You've heard ChatGPT can save you time at work, but you're not quite sure where to start.

The answer fits in one idea: an AI assistant is excellent at turning text you already have (notes, emails, documents) into something polished. It's much less reliable at pulling facts out of thin air.

Here are five everyday work situations, each with a prompt to copy and what to double-check. They also work with Claude, Gemini or Le Chat by Mistral.

And if you're looking for a job instead, the same logic applies to your application: here's how to write a cover letter with ChatGPT that doesn't sound generic, and how to write a resume with ChatGPT without letting the AI invent your experience.

The golden rule before you start

The assistant prepares, you approve.

Nothing it produces goes out without your review: an assistant generates plausible text, not guaranteed facts. It can add a date, a commitment or a figure that isn't in your notes. To understand why, read why ChatGPT makes things up.

Case 1: emails (writing, replying, toning down)

The most rewarding use, in three variations.

Write an email from three lines

Write an email to a supplier to let them know Tuesday's delivery arrived incomplete (two boxes out of five are missing). I want to know when the rest will arrive. Tone: polite and direct, professional, five lines maximum. No threats, and don't mention penalties.

Reply to an email you received

Here is an email from a customer, between the tags. Suggest a reply that thanks them, confirms we're handling their request and says I'll get back to them by Friday. Don't promise anything else. <email> [paste the email, without names or contact details] </email>

Tone down an email written in the heat of the moment

Here is an email I wrote while angry. Keep exactly the same substance (I'm refusing the extra delay), but make the tone calm and professional. Don't add any apology I didn't write.

What to double-check:

  • added promises ("we will refund you," "starting tomorrow");
  • dates and amounts: are they really yours?
  • the tone: does it sound like you?

Case 2: meeting minutes from rough notes

Took messy notes, full of abbreviations and arrows? The assistant puts them in order.

You are in charge of writing a team's meeting minutes. From my rough notes below, write minutes in three parts: decisions made, action items (who, what, by when), open questions. Don't add any information that isn't in the notes. If an action item has no owner or no date, write "TBD". <notes> [paste your notes] </notes>

The "TBD" instruction is essential: without it, the assistant fills the gaps with a plausible date or name.

What to double-check:

  • each decision: was it really made, or only discussed?
  • who owns each action item;
  • nuances: a "we'll see" must not become "approved."

Case 3: preparing a meeting

Before an important meeting, the assistant can help you structure it and anticipate.

I'm preparing a one-hour meeting with my team to decide whether we switch project management tools. Attendees: five people, two of whom are attached to the current tool. Suggest a timed agenda, then five questions to ask to bring out the real objections. End with the three risks of deadlock I should anticipate.

A useful variation: "Play a skeptical attendee and ask me the three most awkward questions about this project."

What to double-check:

  • the timings: are they realistic for your team?
  • the questions: do they match the real stakes, or are they generic?

Case 4: summarizing a long document

A forty-page report, an internal memo, a set of specifications. The assistant can give you the essentials in a few lines, as long as you tell it what you're looking for.

Summarize the document below for a manager who has five minutes. Format: three summary sentences, then the points that need a decision from me. For each point, give the section of the document where it appears. <document> [paste the text or attach the file] </document>

Asking for the source section changes everything: you can check in ten seconds that the point really exists and hasn't been distorted.

What to double-check:

  • the points flagged as important, directly in the source document;
  • what's missing: a summary can skip a short but decisive clause;
  • figures, always, against the source.

Case 5: rewording

Text that's too technical for a customer, a procedure that's too long, an announcement that's too curt: this is where the assistant excels.

Rewrite this paragraph so a non-specialist customer can understand it. Keep all the information, remove the jargon, use short sentences. Give me two versions: one neutral, one warmer.

What to double-check:

  • the meaning: simplifying can turn "usually" into "always";
  • contractual or regulatory terms, which are sometimes best left as they are.

What you never paste

This is the most important part. What you type into an assistant goes to an external service and, depending on the tool and your plan, may be stored or used to improve the service.

Never paste:

  • confidential information: unpublished figures, ongoing projects, contracts, strategy, internal source code;
  • personal data about customers or colleagues: names, addresses, phone numbers, HR files, health data;
  • credentials: passwords, access keys, private sharing links.

And above all: read your company's rules on using AI. If there aren't any, ask your manager or the IT department.

The right reflex: anonymize. Replace names with [customer], [colleague A], amounts with [amount], product names with [product]. The assistant doesn't need to know who the customer is to write a good email.

The checklist before sending

  • I've removed all confidential or personal data.
  • I've checked every figure, date and commitment.
  • I stand behind every sentence as if I'd written it myself.

If your results stay too generic, the problem often lies in the request. The six-move brief method helps you write prompts that work the first time.

Going further

To use ChatGPT well at work, you need to understand what a model generates and what it actually knows, and why to be wary of a suspiciously precise figure. That's the topic of the first chapter of the Lantern island, free and playable in your browser with no account: Light the lantern. And to test your reflexes with your colleagues, start the friend challenge: ten timed questions about AI, and it's up to you to beat their score.

For personal and administrative emails (complaints, cancellations, requests to your landlord), the guide to writing an email with AI gives you a prompt template and a checklist before you hit send.

If your work involves spreadsheets, our guide on how to use ChatGPT for Excel formulas shows how to get a formula, understand it and test it on a copy before trusting it.

Before you paste a work document, remove what the assistant doesn't need: the free game Safe to paste? trains you to spot names, contact details and confidential details in a text, and gives you a redacted version to copy.

And to fit all of this into a busy week, see how to plan your week with ChatGPT: a clear to-do list, real priorities and a plan that respects your fixed commitments.

Frequently asked questions

Am I allowed to use ChatGPT at work?

It depends on your company. Check whether there's a policy on AI use or a tool approved internally, and when in doubt, ask your manager.

What information should you never give ChatGPT?

Your company's confidential information, personal data about customers or colleagues, and credentials. Anonymize by replacing names and amounts with [customer] or [amount].

Can you send an email written by ChatGPT straight away?

No, always reread it. The assistant can add a promise, a date or a figure that wasn't in your request, and you're the one responsible for what you send.

Can ChatGPT write reliable meeting minutes?

Yes, from your notes, if it's told not to add anything and to write "TBD" when information is missing. Then reread each decision before sharing.