Coaley Peak GEO Research Centre Engine guide
Perplexity How the answer engine that cites everything chooses its sources.
Perplexity is the answer engine built around citations: every answer links its sources by number. It runs its own index and, since September 2026, its own ranking engine, while answering with models from OpenAI, Anthropic and others. This guide covers its timeline, the four layers behind its answers, its two crawlers and how to opt out.
Independent guide: Coaley Peak is not associated with Perplexity AI, Inc. Trademarks
- Owner
- Perplexity AI, Inc.
- Launched
- December 2022
- Home market
- Worldwide; no official country list
- Languages
- Not published
- Open weights
- No for the models it answers with
- Crawlers
- PerplexityBot and Perplexity-User
Overview
What Perplexity is, and who uses it
Perplexity AI was founded in August 2022 by Aravind Srinivas, Denis Yarats, Johnny Ho and Andy Konwinski, and launched its answer engine that December. In April 2023 it raised a US$26 million Series A and launched an iOS app. Its chief executive said it handled 780 million queries in May 2025; Perplexity does not publish monthly user numbers.
Perplexity does not build foundation models. It answers with third-party frontier models, such as OpenAI's GPT-5 and Anthropic's Claude Sonnet, chosen by a 'Best' setting or by Pro users, and with Sonar, its own model built on Meta's Llama 3.3 70B. What it does build is search: its own index, crawled by PerplexityBot, and since September 2026 Photon, an in-house retrieval and ranking engine.
It has grown beyond search. Comet, its AI browser, launched in July 2025 and became free that October, and Perplexity Computer, a multi-model agent, launched on the Max plan in February 2026. Publishers can share revenue through Comet Plus, launched in August 2025.
It has also been at the centre of disputes over crawling. Cloudflare accused it of undeclared crawling in August 2025, which Perplexity rejected; the BBC threatened legal action in June 2025, and the New York Times sued it in December 2025. For SEO teams the practical points are its documented crawlers, its published IP lists and the citations on every answer.
Timeline
Perplexity, from launch to today
Perplexity AI is founded by Aravind Srinivas, Denis Yarats, Johnny Ho and Andy Konwinski.
Perplexity launches its AI answer engine.
Perplexity raises a US$26 million Series A and launches an iOS app.
Perplexity launches its Publishers' Program, sharing ad revenue with TIME, Der Spiegel, Fortune and others.
Perplexity rolls out a new version of Sonar, its in-house model built on Llama 3.3 70B.
Perplexity's chief executive says it handled 780 million queries in May 2025.
The BBC threatens legal action, demanding that Perplexity stop using and delete BBC content.
Perplexity launches Comet, its AI-powered web browser.
Cloudflare accuses Perplexity of stealth crawling and removes it as a verified bot; Perplexity rejects the claims.
Perplexity announces Comet Plus, a US$5 subscription that shares revenue with publishers.
Perplexity makes the Comet browser free for everyone.
The New York Times sues Perplexity for copyright infringement.
Perplexity launches Perplexity Computer, a multi-model agent, on its Max plan.
Perplexity moves search retrieval and ranking to Photon, an engine it built in-house.
Things to know
Things to know about Perplexity
To appear in Perplexity, allow PerplexityBot in robots.txt and its published IP ranges in any firewall. Perplexity suggests rules that match both the user agent and the IP list.
Every Perplexity answer carries numbered citations linking to the original sources.
Perplexity says it does not build foundation models, and that content PerplexityBot crawls is not used for pre-training.
Disallowing PerplexityBot does not make a page vanish: Perplexity may still index its domain, headline and a brief factual summary.
Perplexity-User, which fetches pages for live questions, generally ignores robots.txt because a user requested the fetch.
Perplexity also builds its index with third-party crawler partners, which it says have agreed to respect robots.txt, particularly for news publishers.
Since September 2026, retrieval and ranking run on Photon, Perplexity's in-house engine, replacing an adapted open-source one.
Answers can come from different models: 'Best' picks one, and Pro users can choose others, such as GPT-5 or Claude Sonnet.
Factual errors can be flagged under the answer, or through a support ticket that includes the query URL.
Publishers can share revenue through Comet Plus, which allocates it across human visits, search citations and agent actions; the contact is publishers@perplexity.ai.
The four layers
How Perplexity builds an answer
Every AI answer is made in four layers: what the model learned in pre-training, how it was tuned, what it retrieves live, and how it puts the answer together. Here is each one for Perplexity, and what it means for a brand that wants to be cited.
Layer 1
Pre-training: the base (foundation) model
Perplexity says it does not build foundation models. Answers come from third-party frontier models, including OpenAI's GPT-5 and Anthropic's Claude Sonnet, chosen by its 'Best' setting or by Pro users, and from Sonar, built on Meta's Llama 3.3 70B. Knowledge cut-offs depend on the model in use. Perplexity says PerplexityBot's crawl is not used for pre-training. Free, Pro and Max users' data may be used for training unless they switch off 'AI data retention'.
What it means for you: The models' own knowledge matters less here than in almost any other assistant, because nearly every answer is built from live search. What those models learned elsewhere shapes how they phrase answers about your category, but retrieval decides who is cited.
Layer 2
Post-training: tuning and alignment
Sonar was trained further from Llama 3.3 70B for factual accuracy and readability. For Perplexity Computer, Perplexity post-trains open models with reinforcement learning in synthetic environments, then fine-tunes them on real user sessions. Enterprise customers have zero-retention, no-training agreements with the model providers.
What it means for you: Sonar is tuned for factuality, which favours sources that state facts plainly and consistently. Pages that hedge, bury their claims or contradict themselves are less useful to it.
Layer 3
Retrieval: live search
Perplexity searches the web in real time against its own index, crawled by PerplexityBot and supplemented by third-party crawler partners. Since September 2026, retrieval and ranking have run on Photon, its in-house engine. Perplexity-User fetches pages live for users' questions, and premium data sources, such as finance data, and connectors add non-web data.
What it means for you: PerplexityBot's index is the way in. Allow it, keep your pages fast and crawlable, and make sure each one answers specific questions clearly, because Photon chooses passages, not just pages.
Layer 4
The answer: what people read
Every answer includes numbered citations linking to the original sources. Modes include Pro Search with a model selector, Research, which runs dozens of searches and reads hundreds of sources in a few minutes, the creation of files and apps, Comet's browser assistant and Perplexity Computer. TechCrunch reported that Perplexity abandoned advertising in late 2025.
What it means for you: Citations are numbered and prominent, so being cited sends visits. Research mode reads hundreds of sources, which rewards depth: comprehensive, well-structured pages are more likely to be among the few it quotes.
Crawlers
Perplexity's crawlers and user agents
The user-agent tokens to look for in your server logs and to name in robots.txt.
| User-agent token | What it does | robots.txt |
|---|---|---|
| PerplexityBot Documentation | Surfaces and links websites in Perplexity's search results. Perplexity says it is not used to crawl content for AI foundation models. User agent: Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; PerplexityBot/1.0; +https://perplexity.ai/perplexitybot). | Follows robots.txt, though it may still index a disallowed site's domain, headline and a brief factual summary. Changes can take up to 24 hours. |
| Perplexity-User Documentation | Supports user actions: when a user asks a question, it may visit a page to answer and link to it. Not used for crawling or to collect training data. User agent: Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; Perplexity-User/1.0; +https://perplexity.ai/perplexity-user). | Generally ignores robots.txt, because a user requested the fetch. |
Perplexity also uses third-party crawler partners whose user agents it does not publish. It says its agreements with them require them to respect robots.txt, particularly for news publishers.
Crawl sources
Where Perplexity crawls from, and how to verify it
Perplexity publishes JSON IP lists for both its agents, perplexity.com/perplexitybot.json and perplexity.com/perplexity-user.json, which now redirect to the same files on perplexity.ai. It publishes no reverse DNS hostnames or networks.
Perplexity recommends firewall rules that match both the user agent and the published IP ranges. Treat a request with a Perplexity user agent from outside the lists as unverified.
In August 2025 Cloudflare reported crawling it attributed to Perplexity from addresses outside these lists, using a generic browser user agent. Perplexity disputed this, arguing that agents acting for users are not bots. Whatever you make of that dispute, the published lists remain the only official reference.
Opting out
How to opt out of Perplexity, or limit what it uses
Block PerplexityBot
User-agent: PerplexityBot with Disallow: / stops it indexing the full or partial text of your pages. Perplexity may still index the domain, headline and a brief factual summary, and changes can take up to 24 hours.
Official pageFirewall rules for Perplexity-User
Perplexity-User generally ignores robots.txt. Perplexity publishes its user agent and IP list, so a firewall rule matching both is the firm control.
Official pageTalk to Perplexity as a publisher
Perplexity shares revenue with publishers through Comet Plus. Publishers can contact publishers@perplexity.ai.
Official pagePersonal data requests
For access, correction, deletion, objection or restriction, use Perplexity's data request form or email support@perplexity.ai. VeraSafe acts as its EU and UK representative.
Official pageReport a wrong answer
Use the flag icon under the answer, or open a support ticket with the query URL and a description of the error.
Official pageKeep your own data out of training
Free, Pro and Max users can switch off 'AI data retention' in their account preferences. It applies from the date you opt out.
Official page
Blocking PerplexityBot keeps your page text out of Perplexity's index, but not necessarily your domain and headline, and it cannot stop user-requested fetches. Because Perplexity is built on citations, leaving it means giving up a reliable source of referral visits.
How we help
How we get brands cited in Perplexity
The Coaley Peak GEO Research Centre researches and implements GEO for SEO agencies and in-house teams, with foreign markets as our speciality.
Perplexity visibility audit
We run the questions your customers ask through Perplexity, across models and markets, and record whether you are cited, in what position, and who is cited instead.
Crawl access you can verify
We set robots.txt and firewall rules against Perplexity's published IP lists, so PerplexityBot can index you and impostors cannot.
Passage-level content
Photon ranks passages. We restructure key pages so each section answers one question completely, which is what earns a numbered citation.
Depth for Research mode
Research mode reads hundreds of sources. We build the comprehensive, well-sourced pages it prefers, so you are among the few it quotes.
Measurement
We track referral visits from Perplexity and Comet, and your citation share against competitors, market by market.
More on our research, services and team at the Coaley Peak GEO Research Centre.
FAQ
Perplexity GEO: questions
Does Perplexity use my content to train AI models?
Perplexity says it does not build foundation models and that content PerplexityBot crawls is not used for pre-training. Users' own data may be used unless they switch off AI data retention.
How do I get cited by Perplexity?
Allow PerplexityBot and its IP ranges, keep your pages fast and crawlable, and structure them so each section answers a specific question clearly. Every Perplexity answer carries numbered citations to its sources.
Does blocking PerplexityBot remove my site from Perplexity?
Not completely. Perplexity says it won't index the full or partial text of a site that disallows it, but may still index the domain, headline and a brief factual summary. Perplexity-User may still fetch pages users ask about.
Which models does Perplexity use?
Third-party models such as OpenAI's GPT-5 and Anthropic's Claude Sonnet, chosen by its 'Best' setting or by Pro users, and Sonar, its own model built on Meta's Llama 3.3 70B.
Where do Perplexity's crawlers come from?
Perplexity publishes JSON IP lists for PerplexityBot and Perplexity-User, and recommends firewall rules that match both the user agent and those ranges. It publishes no reverse DNS scheme.
Does Perplexity pay publishers?
Through Comet Plus, launched in August 2025, Perplexity shares revenue with participating publishers across human visits, search citations and agent actions. Publishers can contact publishers@perplexity.ai.
Want a free chat about Perplexity?
Tell us which market and brand you are working on. If we can help, we will set up a free, no-obligation chat with Stephen or Ethan.
More engine guides
Related engines
Independence and trademarks. Perplexity is a trademark of Perplexity AI, Inc. Other model and product names are trademarks of their respective owners. Coaley Peak is independent: we are not associated with, endorsed by or a partner of Perplexity AI, Inc. This guide is our own research from public documentation and reputable reporting, checked in October 2026.
Reviewed by Stephen Grindley, research centre lead. Last checked .
Document reference: ISO_webpage_geo_engine_perplexity_v1
Last modified: 7 October 2026
GEO Research Centre·Engine guides·Perplexity