GinfoCoach.
← All articles
Newbie Part 1 of 5 3 min read

Your first good prompt: talk to AI like a brilliant new colleague

Why "write an email" gets you a bland answer, and the four questions that turn any request into a prompt that works.

Imagine a brilliant new colleague on their first day. They are fast, well read and eager to help. But they have never met your team, they don't know your project, and they can't read your mind. If you hand them a sticky note that says "write an email to the team", what will you get back?

Something generic. Polite, correct, and almost useless.

That is exactly what happens with an AI model. A prompt is simply the message you give it, and the model can only work with what you put in that message. Most disappointing AI answers are not the AI's fault. They are the result of a sticky-note prompt.

The sticky-note problem

Here is a typical first prompt:

Write an email to my team about the new project.

The model now has to guess:

  • Who is on the team, and what do they already know?
  • What is the project, and why does it matter?
  • What should people do after reading the email?
  • Should it be formal or friendly? Three lines or a full page?

Every guess is a coin toss. And when a model guesses, it picks the most average option, which is why the answer feels bland.

The four questions

Before you send any prompt, ask yourself four questions. You don't need to answer all of them every time, but the more you answer, the better the result.

  1. Who is the answer for? (Your manager, a customer, a ten-year-old, yourself.)
  2. What exactly do you want? (One clear task, with the facts the model needs.)
  3. Why do you need it? (The goal behind the task changes the best answer.)
  4. How should it look? (Length, tone, format.)

Let's apply them to our email:

Write a short, friendly email to my team of five designers.
We start a new project on Monday: redesigning the mobile app
for a bank client. I want everyone to block two hours on
Monday morning for a kick-off workshop. Keep it under 120
words and end by asking them to confirm by Friday.

Same request. Completely different result, because the model no longer has to guess.

Write it the way you would brief a person

A useful test: would a smart human understand exactly what you want from this message alone? If a colleague would need to come back with questions, the model needs those answers too. It just won't ask; it will guess.

This also means you can write naturally. You don't need special keywords, magic phrases or capital letters. Clear, plain sentences work best.

Three habits to start today

Be direct. Start with what you want: "Summarise…", "Write…", "Compare…". Avoid vague openers like "Can you maybe help me with something about…".

Give the facts only you know. The model knows a lot about the world, but nothing about your team, your client or your deadline. Those details are what make the answer yours.

Read the answer critically. A confident tone is not proof of accuracy. Check names, numbers and facts, especially anything you will send to someone else.

Why writing it yourself matters

It is tempting to let the AI do all the thinking. But a study by Microsoft Research and Carnegie Mellon University, which surveyed 319 knowledge workers in 2025, found that the more people trusted the AI, the less critical thinking they reported putting into their work, while people who were confident in their own skills kept thinking critically.

Writing a good prompt is thinking. Answering "who, what, why and how" forces you to clarify your own goal before you ask for help. That is why GinfoCoach asks you to write first and only then shows you a better version: the skill stays yours.

Try it now

Take something you asked an AI recently. Rewrite it with the four questions in mind, then run both versions through the coach and compare the scores.


Sources