What makes a good prompt for LeChat (Mistral)?
LeChat is the AI assistant from Mistral, a French company. That detail matters more than it sounds: LeChat is built in Europe and is particularly strong in European languages — French first and foremost, but it handles English and multilingual tasks with ease. Here's how to get the best out of it, following on from our articles on ChatGPT and Claude — the basic principles are similar, but each assistant has its own strengths.
LeChat's personality
Two things set LeChat apart.
A genuinely multilingual, Europe-first assistant. Built by a French team, LeChat is especially at home in French and other European languages, with natural phrasing and an appropriate register. If you work across French and English, or need text aimed at a European audience, that's a real asset — including for translation between languages.
A European approach to data. Mistral is a European company, subject to European data-protection rules. For individuals and for professionals who care about confidentiality and digital sovereignty, that's an argument that carries weight.
LeChat is also a direct and efficient assistant: it responds well to clear, concise instructions, with no need for long-winded preambles.
The basics, still valid
As with the other assistants, a good prompt rests on five ingredients: be specific, give context, assign a role, state the format you want and, if needed, provide an example. If you've already mastered these reflexes, you already know how to talk to LeChat. Let's look at what suits it particularly well.
What LeChat appreciates: clear, structured instructions
LeChat responds very well to organised requests. When you have several criteria, lay them out neatly rather than in one block:
Write a listing to sell my second-hand bike.
- Tone: friendly and honest
- Length: around 80 words
- Must mention: Decathlon brand, very good condition, sold with a lock
- Avoid: technical jargon
This formatting helps the AI forget nothing and respect each instruction. It's true for every assistant, but LeChat takes to it particularly well.
What LeChat appreciates: putting its language skills to work
Don't hesitate to ask it for a precise register, which it handles finely:
- "Write it in formal language, for an official letter."
- "Rephrase this paragraph in a simpler, more accessible style."
- "Fix the mistakes and improve the style, without changing the meaning."
It's also a good tool for translating to or from French, or for adapting a text to a particular audience.
A before / after example
Weak prompt:
Write me an out-of-office message.
Result: a one-size-fits-all message, possibly clumsy or too generic.
Effective prompt:
You're my writing assistant. Write my automatic out-of-office reply for my work emails. Context: I'm on leave from 1 to 15 August, my clients are business professionals. I want to stay warm but professional, state the return date, and give a colleague's contact (Marc Dupont, marc@example.com) for urgent matters. Format: 4 to 5 lines, in a polished tone.
Result: a complete message, in the right register, ready to use.
Good habits
- Make the most of its language range. Ask explicitly for a register (casual, neutral, formal); LeChat renders it well.
- Structure your instructions as a list when there are several.
- Stay clear and direct. No need to beat around the bush: a crisp instruction is enough.
- Verify what matters. Like any AI, LeChat can be wrong about a fact or a figure. Cross-check what counts.
In short
LeChat is Mistral's assistant, with two advantages worth noting: strong multilingual, Europe-first language skills and a European approach to data. To steer it well: clear, structured instructions, a specified register, and the trusty trio of specificity, context and format. The result: natural, polished text, with little effort.
And that completes our series: you now know what generative AI is, what an LLM is, and how to talk well to ChatGPT, Claude and LeChat. What do these three assistants have in common? They always do their best when your instruction is clear. A good prompt is 90% of the result.