How to ask AI for what you need
An AI model only knows what is in your message and the conversation so far. Most disappointing answers come from a vague question, not a weak tool.
Five things to include
1. The task
Say exactly what you want done, in one sentence. “Summarise this” is vague. “Summarise this for a customer who has never heard of the product, in five bullet points” is clear.
2. The context
Who it is for, why you need it, and what you already know. The model cannot see your office, your customers or last week’s meeting unless you tell it.
3. The material
Paste the text, figures or notes it should work from. Ask it to use only what you have given it, so it doesn’t fill gaps from memory.
4. The format
Length, layout, tone and audience. “A 150-word email, friendly but direct” gets you something you can use straight away.
5. The limits
What to leave out, what to avoid, and what would make the answer wrong. “Don’t guess prices. If something is missing, say so.”
A before and after
Vague
You will get generic text you can’t use.
Specific
You get a draft close to what you need, to check and edit.
Ask it to check its own work
- “List any claim in this answer that you are not sure of.”
- “Which parts of this answer depend on information I did not give you?”
- “Give me the answer again, but only using the text I pasted.”
This doesn’t make the answer correct, but it shows you where to look. Anything that matters, such as a price, a legal point or a medical detail, still needs checking against a real source.
Keep going
The first answer is a draft. Tell it what to change, one thing at a time: “shorter”, “more formal”, “drop the second point”. A short follow-up usually beats starting again.
If it keeps missing, the problem is usually the context. Add the missing detail rather than repeating the same request louder.
What not to paste
Before pasting anything, check whether it is allowed. Customer records, passwords, health details and confidential contracts should not go into a tool unless your organisation has approved that tool for that data. Remove names and account numbers where you can.
Next: once you can ask well, see When to Use AI for where it helps, and AI Evals for how teams check it is working.