Perplexity is a search product built around language models rather than a chatbot with search bolted on. You ask a question in ordinary language, it retrieves current web pages, and it returns a written answer with numbered citations beside each claim. The citations are the product: they let you check whether the answer is supported, and they make the failure mode visible instead of hidden.
The behaviour that distinguishes it from traditional search is follow-up. Because the conversation has state, you can ask a broad question, then narrow it, then ask for a comparison, without restating context each time. Suggested follow-ups nudge you toward the questions you did not know to ask, which is genuinely useful when researching an unfamiliar subject. Focus modes restrict retrieval to academic sources, discussion forums, video or the wider web, and pointing it at academic papers produces noticeably more careful answers than the general mode.
Spaces collect research on one topic, optionally with your own uploaded documents, so a PDF or a spreadsheet can be questioned alongside the live web. For competitive research, literature reviews and due diligence, that combination of your documents plus current sources in one thread is the workflow people stay for.
The limitations need stating plainly. Answers are only as good as the pages retrieved, and when the web is wrong, thin or contradictory, the answer is confidently wrong in the same way. Citations prove a source exists, not that the source is reliable or that the sentence accurately reflects it, and the mismatch between summary and source is the error most people fail to check for. For anything consequential, the citation is a starting point for verification rather than a substitute for it.
There is also a structural question worth being aware of: this is a product that summarises publishers' work in place of sending them readers, and the relationship between answer engines and the sites they rely on is unsettled both commercially and legally.
Who it suits: anyone doing research where sources matter and where the question is more complex than a single query. Who should be careful: users who will treat a cited answer as verified, and anyone researching very recent or contested events.