How to use Perplexity: the six modes, 15 prompts and what free gets you

How do you start using Perplexity?

Perplexity is a search engine that answers in sentences and numbers every source it used. On 17 October 2026 its help center listed 11 Search models. Pro subscribers can choose 7 of them by hand, Max subscribers can choose all 11, and free accounts can't choose any. So the automatic setting, Best, is the one you'll use most, and Perplexity's Pro page says Best "is available without quota limits."

Seven steps get you from a blank page to an answer you can trust:

1. Open it. Go to perplexity.ai, or install the iOS, Android or desktop app. You can ask without an account.
2. Sign in first. Sessions you start signed out are anonymous, and Perplexity deletes them after 14 days. They can't be recovered.
3. Ask a full question. "What is the current inflation rate in the US?" beats "inflation". That pair is Perplexity's own example.
4. Leave the model on Best. It picks a model per question and costs you nothing from any allowance.
5. Open the citations. Select a numbered source to read the page behind a claim.
6. Stay in the session. Ask the next question in the same place, and Perplexity keeps the context.
7. Change mode only when you need to. Pro Search and Deep Research both draw on a limited allowance.

To change the model on the web, open the dropdown at the bottom right of the question box and choose Custom. Perplexity's model page also lists a Thinking toggle, which gives a model more time to reason before it answers. You can switch the model or the mode in the middle of a session. Perplexity also warns that a model inside Perplexity "can behave differently from the same model in the provider's standalone product", because it runs with Perplexity's own search, prompts and limits.

Which Perplexity mode should you use for each question?

Use Search on Best for quick facts, Pro Search for questions with several parts, and Deep Research when you want a report you'd otherwise assemble by hand. Perplexity's help center names six modes. A free account gets Search and one Deep Research report a month. Model Council, the Comet Assistant and Computer all need a paid plan.

Every cell comes from Perplexity's plan comparison, its mode pages and its Computer page, read on 18 October 2026. Prices are United States prices from the App Store listing and Perplexity's own pricing page.

Pro Search is the easiest allowance to waste. Perplexity's own comparison table calls standard search "Fast, basic answers" from one or two sources, and Pro Search "Slower, in-depth answers" from dozens. A question like "what time does the Louvre open" doesn't need dozens of sources. A question like "compare three budget airlines on fees for a family of four" does.

Deep Research takes longer. Perplexity says most reports finish in under 3 minutes and some take 4 to 5. You can't pick its model, because it combines several on its own. On a free account that one report is your whole month, so draft the prompt in an ordinary Best session first, then paste it into Research once it's right.

Older guides tell you to pick a Focus mode such as Academic, Writing or Social. Perplexity's Pro Search page still lets you narrow a search to one kind of source, such as Academic, Finance or your own files. Signed in to an organization account, you won't see focus mode at all. Its focus mode page points to a Choose sources button there, and None answers without searching, which suits a rewrite of text you paste in.

How do you write a Perplexity prompt that gets a better answer?

A good Perplexity prompt says what you want, who it's for, which sources count, and what shape the answer should take. Perplexity's own prompting guide opens with "The golden rule: State your intent clearly and provide detail." Its practical tips page breaks that into five parts: an instruction, context, the input, keywords, and the output format.

Two habits matter more here than in ChatGPT or Claude. Perplexity searches the web on almost every question, so the words you use become its search terms. Put the exact product name, the place and the date range in the prompt. And ask for a table whenever you're comparing things, because a table forces one source per row and makes gaps obvious.

Copy and paste skeleton: [Task] about [topic]. I am [who you are] and need this for [purpose]. Use [source types, such as peer reviewed studies or .gov sites] from [date range]. Format: [table / five bullets / 200 words]. Cite a source for every number.

Research and fact checks

1. What does the latest research say about [topic]? Use peer reviewed studies from the last three years only. Give me five findings in a table with the study, year, sample size and result.
2. Is this claim accurate: "[paste claim]"? Find the original source of the claim, then tell me what it actually says. Quote the exact sentence.
3. Summarize this page in five bullets and list anything it states without a source: [paste URL].
4. Give me a balanced briefing on [news story] from at least three outlets with different viewpoints. Note where they disagree on the facts.

Shopping and travel

1. Compare [product A], [product B] and [product C] in a table: price in [country], battery life, warranty, and the most common complaint in owner reviews. Link every price.
2. Plan a [number] day trip to [city] in [month] for [who]. Include opening hours and ticket prices, and flag anything that needs booking ahead.
3. What fees does [airline or bank] charge for [service] as of this month? Cite the provider's own fee page, not a blog.

Work and job search

1. Find the last 12 months of news about [company]: funding, layoffs, product launches and leadership changes. One line each, newest first, with links.
2. I'm applying for [job title] at [company]. Summarize what the company says about its values and recent priorities, and give me three questions I could ask in the interview.
3. What is the typical salary range for [job title] in [city]? Use at least three sources, show each source's figure separately, and say which year each one is from.
4. Write a cold email to [person or role] at [company] about [topic]. Under 120 words, plain tone, one specific reference to something they published recently.

Study

1. Explain [concept] to a first year student. One analogy, one worked example, then three practice questions with answers hidden at the end.
2. Here are my lecture notes [attach file]. Cross check them against current sources and list anything that looks outdated or wrong.

Deep Research

1. Research the market for [product type] in [country]: size, growth, the five biggest sellers and their prices. End with the three biggest risks for a new entrant, with sources.
2. Build a report on [policy or regulation]: what it requires, who it applies to, key dates, and how [group] is affected. Separate official sources from commentary.

How do you check whether a Perplexity answer is right?

Open the citation behind every number you plan to repeat, and read the sentence it came from. Citations make checking fast, but they don't make an answer correct. In a March 2025 test by Columbia Journalism Review's Tow Center, Perplexity answered 37 percent of its queries incorrectly. The test gave eight AI search tools the same 200 excerpts from news articles, 1,600 queries in all, and asked for each article's headline, publisher, date and link. Grok 3 got 94 percent wrong, and the eight tools together got more than 60 percent wrong.

Three controls help. Select or hover over a source to see its label: Perplexity marks rated domains with a shield and one of three labels, Government, Academic or Trusted. Perplexity's own source label page adds that "No label does not mean a site is low quality." Second, the Rewrite button answers the same question with a different model, which is a quick way to see whether the answer survives a second opinion. Third, the source button under an answer shows which model, mode and source setting produced it.

When two citations disagree, ask Perplexity directly: "Sources 2 and 5 give different numbers. Which is newer, and which is the primary source?" It'll usually find the original. If it can't, treat the number as unverified.

How do you use Perplexity with your own files?

Attach a file with the + Attach button in the question box, or drag a file or folder onto it, then ask about it in the same session. Perplexity's file upload page caps every file at 40 MB. It accepts text, code, PDFs, images, audio and video. Audio and video get transcribed with speakers labeled, but the pictures in a video aren't searchable.

Long files get trimmed. Perplexity says it reads short files whole and pulls "the most important parts" from long ones. So ask about a named section ("what does section 4 say about termination?") instead of asking for a summary of a 200 page PDF, and check the quotes it gives you against the file.

Uploads run on a rolling weekly allowance on Free, Pro and Max. Pasted images and files pulled in from Google Drive count too. Perplexity doesn't show how many you have left. Each upload stops counting about a week after you attached it, so there's no single reset day.

For files you use again and again, make a Project. A Project groups sessions and files by topic, carries its own instructions, and can be shared with other people. Perplexity's plan page allows up to 50 file uploads per Project on Pro.

How do you use Perplexity for free?

Use Best for almost everything, save the 3 daily Pro Searches for questions with several parts, and spend the 1 monthly Deep Research report on the question that would take you longest by hand. Perplexity's plan page describes free as "Practically unlimited basic searches" with a "Very limited amount of Pro Searches". The table on the same page puts those at 3 a day and 1 Research query a month.

The same table marks advanced models, image generation, file and app creation, and the Comet Assistant as No on Free. File uploads are listed as Limited. You also can't pick a model by hand, so Best always chooses for you.

If you outgrow that, Pro is $20 a month in the US App Store, or $17 a month billed yearly on Perplexity's pricing page. Verified students can get Education Pro for $10 a month through SheerID. Before you pay, read what Pro's weekly limits actually allow in practice, because Perplexity doesn't print them.

When is Perplexity the wrong tool?

Perplexity is built to find and cite. It's the wrong first choice when the job doesn't need the web. A long draft in your own voice, a story, or a rewrite of text you already have gets little from twenty search results, and the citations can pull the draft toward whatever the top pages say. How to use Claude covers a drafting first workflow.

It's also the wrong plan for some jobs on Free and Pro. Perplexity's plan table lists image generation as No on Free. Model Council, the feature that asks three models and merges their answers, is Max only at $200 a month. And Computer, the agent that sends email and builds files, needs a subscription and credits.

If what you want is several models on one question, a multi model workspace is the cheaper route. Whizi puts GPT, Claude and Gemini behind one login, with web search on by default in every new chat. Perplexity is the stronger pure search tool, though: a normal Whizi web search pulls up to 3 results, while Pro Search reads dozens. Whizi vs Perplexity Pro compares the two, and how to use multiple AI models together shows the search, draft and review loop.