Perplexity was released in December 2022 by the company of the same name, which calls the service an “answer engine.” Unlike ChatGPT, Gemini, and Claude, which answer many questions from what their language models learned in training and retrieve web content only when needed, Perplexity is designed to draw on current web content for every question by default. That makes Perplexity a typical example of a generative engine. When people ask Perplexity about providers, products, or services, your business can appear in the answer as a mention in the text or as a numbered source; both shape your AI visibility.
How an answer comes together
According to Perplexity, the system first works out what a question means, then retrieves relevant content from the web and condenses the key information into an answer. That content comes mainly from Perplexity’s own index, a collection of crawled web pages; Perplexity can also fetch individual pages directly when someone asks a question. The company put the index at hundreds of billions of pages in September 2025 and updates it continuously. Perplexity says it splits documents into smaller passages and scores each one individually. Citability describes how well a single passage of a page works on its own.
The answer itself is written by a language model. Perplexity combines web retrieval with models it builds itself and with models from other providers such as OpenAI, Google, and Anthropic. By default, Perplexity picks a suitable model for each question; depending on the plan, users can also choose the model themselves. The same question can therefore be answered by different models, which can be one reason answers vary. Pro Search delivers more detailed answers, and a research mode, which Perplexity calls “Deep Research” in some places and “Research” in others, produces longer reports and draws on more sources to do so.
Through application programming interfaces (APIs), other companies can also build Perplexity’s index and its answers with citations into their own applications. Content from that index can therefore appear in AI answers outside Perplexity as well.
How Perplexity cites sources
The numbered citations in a Perplexity answer are AI citations: according to Perplexity, they link to the original sources, and the numbers show which source a statement is meant to come from. For businesses, that is a good starting point: they can see which websites Perplexity relies on for questions about their topics, such as their own website, industry portals and review platforms, or competitors’ pages. Because answers can vary from one run to the next, however, only recurring prompt tracking across many typical questions gives a reliable picture, the same principle that underlies the mention rate.
Perplexity marks some cited websites with a small shield icon and one of three labels: “Government” for the official websites of government organizations, “Academic” for scientific websites, and “Trusted” for sites that come up as sources often enough and publish information within their own area of expertise. According to Perplexity, the label applies to the whole website, not a single page, and is based on questions such as: Does the site correct its mistakes? Does it say who wrote each piece? Does it keep news separate from advertising and opinion? Perplexity does not say whether a label affects which sources it selects for an answer.
How Perplexity accesses websites
Perplexity collects websites for its index with the PerplexityBot crawler, which it says it does not use to train AI foundation models, the large language models that AI systems are built on; individual pages for a user’s question are fetched by Perplexity-User, a user-triggered fetcher. Whether PerplexityBot can reach a website depends on its robots.txt file and on bot protection that works independently of it, known as bot management.