exposes five fields per result and none of them is a schema signal. The honest reading is that structured data helps an answer engine attribute a page to the right entity, which is a different job from getting the page picked in the first place. This scanner treats schema as a discrimination signal among cited pages, which is stated on the methodology page. --- Lantad measures whether AI crawlers can actually read a page: it fetches as a non-rendering crawler, renders as a browser, and reports the gap. Free scan, one URL, no signup. Method and weights: https://lantad.co/methodology | All pages as markdown: https://lantad.co/md | Crawler policy: https://lantad.co/bot