Jason LeePerplexity Pro
Item 15 of 45
Perplexity answers in the voice of a reference desk that does not work there
The best reference desk I have ever seen was in a university library in the years when I still consulted, and the woman who ran it had a habit that took me a while to name. Whatever you asked, she answered in two sentences and then did something no search product has ever done to me: she showed you the shelf. Not the answer. The shelf, with her hand on the spine of the third book from the left, and a sentence about what else was on it. Students treated her as slow, because a search would have been faster, and she treated them as future adults who would one day need to find things without her. I asked her once why she never just told people the answer. She said the answer expires. The shelf does not.
That desk is the argument, and I want to hold it against the product honestly, because Perplexity is the closest thing to that desk the internet has produced, and it is also the furthest thing from it. Perplexity is an answer engine: you ask, it searches live, and it returns a summary with citations numbered and clickable, so that every claim arrives wearing its source like a name tag. The Pro tier, twenty dollars a month, buys a larger allowance of Pro searches, model choice, file analysis, the standard 2026 bundle, per the pricing reviews that have tracked it. The product is genuinely good. The shelf is the part it cannot hand you, and the interesting question is who pays for the difference.
The citation is a receipt, not a source
The mechanism deserves precision, because Perplexity's entire market position is the word "citations," and the word is doing more work than the citations are. When the engine answers, it composes sentences first and cites them second, which is the inverse of the reference desk, where the source is found first and the sentence is composed to survive it. The citation tells you where a claim could be checked. It does not tell you the claim came from there, and the difference is invisible until the day you check one. Wired's reporting in 2024 documented answers that reproduced the outlet's work nearly whole while citing it thinly, and later analyses, including one by Cloudflare, described the company's crawlers ignoring robots.txt, the web's polite sign asking machines to stay out. Forbes caught it summarizing an exclusive before the ink was dry. The pattern across the episodes is consistent and was reported by Wikipedia's summary of the disputes: Dow Jones and the New York Post sued in June 2024, the BBC threatened action in June 2025, and the Yomiuri Shimbun filed suit that August, while the company's response has been that it aggregates rather than plagiarizes and is open to revenue-sharing. I am going to be unsympathetic for one sentence: aggregation is what plagiarism calls itself when it has a pitch deck.
The user-facing complaints, past the ethics, are the smaller failure mode and worth reporting accurately. Answers can be confidently wrong in the specific way that reads as authority, because the formatting is the authority: numbered citations, a calm register, a structure that looks like scholarship. Reviewers in 2026 praise the speed and the file analysis and complain about the same thing I hit in my own use, that the engine occasionally summarizes a source it did not fully read, and the citation sits there like a receipt for a purchase you never made.
The publisher lawsuit is the incentive audit
Consider the incentive, because the economics of the answer engine are the most nakedly extractive in this series and the lawsuit explains them. A search engine sends you away; the departure is its honesty, and the publisher's page earns the visit. An answer engine keeps you, and the visit it consumes was the publisher's only income, which means every answer is a small withdrawal from an account nobody audited. Perplexity's own response to the suits, offering revenue-sharing, is the tell: the company has proposed becoming the tollbooth on content it did not build, and the publishers are being asked to accept a share of the value their own work creates in someone else's product. Who benefits if you believe the demo? The demo is research at the speed of thought. The meter is the twenty dollars, and the inventory behind the meter is other people's work, acquired at the price of a crawler and defended in court. The company is venture-backed and its costs are model access and litigation, and neither shrinks when you subscribe, which is why the free tier is generous and the Max tier now sells at two hundred dollars a month: the ladder exists, and the bottom rung is where you learn to climb.
My own incentives here are not clean, and I would rather say so than be caught by a reader who knows. I write essays for a living. If answer engines become how people read, my roof is one of the things being aggregated, and my complaint about Perplexity is also a complaint about the floor I stand on. Both things are true, and I have tried to let the good one speak in the next section rather than the last.
The desk was never the point, argued properly
The strongest counterargument is that I am romanticizing a service most people never had. The reference desk was staffed by professionals, in institutions, at public expense, and it is dying everywhere, defunded and unstaffed, while the questions it answered pile up unanswered. Perplexity is what a civilization with no librarians builds instead, and for the person with a genuine question and no shelf of their own, numbered citations that can be checked in two clicks are a real, checkable discipline, closer to honesty than anything the ad-funded incumbents ever served them. The engine is not pretending to be a librarian. It is what the budget replaced the librarian with, and the fault the lawsuits describe belongs to the extractors, but the vacuum was already there. Both things are true. The answer engine is a genuine good built on a genuine taking, and the taking does not undo the good, and the good does not justify the taking.
It depends, then, on what kind of question you carry to it. If it is a question with a right answer, Perplexity is the best tool of its decade and the citations will catch most of its own errors. If it is a question that needs the shelf, the tool will give you a fluent answer and you will never learn what else was on it.
The variable worth watching is the licensing deals. The day the engine's answers are built on paid content, the price of the answer will include the price of the source, and the tollbooth will be legal. The librarian died of a budget. The answer engine was born of one, and it has not yet decided whose.