How to use Perplexity well

What Perplexity is shaped for

Answering a question from the live web with sources you can click, immediately.

• Search-first retrieval with inline, clickable citations
• Good at narrow, checkable, current-facts questions
• A research mode that assembles a longer report across many sources
• Persistent spaces that keep a research thread accumulating

Where it gets its information

Search-first: it retrieves, then answers from what it retrieved, with inline citations. The citations are the product — an answer you do not click through is only half used.

How to prompt it well

• Ask narrow, checkable questions. "What did this organisation announce and when?" is far more effective than "tell me about this" — retrieval rewards specificity.
• Constrain the sources. "Only use official documentation and primary sources" measurably changes what comes back.
• Ask for the date of each source. Web results mix years together; recency is the thing you most often need and least often get by default.
• Use a persistent space for an ongoing topic so related questions accumulate instead of starting cold each time.

The fundamentals — goal, context, constraints, format — work the same everywhere. Build one in the prompt lab →

Giving it context

• It carries less conversational context than a chat-first assistant — restate the subject rather than relying on "it."
• Ask follow-ups as new specific questions, not pronouns referring back.
• For a big topic, ask several narrow questions rather than one broad one.

Getting the output format you want

• Ask for a claim-per-line format so each citation maps to one statement.
• "Give me a table of claim, source, date" is well handled.
• Ask it to say explicitly when sources disagree with each other.

Files, documents, and research

Its research mode will assemble a longer report across many sources; scope it or it runs long. The workflow that works well is discover here, then verify or synthesise elsewhere — use it to find the primary sources, then read them, or move a fixed source set into a source-grounded tool for careful analysis.

Example prompts

Find five recent, credible sources on this topic, with publication dates and one-line summaries. I'll write from them.

What is this organisation's official position on this topic? Quote the primary source and link it. If there's no official statement, say so.

What's the current recommended way to do this task in this framework? Use only official docs, and give the doc link.

Give me a claim-by-claim briefing on this topic. One claim per line, each with its source and date. Mark any where sources disagree.

Here's a claim from a report I'm reading. Find the original source and tell me whether the report represents it accurately.

What approaches to this problem have organisations actually published about? Link each one.

What are the current requirements for this process? Official sources only, with the date each page was updated.

Common mistakes

• Reading the summary and never opening a citation.
• Assuming a cited source says what the summary claims, without checking.
• Using it for long creative drafting, which is not its shape.
• Not asking for dates, then acting on a page that's years old.

What it will not do well

• Not built for sustained creative or long-form writing
• Carries less nuanced conversational context across many turns than a chat-first tool
• Gives thinner answers when the question is vague or the web is sparse on the topic

Privacy and what you send it

A search-shaped product: your queries are the input, and account settings govern retention and whether queries are used for training. Treat queries the way you would treat search history, and check the controls before researching anything sensitive.

How to check its work

• Open at least the sources behind any claim you will act on.
• Check each source's publication date.
• Verify the cited page actually supports the specific sentence it's attached to — this is the most common failure.