How to ask AI for what you actually need

No jargon, no magic phrases: the four things that change an AI answer, and a template to copy and fill in.

Text drafted with artificial intelligence. The topic, the angle and the final review are Ibrahin's. How AI is used here

What has happened

“Prompt engineering” has become fashionable, with courses and lists of phrases that supposedly make AI answer better. A prompt is just the request you type. And OpenAI's official guidance doesn't talk about magic words: it recommends clear, specific requests, improved in rounds.

Why it matters to you

Because if you think the problem is technique, you end up copying long formulas that don't say what you need. What changes the answer is four pieces of information you already have in your head, and you don't need to know about technology to give them.

What you can do

1. Say what you want to end up with

Not “use your full potential”, but an email ready to review, a shopping list with criteria, an outline for a meeting or a plan with a clear first step. If you don't know yet what you want, ask it first to help you pin down the problem.

2. Give the context that changes the answer

It’s not telling your life story: it’s saying who will read it, what they know, how much time or money there is, what has been tried already and which limits can’t be crossed. Compare:

Write a guide to backups.

Write a guide for someone who keeps their photos on their phone and has never made a backup. They're short of time and need to know what they could lose if they don't check it.

The second isn’t more technical. It’s more specific about the person and what they need.

3. Ask for a format you can check

A three-sentence summary, numbered steps, risks and doubts at the end, a table only if there are comparable options, or a list of claims that need a source. And an approximate length: “between 700 and 900 words” helps more than “not too long”.

4. Explain what “done well” means to you

Clear language, no invented data, a warning when information is missing. If it's your own text: “don't change my opinion or add experiences I haven't had”. And if there are several rules, rank them: if you ask for it to be both exhaustive and very short, it will have to break one.

5. Let it ask you questions, with a limit

Before writing, ask me at most three questions that would change the answer. If a piece of information isn't essential, carry on with a clearly marked assumption.

6. Improve one thing at a time

The first answer is a draft. Review it in parts: does it address the problem? Is it clear what to do first? Is there data to check? Is the tone what you want? And ask for specific changes: “remove the grandiose sentences and keep the practical examples” works better than “make it more human”.

7. Check what matters

That it answered what you asked, respected your limits, didn't add names, figures or dates you didn't give, and that the links exist and say what it claims. For important decisions, don't use the AI itself as your only check.

The recommendation

Copy this template, fill it in and try it with public or made-up information:

I want [result]. I need it for [person or situation]. The essential context is [information]. Give it back as [format and length]. It will be useful if [quality criterion]. Don't make up [limits] and flag anything still to be checked.

Don’t include passwords, keys, bank details, medical records or other people’s information.

Sources

Best practices for prompt engineering in ChatGPT, OpenAI Help Center, checked on 2026-09-02.

Review

Responsible for publication: Ibrahin. Date: 2026-09-02. Last checked: 2026-09-02.