Perplexity AI – the answer engine that cites its sources: how it works and how to get your company described correctly

Definition

Perplexity is an AI-driven answer engine that searches the web in real time and writes a summary answer with source citations instead of showing a list of links. Founded in San Francisco in 2022, it is available as a free version, a Pro subscription, the Comet browser and an API for developers.

What is Perplexity AI and how does it work?

Perplexity answers with a coherent piece of text rather than a list of links, and shows where each piece of information comes from. Unlike a classic search engine, the answer comes first and the links serve as evidence – not the other way round. The format is usually called an answer engine, and the rest of this page builds on that distinction.

The company was founded in August 2022 by Aravind Srinivas (CEO, formerly at OpenAI and DeepMind), Denis Yarats, Johnny Ho and Andy Konwinski, and is headquartered in San Francisco (Wikipedia). Technically, Perplexity is a RAG system at scale: a search component retrieves relevant pages from its own index and from live search, a language model writes the answer based on that material, and each claim is linked to a source citation that the reader can click. The free version uses Perplexity's own models; in Pro, users can switch to third-party models from OpenAI and Anthropic, among others – the same model families that power ChatGPT and Claude.

The product family currently has four parts:

• Perplexity (free and Pro) – the web and app service for questions and research.
• Comet – a free Chromium-based browser with a built-in AI agent that can carry out tasks on web pages.
• API for developers – search plus a language model with citations, now via the Agent API. Support for the older Sonar API ended on 27 September 2026 (Perplexity Docs).
• Perplexity Shopping – product search with purchasing directly in the answer.

The business model combines freemium subscriptions, API revenue and commerce revenue from Shopping.

How does Perplexity differ from ChatGPT and Google AI Overviews?

The most important difference is the starting point: Perplexity always starts with a search and bases its answer on the pages it retrieves, whereas ChatGPT and Gemini are essentially conversational models that switch on web search when a question calls for it. AI Overviews, meanwhile, is built into Google's ordinary results page – nobody actively chooses to use it; it simply appears above the blue links.

That has three practical consequences for anyone marketing a company:

• Search intent. People actively choose to use Perplexity. Someone who opens it is more likely to be in the middle of research – comparisons, supplier choices, technical questions – which makes it particularly relevant for B2B.
• Reach. AI Overviews reaches the widest audience because it sits inside Google. Perplexity is smaller, but its users are more inclined to click on the sources.
• Citation. Perplexity shows its sources on every answer, which makes it easier to measure whether and how your company is mentioned.

For a broader look at the platforms, see What is AI search?; Google's version is covered in depth in AI Overviews and GEO.

How does Perplexity choose which sources to cite?

Perplexity chooses sources in two steps: first, candidate pages are retrieved through search and its own index; then the model decides which passages actually answer the question and cites them. Pages that give a clear, self-contained answer early on, with up-to-date information and a clear structure, therefore stand a better chance than long texts that bury the answer far down.

Several industry analyses suggest that community sources such as Reddit, reference works such as Wikipedia and the video platform YouTube are among the sources Perplexity cites most. That says something about what the engine trusts: first-hand voices and consensus sources. For a company, this means brand mentions in third-party sources – reviews, forum threads, industry media – carry as much weight as your own website. That is where the model forms its picture of you as an entity: who you are, what you do and how you differ from others with similar names.

How do I control PerplexityBot via robots.txt?

Perplexity documents its crawlers openly, and according to Perplexity's crawler documentation, PerplexityBot respects robots.txt rules. To allow indexing, you need do nothing at all; to block it, add a Disallow rule for the PerplexityBot user agent. Note that the documentation also describes a separate agent, Perplexity-User, which fetches a page when a user asks a question about it. That agent generally ignores robots.txt, because the fetch is initiated by a person. So blocking PerplexityBot does not cover every request.

For most companies, the default should be to allow PerplexityBot. Block it and you do not disappear from the service – you only disappear as a source. Questions about your company are then answered from what others have written, which rarely works in your favour.

What are Comet and Perplexity Shopping?

Comet is Perplexity's own browser, built on Chromium, with an AI agent that can read pages, fill in forms and carry out multi-step tasks for the user. Perplexity Shopping and the "Buy with Pro" feature launched in November 2024. Since May 2025, purchases have been completed via PayPal checkout with no commission charged to the merchant, and in November 2025 the feature was opened up to all users, not just Pro subscribers (PayPal Newsroom, Digital Commerce 360).

Together, Comet and Shopping point towards a world in which an AI agent, not a person, makes the comparison and the purchasing decision. For online retailers, that means product data, stock levels and prices have to be machine-readable and correct – otherwise the product is passed over without anyone ever having seen it. We explain what that takes in Agentic commerce: AI agents that shop for the customer.

How do you get cited by Perplexity?

You get cited by having the page that answers a specific question most clearly – a page that can also be retrieved and verified from the outside. In practice, that means:

• Let the crawler in. Check that robots.txt and your firewall do not block PerplexityBot.
• Answer first, explain later. Open each section with the direct answer. Headings phrased as questions help the model match sections to questions.
• Write the content in HTML, not in scripts. AI crawlers generally do not read content that only exists after JavaScript has run.
• Keep your fact pages fresh. Prices, features, integrations and contact details should be correct – those are the details Perplexity retrieves when someone asks "does X support Y?".
• Be visible beyond your own website too. Relevant forum threads, industry lists, reviews and an accurate Wikipedia presence, where justified, give the model more independent evidence.

This is the same groundwork as the rest of AEO, but Perplexity's visible citations make the results unusually easy to track: ask ten questions your potential customers typically ask and note which sources the answers draw on.

What controversies surround Perplexity?

Perplexity is controversial in three respects, all of them relevant to how a company should approach the service.

The crawler debate. On 4 August 2025, Cloudflare published an analysis claiming that Perplexity had used undeclared crawlers that rotate user agents, IP addresses and networks to get around robots.txt blocks. Cloudflare also removed Perplexity from its list of verified bots (Search Engine Land). Perplexity has disputed this account. Whoever is right, the episode shows that robots.txt is an agreement, not a technical lock.

The publisher lawsuits. News Corp sued Perplexity in October 2024 (Washington Times), Encyclopaedia Britannica and Merriam-Webster in September 2025 (Britannica), the New York Times in December 2025 (TechCrunch) and CNN in May 2026 (NPR). The core question is whether it is lawful to summarise copyrighted material in an answer that replaces a visit to the source. The outcome will affect how all answer engines may use web content.

Comet security. Security researchers at Brave showed that Comet's agent could be tricked by hidden instructions on web pages – a technique known as prompt injection (Brave). LayerX described a variant called "CometJacking" in which a single click could turn the agent against the user (LayerX), and new variants were reported as recently as March 2026 (The Hacker News). The problem is not unique to Comet – it applies to every browser with a built-in AI agent – but it is a reason not to let an agent stay logged in to sensitive accounts.

How do you get your company described correctly in Perplexity?

You get your company described correctly in Perplexity by owning the fact pages the engine retrieves its information from – what is usually called owned AEO. It comes down to four things:

• One clear page per offering.
• Up-to-date information on prices, features and integrations.
• Consistent company information on your website and in external registers.
• A robots.txt that lets PerplexityBot in.

When the information is missing, an answer engine tends to answer "no" rather than "don't know" to questions about what you offer. The simplest way to check where you stand is to ask the questions yourself in Perplexity, in the language your customers use, and see which sources the answer draws on.