Perplexity Prompts: Search Better and Verify Your Sources

Type "VAT micro-entreprise" into Perplexity and you'll get a general summary, with a few links. Ask this instead:

I'm a micro-entrepreneur in France, selling graphic design services to individuals and businesses. My revenue is approaching the VAT exemption threshold (franchise en base de TVA).

What happens in practice if I exceed it during the year: from when do I have to charge VAT, and what steps do I need to take?

Rely primarily on impots.gouv.fr, service-public.fr, and official sites. Give the last-updated date of the pages you cite. If the rules have changed recently or if sources disagree, say so.

You don't get the same answer because you didn't ask the same question. The first asks for a topic, the second asks for an answer to a situation.

That's the principle to remember. Perplexity queries the web in real time, then synthesizes what it finds with numbered citations. It behaves like a search engine that writes its own answer. A prompt for it is best thought of as a question put to a research librarian, and less like a text order.

Give your situation, not just your topic

Perplexity's tips are simple: ask a complete question rather than a single word, add your situation, include the specific terms that matter, and say what form you want the answer in. It also recommends writing as you would to a colleague and making just one request per prompt.

If you ask in one go to compare three software tools, check their compliance, and draft an email to your director, the search scatters. Split those requests into several questions within the same session: follow-up questions keep the conversation's context.

You don't need everything every time. Above all, add the location and the period whenever the topic touches law, taxes, or health: those are the two details that most easily skew an answer. You can add the rest in a follow-up if the result stays too broad.

Treat sources as witnesses, not proof

The citation number next to a sentence gives an impression of rigor that can mislead. It indicates that Perplexity used that page. It doesn't say the page is reliable, or up to date, or that the sentence written actually reflects its content.

Three quick checks cover most of it. Click the source behind the most important claim and look for the sentence yourself. Look at who's speaking: a software vendor's blog that ranks its own software first is not an independent test. Look at the date. If several citations all point back to the same information picked up from a single article, you actually have only one source.

You can also have Perplexity do part of this sorting. Here's a prompt to send as a follow-up after an important answer, useful for example for a trainer preparing material on the right to training:

Take your previous answer and critique it:

1. For each factual claim, indicate the type of source that supports it: official text, public body, media, commercial site, or forum.
2. Flag the claims that rest on a single source, or on sources that copy each other.
3. Actively look for sources that contradict or qualify your answer, and present them.
4. Indicate any information that is more than [DURATION] old or whose date isn't visible.

End with what I should check myself before relying on it.

This prompt doesn't replace your own check, because the tool can also be wrong in its own critique. It simply tells you where to look first. The detailed method is in the article on verifying an AI answer.

For topics where tools diverge, be aware that source-selection settings vary by plan and interface: some presentations talk about Focus, others about choosing sources. At the time of writing, look at the options under the search bar. When you don't have any, you can simply write the type of sources you want in your question.

Frame Deep Research with a narrow question

Pro Search now applies automatically when your plan includes it. Deep Research, for its part, produces a longer report. Perplexity is in the process of integrating it into the skills of Computer, its tool that carries out tasks, with a gradual rollout: depending on your device and plan, you'll find it in the search menu or among those skills.

Perplexity's advice is to ask a targeted question, since a narrow question gives a targeted report. You can start broader and refine, and you can also attach your own files to have them analyzed alongside web sources.

Weak: "Write a report on selling local products online."

Better, for a honey producer who wants to sell remotely:

I produce honey in [REGION] and I want to sell online to consumers in France starting [DATE].

Compare three options: my own online store, a general marketplace, and a direct-from-producer sales platform.

For each option, I want to know:
- the fees and costs to plan for, with the date of the information
- the actual management time, based on feedback from producers
- the shipping constraints for glass jars
- the labeling requirements to comply with

My attached file contains my volumes and prices: use it to estimate what I'd have left per jar in each case, and show the detail of your calculations.

Present a summary table, then a report of two pages maximum. Separate sourced facts from your estimates.

The report will give you a working base, not a decision. Recalculate the margins yourself and confirm the labeling requirements with a competent body.

Keep recurring research in a project

Perplexity's Spaces have become Projects. A project groups your sessions, your files, and shared instructions, and it can be shared with others. It's useful if you keep coming back to the same area: industry monitoring, preparing a thesis, tracking a competitor. Features and access conditions change by plan, so check in the tool.

In the project's instructions, state what applies to all searches: your profession, your country, the sources to favor, and the answer format. A doctoral student might write: "I work on the history of education in France. Favor scientific publications and archives, and always give the author and year. Flag secondary sources." For scientific publications, always check that the reference exists and that it says what it's being made to say. It's a classic case of error, covered in the article on reducing hallucinations.

If your work rests on long documents you already own, Perplexity isn't necessarily the best starting point: the guide on long documents shows another way to go about it.

Take the last search you did on Google for a work decision. Rephrase it by adding who you are, the location, the period, and the type of sources, then open the main source cited in the answer and look for the sentence used. In two minutes, you'll know whether you can trust the result.