Perplexity

Perplexity is an AI-native search engine that answers questions with cited sources rather than returning a list of links. Founded in 2022 by Aravind Srinivas and former Google/DeepMind researchers, Perplexity represents the vanguard of how AI agents are disrupting traditional search and reshaping web discovery.

Answer Engines vs. Search Engines

Perplexity's core insight is that users want answers, not links. By combining large language models with real-time web retrieval, Perplexity synthesizes information from multiple sources into coherent, cited responses. This is fundamentally different from traditional search: instead of directing users to websites, Perplexity reads the web on their behalf — an early implementation of the agentic web paradigm.

The Discovery Layer of the Agentic Web

As Jon Radoff has analyzed in his work on the agentic web, AI-powered discovery is restructuring how humans find information, products, and services. Perplexity is the most visible consumer example of this shift. Its Pro Search feature demonstrates multi-step reasoning: decomposing complex questions, searching multiple sources, and synthesizing results — behavior that approaches autonomous agent capabilities.

LLM Optimization

Perplexity's rise has created a new discipline: LLM optimization — ensuring that brands and content are discoverable and accurately represented in AI-generated answers. As Jon Radoff has explored, your SEO might be perfect, but to the AI replacing Google for your customers, you might not exist. This is a fundamental challenge for businesses operating in the emerging agentic discovery layer.

How Perplexity Cites in 2026

Perplexity cites more sources per answer than its rivals and leans on each one less. Zhang, He and Yao (arXiv, April 2026; 21,143 citations) measured a mean of 16.35 citations per prompt on Perplexity, against 12.06 for Google and 6.88 for ChatGPT. The mean influence of a single cited page on the final answer ran the other way: 0.065 on Perplexity and 0.058 on Google, against 0.27 on ChatGPT. The authors separate "selection" (getting cited) from "absorption" (shaping the answer). On Perplexity a citation is comparatively easy to earn and comparatively light in effect, so a count of AI citations overstates how much a page actually steered what the user read.

The citations cluster on a few community and reference sites. In Ahrefs' September 2026 analysis of 3.1 million US queries, Reddit took 21.6% of citations among Perplexity's 50 most-cited domains ("the most concentrated top source of any assistant we looked at"), followed by YouTube at 20.8%, Wikipedia at 6.3%, Facebook at 4.5% and Amazon at 3.6%. These are shares of the top 50 sources, not of every citation, and panels that use a different denominator report much smaller numbers. Perplexity also appears less tied to recently edited pages than other engines: Seer Interactive (July 2026) found 65% of the pages it cites had been updated within the past year, against 73% for ChatGPT and 78% for Gemini.

Being visible on Perplexity says little about visibility elsewhere. A SIGIR 2026 study of 11,500 real queries put cross-platform source overlap below 0.2 on the Jaccard index, and repeated sampling shows roughly 65% of the sources cited for a prompt change from one day to the next (Schulte, Bleeker and Kaufmann, April 2026). Any single snapshot is one draw from a distribution.

The Crawlers Perplexity Operates

Perplexity's documentation lists two user agents. PerplexityBot builds the search index: it is "designed to surface and link websites in search results on Perplexity", honours robots.txt, and according to the company "is not used to crawl content for AI foundation models". Perplexity-User fetches a page when a person's question requires it, and Perplexity states that "since a user requested the fetch, this fetcher generally ignores robots.txt rules". Both publish their IP ranges as JSON files. A site that disallows PerplexityBot is therefore opting out of Perplexity's index, not out of model training, and a robots.txt rule alone does not stop user-initiated fetches. The same three-way split between training, search-index and user-fetch bots now runs across the major AI crawlers.

Implications for Publishers

Perplexity's model raises profound questions about the economics of web publishing. If AI agents synthesize and deliver answers without users visiting source websites, the advertising and traffic models that fund content creation collapse. This tension between AI utility and publisher economics is one of the defining challenges of the agentic web transition.