This is the one-page reference for prompting Perplexity — the answer engine that grounds every response in live web sources with inline numbered citations. You don't prompt it like a chatbot; you brief it like a research assistant. State what to find, how deep to go, where to look, how recent it must be, and how to format the answer. Below is the 5-part formula and copy-paste tables for the modes, source toggles, models, and modifiers that reliably move the output.
New to it? Start with the best Perplexity prompts roundup, then use this sheet while you write your own. For a deeper walk-through, see how to prompt Perplexity for research.
The prompt formula
A strong Perplexity prompt is one clear request built from five parts. Cover the task and scope first; sources, timeframe, and format sharpen the answer and make it verifiable. The order is a guide — Perplexity reads the whole request and plans its searches around it.
- Task — the exact job: compare, summarize, find, brief, list, build.
- Scope — the boundaries: which products, regions, segments, or questions are in and out.
- Sources — where to look and what counts (web, academic, primary sources, peer-reviewed only).
- Timeframe — how recent it must be ("last 12 months", "as of July 2026").
- Format — how to return it (comparison table, executive brief, bullet summary with sources).
Skeleton to copy:
[Task: compare / summarize / find / brief] [scope: exactly what and its
boundaries], using [sources: web / academic / primary sources] from
[timeframe]. Return it as [format], and link every claim to a citation.The same skeleton, filled in:
Compare the three leading open-source vector databases for a startup RAG
stack, focusing on self-hosting cost, query latency, and hybrid search,
using technical docs and primary benchmarks published in the last 12 months.
Return a comparison table with a source column, link every claim to a
citation, and flag any conflicting numbers.Why it works: the task is explicit, the scope is bounded to three tools and three criteria, the sources and timeframe are named, and the format plus citation instruction make the answer easy to verify. For a step-by-step build, see Perplexity research prompts.
Modes & when to use them
Perplexity has several depth tiers and tools. Pick the mode before you write the prompt — it decides how many sources get read and whether you get an answer or a deliverable.
| Mode | Best for |
|---|---|
| Search | Quick, cited answers to a single question — reads a handful of sources and replies in seconds |
| Research (Deep Research) | Multi-step questions — runs many searches, reads dozens of sources, and returns a structured, fully cited report |
| Labs | Building deliverables — spreadsheets, dashboards, slide decks, and simple web apps from your research |
| Comet | Agentic browsing — the AI browser that navigates sites and completes multi-step tasks for you |
| Model Council (Max) | Highest-stakes questions — runs three frontier models and synthesizes their answers into one |
Rule of thumb: Search for facts, Research for reports, Labs for files. Reach for Model Council only when an answer is worth the extra time and cost.
Sources & focus
Source focus restricts where Perplexity looks, which cuts noise and improves citation quality. Set the toggle, or name the focus directly in the prompt ("search only academic sources").
| Focus | What it searches | Best for |
|---|---|---|
| Web | The open internet — news, blogs, docs, official sites | General questions, current events, product research |
| Academic | Published papers and peer-reviewed journals | Evidence, literature reviews, citing primary studies |
| Social | Reddit and X — real-user opinions and discussion | Sentiment, honest reviews, "what do people actually think" |
| Finance | Tickers, analyst ratings, and SEC filings | Market data, earnings, company and stock research |
Models
The model decides how hard Perplexity reasons. Sonar Pro is the default and covers most cited answers; step up only when the question needs it. Frontier models and Model Council are on the Pro and Max plans.
| Model | Strength |
|---|---|
| Sonar | Fast, lightweight cited answers — good for quick lookups and high volume |
| Sonar Pro (default) | The balanced everyday model — solid reasoning and citations for most research |
| Sonar Reasoning Pro | Multi-step logic and harder analysis — comparisons, math, chained reasoning |
| Frontier models (Pro/Max) | GPT-5.2, Claude Opus 4.6, or Gemini 3 Pro — the toughest writing and reasoning |
Tip: pairing Sonar Reasoning Pro with Deep Research handles nearly every serious research task without touching frontier models.
Citation & source modifiers
These phrases tell Perplexity what counts as a good source and how to handle evidence. Drop one or two into any prompt to raise citation quality — they pair especially well with the Academic and Finance focus.
| Phrase to add | Effect |
|---|---|
| "cite only peer-reviewed sources" | Restricts evidence to published, reviewed research — filters out blogs and marketing |
| "use primary sources, not summaries" | Pulls original filings, papers, and data instead of second-hand reporting |
| "link every claim to a citation" | Forces a numbered source next to each statement so nothing is unsupported |
| "flag any conflicting sources" | Surfaces disagreement between sources instead of averaging it away |
| "as of [date]" | Anchors the answer to a point in time and pushes for the most recent data |
| "list the limitations of this answer" | Adds a short honesty section — gaps, assumptions, and where evidence is thin |
| "quote the exact figure and its source" | Ties every number to a verbatim source line so you can check it fast |
Output & scope modifiers
These phrases shape the format and boundaries of the answer. State one clearly so Perplexity returns something you can use as-is rather than a wall of text.
| Phrase to add | Effect |
|---|---|
| "return a comparison table" | Side-by-side rows and columns — ideal for tools, products, or options |
| "write a one-paragraph executive brief" | Tight, decision-ready summary for a busy reader |
| "bullet summary with a source after each point" | Scannable takeaways, each backed by its own citation |
| "only the last 12 months" | Caps the timeframe so stale results drop out |
| "focus on [region] only" | Limits scope to a market or geography — US, EU, India, and so on |
| "go deep: dozens of sources" | Signals Deep Research breadth over a fast surface answer |
| "keep it to five key points" | Forces prioritization instead of an exhaustive dump |
Copy-paste example prompts
Five complete prompts that assemble the modifiers above. Paste any into Perplexity and tweak the bracketed parts. For more, browse the best Perplexity prompts.
1. Quick cited answer (Search)
What is the current EU regulatory status of the AI Act's high-risk system
rules, as of July 2026? Give a five-point bullet summary with a source after
each point, use primary sources like official EU publications, and flag any
conflicting sources.Best for: getting a fast, verifiable answer to a single factual question without a full report.
2. Deep Research report (Research)
Run a deep research report on the state of the electric cargo bike market for
last-mile delivery. Scope it to North America and Europe, cover market size,
top manufacturers, unit economics, and regulation, using web and primary
sources from the last 12 months. Return a structured report with sections,
link every claim to a citation, list limitations, and go deep across dozens
of sources.Best for: market scans and literature reviews where breadth and full citations matter. See the research guide for the full workflow.
3. Comparison table
Compare the top five project-management tools for a 20-person remote startup,
focusing on price per seat, integrations, automation, and mobile app quality.
Use official docs and recent hands-on reviews from the last 12 months. Return
a comparison table with a source column, quote the exact price and its source,
and flag any conflicting numbers.Best for: buying decisions — a side-by-side table you can drop straight into a doc.
4. Finance / ticker brief (Finance focus)
Give me an executive brief on NVDA as of today, using the Finance focus.
Cover the latest quarterly results, current analyst ratings and price
targets, and any recent SEC filings. Write a one-paragraph executive brief
followed by a bullet list of key figures, quote the exact figure and its
source for each, and flag conflicting analyst views.Best for: a fast, cited snapshot of a stock before a deeper dive. Always click through the citations to confirm the numbers.
5. Labs deliverable (Labs)
Build a spreadsheet comparing the ten most-used open-source LLM serving
frameworks across throughput, GPU support, quantization, license, and GitHub
stars. Research each from official docs and primary benchmarks in the last 12
months, add a source column linking every value, and include a short summary
tab with the top three picks and their limitations.Why it works: the task names the exact artifact, the columns are specified, and the source and limitations instructions keep the built file honest — exactly what Labs is for.
Frequently Asked Questions
How is prompting Perplexity different from ChatGPT?
Perplexity is an answer engine, not a chatbot — every response is grounded in live web sources with inline numbered citations. So you prompt for research, not creativity: state the task, the scope, which sources to search, the timeframe, and the output format. Skip persona role-play and long system prompts; instead be specific about what to find, how recent it must be, and how you want it summarized so you can verify each claim against its citation.
When should I use Research mode instead of Search?
Use Search for a quick, cited answer to a single question — it reads a handful of sources and replies in seconds. Switch to Research (Deep Research) when the question is multi-part or high-stakes: it runs multiple searches, reads dozens of sources, reasons across them, and returns a structured, fully cited report. Deep Research takes a few minutes but is worth it for market scans, literature reviews, and comparisons where breadth and citations matter.
What do the source focus toggles actually do?
Each focus restricts where Perplexity looks. Web searches the open internet; Academic searches published papers and journals; Social pulls opinions and discussion from Reddit and X; and Finance searches tickers, analyst ratings, and SEC filings. Setting the right focus cuts noise — pick Academic for evidence, Social for real-user sentiment, and Finance for market data — and you can also name the focus in the prompt, e.g. "search only peer-reviewed academic sources."
Which model should I pick in Perplexity?
Sonar Pro is the default and handles most cited answers well. Use Sonar Reasoning Pro for multi-step logic and harder analysis. On Pro and Max plans you can route to frontier models — GPT-5.2, Claude Opus 4.6, or Gemini 3 Pro — for the toughest writing and reasoning. Max also unlocks Model Council, which runs three frontier models and synthesizes their answers into one. For most research, the default plus Deep Research is enough.
How do I get better citations and fewer weak sources?
Tell Perplexity what counts as a good source directly in the prompt. Add lines like "cite only peer-reviewed sources," "use primary sources, not summaries," "link every claim to a citation," and "flag any conflicting sources." Combining a source-quality instruction with the Academic or Finance focus gives the cleanest, most verifiable answers. Always click through the numbered citations to confirm the underlying source actually says what the answer claims.
What is a Space and when should I use one?
A Space is a persistent workspace where you can upload files and set custom instructions that apply to every question you ask inside it. Use a Space when you have an ongoing project — a company you're researching, a codebase, a set of PDFs — so you don't have to re-explain context each time. Your uploaded files become part of the searchable context alongside the live web, which keeps long research threads consistent.
Can Perplexity build spreadsheets or dashboards, not just answers?
Yes — that's Labs. While Search and Research return written answers, Labs builds deliverables: spreadsheets, dashboards, slide decks, and simple web apps from your prompt and its research. Ask for the artifact explicitly, e.g. "build a spreadsheet comparing these ten tools across price, features, and rating, with a source column." Labs runs longer than a normal query because it's assembling a file, not just replying.
Does Perplexity ever make up numbers or citations?
Perplexity is designed to ground every claim in a real, linked source and avoids inventing figures, but no answer engine is perfect. Reduce risk by adding "as of [date]," "link every claim," and "flag conflicting sources" to your prompt, and by using Research mode for anything important. Then verify: open the numbered citations and confirm the numbers match the source before you rely on them.