Perplexity: search that cites itself, and the $200 tier most people skip

Perplexity search answers a question and puts a link against each claim in the answer. Under the hood it runs retrieval-augmented generation: the model searches first and writes from what it found, instead of writing from memory. RAG (Retrieval-Augmented Generation) is the part that makes a citation possible at all. Aravind Srinivas, Denis Yarats, Johnny Ho and Andy Konwinski founded it in 2022, coming out of OpenAI, Meta and Databricks. The decision that separated it from a chatbot was editorial: you check the answer line by line, and a wrong link shows up as a wrong link.

“Search that shows where the answer came from”

— Intokened, Our card on the service

What it does

That design makes it a working tool where an error costs money or reputation. A lawyer checking a precedent, an analyst pricing a market and a journalist verifying a claim all need the citation more than the prose. Our service card keeps the features and the checked prices together.

The features follow from that. Multi-step deep research runs one question across many searches and returns a report with its sources. The service reads files you upload, works on a phone and exposes an API. The free tier is real: sourced search without a cap on ordinary questions, with deep research limited.

What the tiers cost

Three tiers hold the pricing. Free costs nothing. Pro costs $20 a month and adds model choice and deep research. Max costs $200 a month for maximum model access, ten times Pro, which is $2,400 a year against Pro's $240.

Pro buys the product most people mean when they recommend Perplexity. Max buys the heaviest models for someone running research all day. At $200 a month it competes with hiring help, and the question to ask is how many hours of your own time it has to save each month to pay for itself.

Who it fits

Who it fits: anyone who has to show their working, and anyone building a pipeline that needs citations it can pass on. Someone who wants a model to draft an email needs none of it, and our automation selection points at the tools that do that job instead.

We measured the other side of this market yesterday, when English Wikipedia served 9.3% fewer human pageviews than a year earlier, because assistants answer out of their weights and send nobody anywhere. Perplexity sends the reader to the source, which is the whole product and the reason its answers can be audited at all.

Informational material, not a purchase recommendation. Plan prices come from the company's own page and were checked on 12 October 2026.