// AI guide for professionals

How to write prompts that actually work

2026 · 07 · 28 — AI GUIDE FOR PROFESSIONALSHow to write prompts that actually work

You type “write me an email for a client” and ChatGPT hands you the same old mush: formal, empty, identical to everyone else's. You toss it, rewrite it by hand and conclude that “AI isn't for me.” Hold on a second: the problem is almost never the tool, it's the request. A prompt is a work assignment — and if you give a vague assignment, you get vague work. The same goes for a human colleague, after all.

Here are the 5 rules we use every day at MariVision, with before/after examples for emails, quotes and posts. No jargon: just the method. Ten minutes now, hours saved every week.

1 · Give context: who you are, who you're writing to, what you want to achieve

AI knows nothing about you until you tell it. It doesn't know you're a plumber in Bergamo, that the client has been waiting three weeks, or that you want to keep the job without over-apologizing. The first line of your prompt needs to give it three pieces of information: who you are, who you're writing to, what you want to achieve.

Same AI, same cost, a result in a completely different league. 70% of an answer's quality is decided right here, before we even talk about “better models.”

2 · Show an example of your tone (don't describe it)

“Write in a professional but friendly tone” is the most useless sentence in the history of prompts: it means everything and nothing. The shortcut that always works: paste in a real text of yours — an email that landed, a post that performed — and ask it to “write in this tone.”

One concrete example is worth more than ten adjectives. That's why the results suddenly improve when you stop describing the style and start showing it.

3 · Ask for the exact format: length, structure, limits

If you don't specify the format, AI decides — and it usually decides badly: too long, too much like a “school essay,” full of preambles. Spell it out yourself: how many lines, what structure, what must NOT be in it.

4 · Follow up: the first draft isn't the answer

The most common mistake: reading the first answer, wrinkling your nose and closing the chat. AI isn't a vending machine, it's a conversation. The first draft is only there to show you what to fix — and the fixes need to be targeted, not “do it better.”

Three targeted follow-ups beat any “perfect” prompt written on the first try. If you make the same request often, save the final version of the prompt in a notes file: it's your personal library, and it's worth its weight in gold.

5 · Always verify: the facts are your responsibility

The final rule, the one that saves you from embarrassment: AI makes things up with total confidence. Dates, prices, regulations, product names: when it doesn't know, it doesn't say “I don't know” — it improvises. Before a text goes out under your name, you check the facts: numbers, references, promises to the client.

In short

Context in the first line, a real example of your tone, an explicit format, targeted follow-ups instead of starting over, a final fact-check. Five rules, zero tools to buy: the difference between people who “tried AI” and people who use it every day comes down to this. And when prompts become routine — the same documents, every week — that's the moment to turn them into content and processes that run on their own.

Frequently asked questions