What is Perplexity and how does it choose the products it recommends?

Perplexity is both the name of a company and the name of its product, and almost everyone who searches for it wants to get into it. A store has a different reason to care: Perplexity is an answer engine that recommends products by name and with a link, and in the United States it already sells them without the buyer visiting the store. If a customer asks it what to buy, the answer comes from the pages Perplexity was able to read and choose. This entry explains how it gets to those pages, which part you control and which part can only be measured.

How it finds pages

Perplexity reads the web with two agents, and each is managed separately in robots.txt. Changes can take up to 24 hours to show in its systems.⁠^{[1]Perplexity. Perplexity Crawlers}

The practical difference is big. PerplexityBot is a crawler in the classic sense: it walks the web and feeds what Perplexity can show. Perplexity-User acts in the moment, for one specific question. Both identify themselves with their own user agent and Perplexity publishes each one's IP addresses in a JSON file, so you can check whether a visit that claims to be theirs really is.⁠^{[1]Perplexity. Perplexity Crawlers}

What it uses to choose

How Perplexity ranks the sources in an answer and the products it recommends is not visible from outside. The closest thing it documents is its search API, the one it sells to developers: it returns ranked, real-time web results, with controls for domain, language and region, and with the content extracted from each page.⁠^{[2]Perplexity, developer documentation. Perplexity Search API} That is the API, not the app that shoppers use, but it shows the raw material: a relevance order, filters and the text of each page. In the app, when it visits a page to answer, Perplexity links it in the answer;⁠^{[1]Perplexity. Perplexity Crawlers} grounding what it writes in retrieved pages is what is called grounding.

There is one more clue, an academic one. The paper that named GEO (Generative Engine Optimization) tested its rewriting methods on Perplexity with 200 queries. Because Perplexity does not let you specify source URLs, the authors gave it the text as uploaded files and asked it to answer only from them. Adding quotations improved by 22% the share of the answer attributed to the source, and the best methods reached 37% more in the subjective impression of visibility; keyword stuffing, by contrast, lowered the share attributed to the source below leaving the text untouched.⁠^{[3]arXiv (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande; KDD 2024). GEO: Generative Engine Optimization} It is a 2023 and 2024 experiment on given sources, not on how Perplexity chooses what to read.

Buying without leaving Perplexity

Since November 25, 2025, users in the United States can buy inside the Perplexity chat from stores that take PayPal. PayPal presented it as the culmination of a strategic partnership announced earlier that year, with stores such as Abercrombie & Fitch, Ashley Furniture, Fabletics, Adorama or Newegg.⁠^{[4]PayPal, newsroom (November 25, 2025). PayPal and Perplexity Launch Instant Buy Ahead of Black Friday}

What PayPal says about how a store gets in: one sync service makes its catalog discoverable in Perplexity's results, and another lets it get paid there with the PayPal setup it already has. The purchase happens in Perplexity's interface, but the store remains the seller of the order (merchant of record) and gets paid directly.⁠^{[4]PayPal, newsroom (November 25, 2025). PayPal and Perplexity Launch Instant Buy Ahead of Black Friday} It is an agentic checkout: the assistant recommends and the payment happens in the same conversation.

In a store

A made-up example: a coffee store in the United States with a good product page for its hand grinder. Someone asks Perplexity: "which hand grinder works for espresso for under $100?" The routes by which that grinder can end up in the answer:

On the first two routes, what decides whether the page is useful is what it says: if "suitable for espresso" and the price are in the text, the answer can use them; if they are in an image, it is harder.

How to check it

1. Open your robots.txt and see whether PerplexityBot can get through. If you change it, allow up to 24 hours.⁠^{[1]Perplexity. Perplexity Crawlers}
2. Review your firewall or CDN. Perplexity explains how to allow its agents by combining the user agent with its published IP ranges, because a firewall can block them even if robots.txt lets them through.⁠^{[1]Perplexity. Perplexity Crawlers}
3. Look for its visits in your logs and separate PerplexityBot's from Perplexity-User's. Check the IP against the JSON files Perplexity publishes to rule out impostors.⁠^{[1]Perplexity. Perplexity Crawlers}
4. Ask your customer's question in Perplexity several times, on different days and in new threads, and write down the sources and the products. A single answer is a sample of one.
5. If you sell in the United States and take PayPal, ask about the catalog sync that gives access to purchases inside Perplexity.⁠^{[4]PayPal, newsroom (November 25, 2025). PayPal and Perplexity Launch Instant Buy Ahead of Black Friday}

Related concept: answer engine

Answer engine is the category, and Perplexity is one of its clearest examples: PayPal's own announcement refers to Perplexity as "the answer engine".⁠^{[4]PayPal, newsroom (November 25, 2025). PayPal and Perplexity Launch Instant Buy Ahead of Black Friday} The category also includes Google's AI summaries, ChatGPT with search and Copilot, each with its own way of retrieving pages; AI search compares them. What is specific to Perplexity, for a store, is its two agents, which are managed separately, and a purchase that in the United States already happens inside the answer.

To go further, the AI search reading path orders the rest of the terms, and the guide on what an AI crawler sees in your store goes down to the detail of each agent.