What Are Perplexity AI's Features, and Which Ones Matter?

Perplexity started as one simple idea: a search box that answers your question in a paragraph and shows you where each claim came from. That is still the core of it. But the product around that core has grown enough that "what does Perplexity do" no longer has a one-line answer.

There are now several distinct search modes, a research agent that works for minutes at a time, a build tool that produces spreadsheets and small web apps, persistent workspaces, and a browser. Some of these are genuinely useful. Some are features you will open once.

This guide walks through what is actually there, based on Perplexity's own documentation, and flags which parts earn their place in a working routine.

What is Perplexity AI, in one paragraph?

Perplexity is an answer engine. You ask a question in natural language, it searches the live web, and it returns a written answer with numbered citations pointing at the sources it used. The citation layer is the whole point — it is what separates Perplexity from a chatbot answering from memory, and it is why people reach for it when the answer needs to be checkable.

Everything else in the product is built on top of that loop: search, read, cite, answer.

Getting cited by AI search is the new front page of Google. Get a free SEO audit and see whether your site is in the running.

The search modes

Perplexity's documentation describes several modes rather than one search, and choosing between them is the single biggest lever on answer quality.

Best mode

The default. Perplexity describes it as intelligent model selection — it looks at your question and routes it to whichever underlying model suits it. For most questions this is the right setting, and the time you would spend choosing a model yourself is rarely repaid.

Pro Search

The step up for questions that need more than one lookup. Perplexity frames it as extended access for multi-step reasoning: instead of one search and one answer, it breaks the question down, runs several searches, and assembles the result.

Use it when the question has parts. "Which of these three tools supports SSO and what does each charge for it" is a Pro Search question. "What year was this company founded" is not.

Research mode

This is the heavyweight, and the one most worth understanding. Perplexity's help documentation calls Research "Perplexity's advanced research feature that conducts in-depth research and analysis on your behalf, saving you hours of time."

Mechanically, it uses "search and coding capabilities to iteratively search, read documents, and reason" about what to look at next, then "synthesizes all the research into a clear and comprehensive report." You can then "export the final report to PDF or document."

It is not instant. Perplexity says Research "completes most research tasks in under 3 minutes," with responses taking "around 4 to 5 minutes to generate." That is the correct trade: you are buying a sourced report, not a chat reply.

On access, the documentation is straightforward: "Free users get limited access to Research. For those who need more, Perplexity Pro subscribers get extended access."

Labs: when you want an artefact, not an answer

Labs is the feature people most often miss, and it does something different from every mode above. Rather than returning prose, it builds things.

Perplexity's own description of what Labs produces covers "reports and spreadsheets to dashboards and simple web apps," plus charts and images. It writes and runs code, collects the results as downloadable assets, and can deploy small interactive apps that Perplexity calls mini apps.

The time profile is different too. Where Research runs for a few minutes, Labs tasks involve "10 minutes or more" of automated work — the pitch being that it can "accomplish in 10 minutes what would previously have taken days."

Labs is reached from the mode selector in the input bar, and at launch it was "available today for Pro subscribers."

When Labs is worth it

Reach for Labs when the output you want is a file rather than a paragraph. A competitor comparison as an actual spreadsheet. A small dashboard over data you paste in. A one-page tool you can send someone. Asking Labs for something you could have read in an answer wastes ten minutes.

The other thing Labs changes is how you phrase the request. Because it is going to run code and produce an artefact, being specific about the format pays off. Saying "build me a spreadsheet with one row per competitor and columns for pricing tier, free plan limits and SSO support" produces something usable. Saying "compare these competitors" produces prose you could have got in thirty seconds.

Ten minutes of automation beats ten hours of guessing. Get a free SEO audit and stop guessing about your rankings.

Projects: the persistent workspace

Perplexity's workspace feature is now documented as Projects, and if you last looked at the product when this was called Spaces, the capability has grown considerably.

The documentation describes Projects as "persistent, shareable Perplexity workspaces that keep searches, Computer tasks, files, instructions, and connected tools in one hub."

What you can put in one:

• Files — up to 5,000 files and 2.5 GB in total
• Custom instructions — up to 8,000 characters, applied to everything in that Project
• Default modes and models — set once, rather than chosen per question
• Collaborators — with Owner, Can edit and Can view roles
• Prioritised web links and domains — sources you want weighted

That combination is what makes Projects more than a folder. A Project with your own documents, a standing instruction about how you want answers framed, and a list of domains you trust behaves noticeably differently from the open search box.

The custom instructions field is the part most people under-use. Eight thousand characters is enough to describe your industry, your audience, the sources you consider credible and the format you want answers in. Written once, it removes the preamble you would otherwise retype into every question.

Prioritised domains are the second quiet win. If you work in a field where a handful of sources are authoritative and the rest of the web is noise, weighting those sources changes the quality of every answer in the Project rather than one at a time.

Your content can be the best answer on the web and still not get cited. Get a free SEO audit and find out why.

Model choice

Pro subscribers can pick the model behind an answer rather than accepting the default. The picker includes Perplexity's in-house model, Sonar, alongside current frontier models from OpenAI, Anthropic and Google, plus reasoning-focused options.

One deliberate omission here: we are not printing specific model version numbers. The contents of that picker change every few weeks, and any list written today will be wrong by the time you read it. Open the model selector and look — that is the only reliable source.

The practical advice is simpler than the menu suggests. Leave it on the default unless you have a reason. The reasons that actually come up are wanting a reasoning model for a genuinely hard analytical question, or preferring a particular model's writing style for drafting.

Comet: the browser

Comet is Perplexity's Chromium-based browser, launched in July 2025 and opened to everyone in October 2025 after an initial subscriber-only period.

The idea is that the assistant travels with you rather than living in a separate tab. It can see the page you are on, summarise it, answer questions about it, and carry context across the pages you visit.

Whether that is worth switching browsers for depends entirely on how much of your work is reading and comparing web pages. For research-heavy roles it removes a lot of copy-pasting. For everything else it is a browser.

Other features you will encounter

• Pages — turns a piece of research into a structured, shareable article-style write-up
• File and media upload — PDFs, CSVs, audio, video and images as context for a question
• Image generation — available to Pro, with limited access to video generation
• Finance — stock data, earnings and company comparisons
• Shopping — product research with buying integrated
• Computer — an agent that executes multi-step tasks across the web and external services, introduced in 2026

Limitations worth being honest about

Citations prove sourcing, not correctness. A numbered reference tells you where a sentence came from. It does not tell you the source was right, or that the summary represents it fairly. For anything consequential, open the links.

It is only as good as what is indexed. Perplexity is searching the live web. Where the web is thin, wrong or dominated by SEO sludge, the answer inherits that. This shows up most on niche technical questions and on anything that happened in the last few hours.

Paywalled and gated material is largely invisible. A great deal of the best writing on any specialist subject sits behind a login. Perplexity will answer confidently from the free web around it, which is not the same thing as answering from the best available source.

Modes are easy to misuse. Running Research on a factual lookup wastes four minutes. Running default search on a genuinely multi-part question gives you a shallow answer you will have to redo.

Plan boundaries move. Which features sit behind Pro, and how much of each you get, has changed repeatedly. Check the current pricing page rather than any article, including this one.

The publisher relationship is contested. Perplexity has faced legal action and complaints from multiple publishers over how it uses their content. That is worth knowing about if you publish for a living, and it is an active situation rather than a settled one.

How it compares to the alternatives

Perplexity vs Google. Google is still better for navigational searches and for anything where you want to see the range of results yourself. Perplexity is better when you want the answer assembled, with sources, in one place. They are not really competing for the same query.

Perplexity vs ChatGPT. ChatGPT is the stronger general assistant — writing, coding, open-ended thinking. Perplexity is the stronger research tool, because citation-first is its default rather than a mode you switch on. If you mostly need sourced answers about current information, Perplexity wins; if you mostly need a collaborator, it does not.

Perplexity vs a dedicated research tool. Research mode and Labs cover a lot of what people previously used specialist tools for, but the output is a starting point. It is a very fast first draft of an analysis, not a finished one.

Final thoughts

The honest summary of Perplexity's feature list is that three things matter and the rest is optional. Research mode, for questions that deserve a report. Labs, when you want a file at the end. Projects, if you use it often enough that a persistent setup pays off. Those three change how the tool works for you.

The search modes underneath are worth five minutes of learning, mostly so you stop running the heavyweight ones on lightweight questions. Model choice you can safely ignore. Comet depends on your browsing habits more than on the feature itself.

If your interest in Perplexity is less about using it and more about appearing in it, that is a separate discipline with its own rules — our step-by-step checklist for ranking in ChatGPT, Claude and Perplexity covers what actually influences which sources these engines cite.

Being the answer beats being a blue link. Get a free SEO audit and see how visible you are in AI search.