An analyst opens her laptop at 7:40 a.m. and types one line: what did the central bank announce overnight, and who reported it first? Ninety seconds later she has a summary with numbered sources she can click. That is the whole loop. To use Perplexity AI, go to perplexity.ai or open its app, ask a full question in ordinary language, read the short answer, then open the citations to check them. Ask follow-ups in the same thread instead of starting over. Quick searches are free and need no account, while Perplexity Pro costs $20/month for deeper modes.
Your First Five Minutes
Treat it as a question box, not a keyword box.
- Open the web app, or install the macOS, Windows, iOS or Android app.
- Type a full question, including the constraint that matters: a date, a country, a budget.
- Read the answer, then click the numbered references before you use any fact.
- Ask a follow-up in the same thread. Context carries over.
- Switch modes when the first answer is too shallow.
Most first-time disappointment comes from typing three keywords out of habit. The tool is built to read a sentence, so give it one.
The reference list is the point. Perplexity summarises web sources and shows which one produced each claim, so you can audit the answer rather than trust it. That single habit separates people who get value from people who get burned.
Setting Up: Account, Apps, and Your First Query
You can run a quick search without signing up at all. An account starts mattering once you want history, saved threads, and file uploads.
Where it runs
Perplexity ships as a web app, mobile apps for iOS and Android, desktop apps for macOS and Windows, and two Chrome extensions, one of which makes it your default search engine, as Zapier’s walkthrough documents. There is also a separate AI browser, Comet.
Writing a query that earns a good answer
- Ask one question, not three stacked together.
- Name the timeframe: “since January 2026”, not “recently”.
- Say what you will do with the answer: compare, buy, cite, summarise.
- Paste the constraint instead of describing it, such as a job spec or an error message.
- End with a format request: a table, five bullets, a paragraph.
Settings worth changing on day one
- Pin a default model if you have a preference, or leave routing automatic and compare results for a week.
- Turn on the browser extension only if you actually want it intercepting your address bar.
- Decide early whether threads stay private or get shared as Pages, because habits harden fast.
- Keep one Space per project so research does not pile into a single scrolling history.
The mode ladder
Quick Search returns fast, basic answers and is available free. Pro Search digs into the specifics of your question and can ask you clarifying questions back; free accounts get a limited number, and it becomes the default once you pay. Research spends longer and produces an in-depth report, while Labs assembles documents, slides and dashboards from the same kind of query, per the same guide. Start low. Climb only when the answer is thin.
What Actually Makes It Different
The differentiator is no longer live web access. ChatGPT, Claude and Gemini all reach the web now. What remains distinctive is focus: every response is built around a search with visible sources, and the official getting-started material frames the product around real-time answers with sources rather than open-ended chat.
- Cited answers. Footnote-style markers show which source produced which claim.
- Model choice. It routes across several large language models, including GPT-5, Claude 4, Gemini 2.5 Pro and Grok 4, plus in-house models, and you can pin one or let it choose.
- Search homes. Finance, Travel and Academic modes lean on different source pools, such as filings or papers.
- Social search. Useful when you want opinions and reviews rather than press releases.
- File upload. Drop in a PDF or spreadsheet and search it alongside the web.
- Threads, Pages and Spaces. Threads are conversations, Pages are shareable reports, Spaces group related work.
- Discover and Tasks. Discover surfaces topical threads; Tasks delivers reports on a schedule.
- Comet. The company’s browser, which adds agentic actions such as comparing products across two tabs.
How the model routing changes answers
Routing matters more than most people expect. A reasoning-heavy model will restructure a messy question before answering it, while a faster model takes the question literally. If an answer feels oddly literal, re-run the same prompt on a different model and compare. The sources usually overlap; the framing rarely does.
A caution about mode sprawl
The naming is genuinely confusing, and the underlying behaviour is more similar than the labels suggest. Every mode still takes your query, finds pages, and summarises them. Do not assume a fancier mode means a more reliable fact.
Putting It to Work on Real Tasks
Research and study
Use Academic mode, ask for the primary source, then open it. A citation you have not clicked is not a citation. It is a rumour with a footnote.
- Draft a literature scan, then verify every paper exists.
- Upload a dense PDF and ask for the method section in plain words.
- Ask what the strongest counter-argument is, not just for a summary.
Work and commercial questions
- Build a shortlist of vendors, then ask for the objection to each one.
- Turn a messy brief into a structured comparison table.
- Ask for the number and the source page, then paste both into your doc.
- Schedule a recurring Task for a market or competitor you track weekly.
Everyday questions
Travel plans, product choices, live news, sports results. In one live-news test, Perplexity surfaced articles roughly half an hour old while another assistant fell back on months-old information, though it took a second prompt to get there, according to Zapier’s hands-on testing. That is the honest picture: fast, but not automatic.
A shopping example
Ask for “a laptop under $1,200 with at least 16GB of RAM for video editing, sold in the UK” and you get a shortlist with retailer pages attached. Ask for “best laptop” and you get a recycled listicle. The constraint is what turns a search engine into an assistant, and it costs you eight extra words.
When Answers Go Wrong: Fixes That Work
Most bad outputs trace back to a small set of habits rather than a broken product.
- Vague prompt. “Best CRM” gets a listicle. “Best CRM under $40 per seat for a 6-person agency” gets a decision.
- Unchecked citation. The summary can be right while one linked page says something narrower.
- Wrong mode. A Quick Search on a legal question will disappoint you.
- Stale thread. After five follow-ups, context drifts. Start fresh.
- Blocked sources. Some sites refuse crawlers, so their view never enters the answer.
- Treating it as a fact database. It is a summariser of pages it found, and pages can be wrong.
If an answer looks confident and thin, ask it directly: which source says this, and what would contradict it? That prompt fixes more problems than any setting.
Three repair prompts to keep handy
- “List every claim above that is not supported by one of your cited pages.”
- “Give me the primary source, not the article summarising it.”
- “What is the strongest argument against the answer you just gave?”
Each one forces the tool back onto evidence. The third is the most useful, because a summary that cannot survive its own counter-argument was never worth pasting into your work.
Privacy and what leaves your machine
Anything you paste becomes part of a query sent to a third-party service, and uploaded files are processed on their infrastructure. Treat client data, unreleased figures and personal records the way you would treat them in any cloud tool: read the current privacy terms yourself, and keep genuinely confidential material out of the box.
Three Workflows With Real Numbers
The freelance researcher
She bills $65 an hour and previously spent about 6 hours a month assembling market briefs. Pro at $20 a month cuts that to roughly 3.5 hours because the sourcing step collapses. The trade is 2.5 hours back for the price of a lunch, provided she still opens every link she cites.
The two-person policy desk
Two Pro seats cost $40 a month. Their rule: Perplexity produces the first pass and the source list, a human writes the analysis. They rejected the $200/month Max tier because unlimited Research and Comet access solved a problem they did not have.
The student who nearly got it wrong
He generated a reading list, cited four papers, and checked none. Two were real, one was mistitled, one did not exist in the journal named. He now verifies every reference in the library catalogue before it enters a draft. Total cost of that habit: four minutes per essay.
Reading these numbers honestly
None of these figures is a benchmark. They are illustrations of a simple arithmetic: the tool is worth its price when it removes more billable minutes than it costs, and only if verification stays in the loop. Run your own version of that sum with your own rate before you subscribe.
What the three have in common
Each one uses the tool for retrieval and keeps judgement in human hands. None of them lets a summary reach a reader unchecked.
Product, Course, App and Platform Experience
Perplexity behaves less like a chatbot and more like a research surface stretched across devices. The web app carries the full feature set. The mobile apps are built for quick questions and voice. The Chrome extensions push it into the browsing you already do, and Comet goes further by acting inside pages, currently for Max subscribers and a limited group of others, as documented in Zapier’s overview.
Here is how the tiers line up in practice.
| Tier | Cost | What you get | Suits |
|---|---|---|---|
| Free | $0, no account needed for a quick search | Quick Search, limited Pro Searches | Casual questions, trying the citation habit |
| Pro | $20 per month | Stronger models, more searches, larger Research and Labs limits | Freelancers, students, small teams |
| Max | $200 per month | Unlimited Research and Labs, top models, Comet access | Heavy daily research, agentic browsing |
Plan names and limits shift often, so check the current terms on Perplexity’s own pricing page before you commit a budget. If you want to understand the models underneath rather than only the interface, you can explore Coursiv AI lessons and build the prompting habits that make any of these tiers pay off.
Decision Framework: What to Know Before Deciding
Run these six checks before you pay for anything.
- Do you need sources, or ideas? Sourced answers favour Perplexity. Long drafting favours a general assistant.
- How often do you search per week? Under ten real research questions and the free tier is likely enough.
- Does your work get published? Then citation visibility is worth money on its own.
- Are your sources behind logins? Uploads help, but paywalled material stays out of reach.
- Who checks the output? If nobody does, no tier is safe.
- What breaks if it is wrong? A dinner recommendation and a compliance memo deserve different care.
Where it should not be your default
Creative long-form writing, code you intend to ship without review, and anything where a single wrong figure has legal or financial consequences. In those cases, use it to find the source, then read the source yourself.
Your next three steps
- Run the same question through a quick search and a Pro search and compare the reference lists.
- Rewrite one habitual keyword search as a full sentence with a constraint.
- Pick one weekly task and set it as a scheduled report for a fortnight, then judge whether it earned its slot.
If you want to keep going, best ai chatbot for everyday use is the obvious next stop, and best ai search engines is close behind.