Perplexity runs a web search, reads the results, and writes an answer with a numbered citation on each claim. You use it well by asking a scoped question, setting a time window, telling it which kinds of sources count, and then opening the citations before you trust the summary. This guide gives you the prompt structure I use, five research prompts described by job, and the checking routine.
What Perplexity does differently
A plain chat model answers from memory and may invent a source when you ask for one. Perplexity searches first and answers second, so every sentence can point at a page that exists. The citations are the product. The answer is a reading guide to them.
That changes how you write the prompt. With ChatGPT you spend words on role and examples. With Perplexity you spend them on scope: what to search, how far back, which sources to prefer, and what shape the output should take.
It also has modes that trade speed for depth, including a longer-running research mode that reads many more pages and returns a report. Use the fast default for a quick fact. Use the deep mode when the answer needs many sources rather than a handful.
The research prompt structure
Six lines. I write them in this order and keep each one short.
| Line |
What it does |
Looks like |
| Question |
One question, scoped |
"How do the leading email marketing tools price their entry-level plans" |
| Time window |
Limits results to a period |
"Sources published in the last six months" |
| Source rules |
Says which sources count |
"Prefer official pricing pages and documentation; skip aggregator blogs" |
| Comparison |
Names the axes |
"Compare on monthly price, contact limit and included seats" |
| Output shape |
Sets the format |
"A table with one row per tool and a source in every cell" |
| Disagreement |
Says what to do when sources conflict |
"Where two sources disagree, show both and flag it" |
The last line is the one people skip and the one that saves you. Sources disagree constantly, and a model told to produce a clean answer will pick one silently. Told to flag it, it does.
Keep the question first. Perplexity builds its searches from your opening line, so a question buried under four lines of setup pulls worse results.
Five research prompt ideas
The full prompts are in the free guide, alongside the Deep Researcher mega-prompt for the long-form version. Here is the job each one does.
- Market snapshot. Builds a one-page picture of a category: size estimates with a source per number, the main players, and the trends the sources agree on.
- Competitor table. Compares named competitors on the axes you choose, with a citation in every cell and a note wherever the sources disagree.
- What changed since. Takes a topic you already know and reports only what has happened in the last 30 days, ranked by how many independent sources cover it.
- Evidence for and against. Takes one claim and finds the strongest sources on each side, then says which side has primary evidence and which has commentary.
- Pre-meeting brief. Profiles a company from public sources: what it sells, who runs it, its latest announcements and what its customers complain about, each line linked.
How to check the sources
Perplexity is honest about where its answer came from. That is the whole reason to use it, and it only helps if you look.
My routine, in order:
- Open every citation behind a claim you plan to repeat. A number headed for a deck gets a click.
- Check the date. An old source cited for a current fact is the most common failure I see.
- Check that the page says what the summary says. Models compress, and compression drops qualifiers. "Up to" disappears, "in one study" disappears, and a hedge becomes a fact.
- Check for circular sources. Three blogs citing each other count as one source. Look for the primary document underneath.
- Rerun the prompt with a stricter source rule when the first pass leans on secondary pages. "Only primary sources: filings, official docs, first-party announcements."
Keep a log: the prompt, the date you ran it, the links you checked. When someone asks where a number came from, the answer is in one place.
If you want the general principles behind these prompts, the prompt engineering guide covers the structure that holds across models, and the Perplexity section of the prompt library has more research prompts sorted by job.
What is inside the free Perplexity guide
The download has four parts.
- Perplexity Prompts and Research Engineering
- Research Prompt Templates
- Research Engineering Mini-Course
- Lifetime access and updates
