Perplexity Pro Review

The dirty secret of AI chatbots is that they are most confident precisely when you can least check them. Perplexity's founding bet is that the fix isn't a smarter model — it's receipts. Ask it anything and the answer arrives footnoted, with the sources lined up for inspection. That one design decision changes what the tool is for: less a conversation partner, more a research assistant who shows their work.

For the tasks it was built for, it is quietly transformative. Comparing appliance warranties, checking what a medication interaction actually says on reputable health sites, getting the state of play on a news story with the coverage laid out side by side — work that used to mean fifteen open tabs collapses into one query and a scan of the sources. The Pro tier adds heavier searches and a choice of which underlying AI model does the reasoning, which matters more than it sounds: different models summarise with different temperaments.

Two honest warnings. First, a citation is not automatically a good citation — Perplexity will sometimes footnote a content-farm listicle as readily as a journal, and the discipline of clicking through remains yours. Second, the tool makes summary feel like reading. For anything consequential, the sourced answer is where research starts, not where it ends. Used with that discipline, it's the closest thing the AI market has to an honesty feature.

There is a subtler failure worth knowing about, because it survives the click-through habit. The sources can each be perfectly respectable while the summary drawn from them is still wrong — a sentence assembled from two different articles can assert something neither of them says, and the footnote at the end of it will point at a page that looks like it supports the claim. The tell is a claim that feels more specific than any single source would plausibly be. When something matters, read the sentence the citation is actually attached to rather than accepting that a citation exists.

How you ask changes what you get here more than it does with a conversational assistant. Vague questions produce vague sweeps of the open web; questions with a constraint in them — a year, a jurisdiction, a named standard, a preference for primary sources — steer it somewhere better. The academic focus mode is the single most useful control for anyone doing serious reading, and the follow-up questions it suggests are worth using rather than ignoring: the second and third query are usually where the actual answer surfaces.

It is also worth being clear about what this category is doing to the web it draws on. An answer engine that summarises publishers' work and satisfies the question in place means fewer people reach the page that did the reporting. That is not a reason to refuse to use the tool — the convenience is real and it is not going away — but it is a reason to click through on the sources you find genuinely useful, rather than treating every article as raw material for a summary.

As a general assistant it is merely fine. Drafting, editing, and creative work are noticeably better on the products built around them, and if you can only justify one subscription, that trade-off decides it: mostly writing, buy elsewhere; mostly finding out and deciding, buy this. A fair number of people will also discover the free tier covers their research life entirely — try it there first, and let your own usage tell you whether Pro earns its fee.