How to give AI good context and goals

AI can feel like a slot machine: one time you get something useful, the next a generic story you can do nothing with. That difference rarely sits in the tool. It sits in the request you make. Anyone who learns to convey their goal and context well gets structurally better answers, no matter which AI they use.

This article is about how you do that. It builds on the basics of using AI in your daily work and goes a step deeper, into the work that really matters.

Why context and goals make the difference

An AI only knows what you put into it. It does not know your customer, your situation, or what you want to achieve with the text. Ask something vague and the AI fills the gaps itself with something generic. Ask in a targeted way, with your goal and the right background, and the answer can suddenly be usable.

This is good news, because it means you are the one at the controls. You do not need to understand the technology. You only need to get clear on what you want and convey it well. That is a skill you can pick up after practising a few times.

Tell the AI five things

A useful request usually contains these five parts. Not as a mandatory list, but as a checklist in your head.

1. Your goal. What do you ultimately want to achieve? "Convince a customer", "inform a colleague", "quickly understand something yourself".

2. The context. The background the AI cannot know: which situation it concerns, who it is for, what already exists.

3. The constraints. The boundaries: how long, what to mention and what not, which tone, when you need it.

4. The form. How you want it back: an email, a bullet list, a table, three paragraphs.

5. An example. If you think something is good, show it. "In the style of this text" works better than a description.

You do not have to name all of them every time. For a quick question, two sentences are enough. For work that matters, all five are worth the effort.

Give your goal, not just your question

The most common mistake is that people name a task but not their goal. "Write an email to this customer" is a task. "Write an email to this customer so he feels heard and still books an appointment" is a goal.

That difference steers everything. The AI makes different choices in tone, length and arguments when it knows where you want to go. So tell it not only what you want done, but why, and what a good result looks like for you.

Give context the AI cannot know

The AI misses everything that sits in your head or in your systems. That context is exactly what makes the answer personal and usable. Think of:

• Who the recipient is and what your history together is.

• What came before: an earlier conversation, an earlier email, an ongoing project.

• What already exists: documents, figures, an earlier version you can hand over.

The more relevant background you share, the less the AI has to guess. If you work with sensitive data, pick a version and setting up front where your data is not used for training.

Work in rounds

Do not expect the first answer to be right straight away, and it does not need to be. You steer with short instructions: "shorter", "less formal", "do not mention the amount", "start with the conclusion". Each round it gets closer to what you mean.

This is exactly how you would work with a colleague. You give an assignment, you look at the result, and you explain what should be different. The difference is that the AI does not get tired on the fifth round.

An example: from vague to usable

A vague request: "Write something about our new service for the website."

What the AI gives back is predictably generic, because it does not know what the service is, who it is for, or what it should set in motion.

The same request with goal and context: "We offer a new service where we help SMEs put their first AI application live within four weeks. Write a short introduction for our website, aimed at business owners who are curious but have no technical knowledge. The goal is for them to request a call. Down to earth in tone, a hundred words at most, no sales language."

The second one produces something you can almost publish. The difference is not in the AI, it is in what you gave it.

Common mistakes

• Too short. "Make a plan" without saying what for, who for and within which constraints.

• Everything at once. One enormous request with ten wishes at the same time. Break it up and build in rounds.

• Giving no example while you do have a clear picture of what you consider good.

• Leaving out the goal. The AI then does not know what to work towards.

How we can help

This is exactly what you practise in the workshop Claude as your daily work partner. You work with your own assignments and notice in one afternoon the difference between a vague and a sharp request. You learn it on your own work, so you can apply it the very next day.

If you also want AI to take on bigger jobs, such as carrying out tasks itself or drafting an approach for a project, it helps to get this right first.

Read more in this guide: how to use ChatGPT or Claude in your daily work and what can AI do beyond chatting.

Frequently asked questions