How to ask AI well – the basics of prompting

What does "prompting" mean?

The text you type to an AI is called a prompt. Prompting simply means how you ask an AI for things. You don't need special commands or code. You just need to explain clearly what you want, as you would to a new colleague.

Think of it this way: if you ask a neighbour to "get something from the shop", they won't know what to bring. If you say "could you pick up a litre of skimmed milk and a rye loaf, the round kind", you'll get exactly what you wanted. AI works the same way. It can't read your mind, but it does a surprisingly good job once it knows what you're after.

Four parts that make a good question

A good prompt can be built from four parts. You won't need all of them every time, but the more you include, the better the answer tends to be.

1. Role – who is the AI? "Act as an experienced bookkeeper", "You are a patient computer teacher" or "You are an adviser who knows marketing". The role steers its vocabulary and point of view.
2. Context – what is the situation? Say who you are and what this is about. "I'm 62 and moving from a house to a flat." Context is the most important part, and it's exactly what most questions leave out.
3. Task – what do you want? Use a clear action word: write, explain, compare, summarise, suggest, check, list.
4. Format – what should the answer look like? "Answer in five bullet points", "no more than 100 words", "as a table", "in plain English" or "polite but firm".

Before and after: three examples

Example 1: An email

The first version produces a generic, empty message. The second produces an email you could send almost as it is.

Example 2: An explanation

Notice the last sentence: you ask the AI itself to tell you where to verify the information. Tax amounts and rules change, so always check them with your tax authority.

Example 3: A post for a small business

Asking for several options lets you pick the best one or mix and match.

Follow-up questions: the most important skill nobody talks about

Many people give up if the first answer misses the mark. That's a shame, because an AI's biggest strength is the conversation. The first answer is a draft, not the end result. You can rework it as long as you like. Good follow-ups include:

• "Cut it down by half."
• "Make the tone warmer / more formal / more relaxed."
• "Rewrite it so my 80-year-old mother would understand it."
• "What did you leave out? What are the risks here?"
• "Explain your reasoning and tell me where I can check this."
• "That's not what I was after. I meant…"

You can also turn things round: ask the AI to question you. Add to the end of your prompt: "Ask me clarifying questions first if you need more information." The AI will interview you and can then help much better.

A ready-made template to reuse

Copy this template into the notes app on your phone. Fill in the parts in square brackets and paste it into the AI:

Once you've used the template a few times you won't need it any more – you'll start writing like this naturally.

The most common mistakes

• Too short a question. "Recipe" gets you anything. "Quick weeknight dinner for two with minced beef, no cream" gets you something useful.
• Too many things at once. Ask for a travel plan, budget and packing list in one message and the answer will be shallow. Ask for one thing at a time.
• Assuming the AI knows your situation. It doesn't know where you live, what your business does or what you meant in another chat yesterday – unless a memory feature is switched on.
• Trusting the first answer. Always check facts, especially figures, dates and legal points. See what AI can't do.

Summary

A good question gives a role, context, a task and a format. Context matters most: describe your situation as you would to a new colleague. Don't settle for the first answer – keep the conversation going and ask for changes. Keep a template handy and you'll get far more out of AI, saving time every time.