Perplexity launches its AI answer engine

Background

A conventional web search engine responds to a query with a ranked list of links, leaving the reader to open pages and assemble an answer. A large language model can instead write an answer directly, but an answer produced by a model alone gives the reader no obvious way to check where the information came from. Perplexity's launch sat between the two approaches: a written answer, with the sources it drew on attached.

What happened

On 7 December 2022 Perplexity launched its search engine ^[1]. The product answers queries with large language models and cites its sources ^[1]. In practice that means a user puts a question to the system and receives a short written response rather than only a list of pages, with the sources behind the response marked so the reader can follow them back ^[1]. The citation of sources is the detail that separates this description of the product from a plain chatbot: the answer is presented together with the material it was built from ^[1].

The company behind it

The engine did not arrive from a long-established search company. The company had been founded on 22 August 2022, a few months before the launch, by Aravind Srinivas, Denis Yarats, Johnny Ho and Andy Konwinski ^[1]. TechCrunch reported in April 2023 that Perplexity was founded in 2022 by the same four people, describing them as engineers with backgrounds in back-end systems, AI and machine learning ^[2].

The same report sketched where the founders had worked before. Yarats had been a staff machine learning engineer at Quora before joining Meta as an AI research scientist; Ho had worked at Quora and then as a quantitative trader; and Konwinski was among the founding team at Databricks ^[2]. Those backgrounds — search and question-answering products, machine learning research, data infrastructure — map onto the parts such a product needs: a way to retrieve material, a model to write an answer, and systems to serve it.

Early growth

Adoption came quickly for a product launched in the final weeks of 2022. By February 2023 the company reported two million unique visitors ^[1]. TechCrunch reported in April 2023 that Perplexity had 10 million monthly visits and 2 million unique visitors in February 2023 ^[2]. The two figures measure different things — visits count sessions, unique visitors count people — and together they show a small but active user base within roughly two months of launch ^[1][2].

Why it matters

The launch is one of the early public examples of a search product built around large language models rather than around a list of links, and it made source citation part of the interface from the start ^[1]. That choice addresses a known weakness of model-written text: a confident sentence is hard to verify unless the reader is told what it rests on. By attaching sources to the answer, the product keeps the checking step that a link list provides while changing the output from links to prose ^[1].

It also illustrates a pattern that became common: a small team of engineers with search, machine-learning and infrastructure experience assembling an answer engine on top of existing models, and reaching a measurable audience within months of founding ^[1][2]. Perplexity was founded in August 2022 and its engine went live in December of the same year; by February 2023 it reported two million unique visitors ^[1].