Perplexity AI: the AI that actually cites its sources

The biggest objection to AI chat isn't that it can be wrong, but that you can't see when it is. A language model writes a made-up year with exactly the same confidence as a correct one. If you need to be able to justify something — for work, for a thesis, for a client — that's of no use to you.

Perplexity is built around exactly that problem. It isn't a chatbot with a search function tacked on, but a search engine that uses AI to summarize the pages it finds. Every answer comes with numbered references to the sources it's based on, so you can click through and check each claim. This page explains how that works, what it costs, and where Perplexity loses out to ChatGPT and Claude.

What does Perplexity do differently from a chatbot?

A regular AI chat answers your question from what the model picked up during training. That often goes well and sometimes goes spectacularly wrong, because a language model predicts which words are likely — not which are true. If there was no good answer in the training data, the model still produces something that looks convincing. In the jargon, that's called hallucinating.

Perplexity reverses the order. First search, then read, and only then write. The answer you get is a summary of sources retrieved at that moment, with a small number after each passage pointing to the page used. So you get two things at once: a readable answer and the ability to check it.

That has another practical advantage: timeliness. Because it searches live, Perplexity doesn't depend on when the model was trained. For questions about prices, rules or recent events, that's the difference between useful and outdated.

In practice you work with it differently than with a chatbot. You start with a broad question, look at the sources that appear below it, and sharpen your question based on what you see. A follow-up question triggers a new round of searching, with new sources. That makes it feel more like researching than conversing: you use the AI to get to the right pages faster, not to do the thinking for you.

Why do source citations matter more than they seem?

For a question like "explain quantum mechanics", it makes little difference whether you see sources. What matters are the cases where you have to base something on the answer.

• Work you're held accountable for. Advice to a client, a justification in a report, a figure in a quote. Without a source, an AI answer is a rumor.
• Dutch rules and amounts. Terms, rates and schemes change, and Dutch information is underrepresented in the training data of American models. In our experience, Perplexity does find Dutch sources and can cite them.
• Studying. With a thesis or paper, it's not just the answer that counts, but also where it comes from.

How good is Perplexity in Dutch?

In our assessment of Dutch language quality, Perplexity gets a 4 out of 5.

The weak spot is in the quotes. When Perplexity finds an English source and translates from it, the quoted part sounds stilted and literal. For reading and understanding that's not a problem, but if you want to use such a passage in your own text, you'll rewrite it. How all providers score on Dutch is in which AI speaks the best Dutch.

How much does Perplexity cost?

Perplexity bills in dollars, excluding VAT. At an exchange rate of $1 = €0.8687 plus 21% VAT, you end up with these amounts.

Two things stand out. First, the free version is unusually useful here: the core feature, search with sources cited, is included. With most providers the free tier is a stripped-down version of the product; with Perplexity you get the same product with fewer searches on the heavy model. If you look something up a few times a week, paying simply isn't necessary. Second, Education Pro at $4.99 per month is one of the sharpest student deals on the market; what else is available for students is listed in AI for students.

Keep in mind that Perplexity bills in dollars. Your monthly cost therefore moves with the exchange rate, whereas Google prices its subscriptions in the Netherlands directly in euros including VAT. On one subscription that difference is small; on three or four you'll notice it.

Perplexity Pro also includes $5 of Sonar API credit every month. That's only interesting if you build something yourself that runs searches; for a regular user it's a footnote.

What is Perplexity less good at?

Perplexity costs as much as ChatGPT Plus and Claude Pro, and that's exactly the problem: for the same money you get far more features with those two. Be honest about what you're giving up.

Long creative texts. Perplexity is built to answer, not to write. Ask for a blog post, a story or a persuasive page and you get something that reads like a summary of sources, because that's what it is. For writing, Claude is in a different class.

Iterating on a single text. In ChatGPT or Claude you keep refining a piece in the same conversation. In Perplexity every question starts over as a search, which breaks your writing rhythm.

Images and video. Perplexity doesn't make images or video. GPT Image 2 and Sora 2 are in ChatGPT Plus, Nano Banana Pro and Veo 3.1 at Google.

Your own documents. If you want to ask questions about a stack of your own PDFs rather than about the web, NotebookLM in Google AI Pro is better made for that. We explain it in Gemini AI explained.

Programming. Perplexity can answer a coding question, but it has no environment in which the AI writes and runs code itself.

When should you use Perplexity, and when not?

The practical rule of thumb: use Perplexity when the question "is this right" matters more than the question "how does this sound".

That list lays bare exactly why so many people end up with several subscriptions: each tool wins at something different. What that costs once you start stacking, we calculate in why multiple AI subscriptions are a waste of money, and how all the subscriptions compare with each other is in our comparison of AI subscriptions.

Conclusion