Writing Instructions in English: The Imperative Behind Every Prompt
A prompt is an instruction, and instructions run on the imperative. How to write clear commands in English that a model — or a person — can follow.
Every prompt is an instruction, and instruction in English has one grammar: the imperative. It is the base form of the verb with no subject — write, summarise, explain, list. You already know it from recipes, signs and directions. Prompts simply use the same form for a machine.
The imperative
The subject is always the implied you, so it is dropped:
- Write a summary. (not
You write a summary.) - Translate this sentence.
- List three examples.
To make it negative, put don't (or do not) in front:
- Don't use jargon.
- Don't translate the technical terms.
The imperative is what makes a prompt read as an instruction rather than a question or a description. Compare:
- ❌ Can you make a list of the benefits? — a request, which the model may treat as optional.
- ✅ List the benefits. — a command, which the model treats as the job.
Politeness is wasted words with a machine. Could you please write a summary? works, but Write a summary. works just as well and is shorter. A model does not need to be softened up; it needs a clear task. Save please and could you for people.
Being specific
An imperative on its own is rarely enough. Summarise it. — summarise what, at what length, for whom? The instruction needs its three pieces:
- The verb — what to do.
- The object — what to do it to.
- The constraint — how, for whom, how much.
| Weak | Strong |
|---|---|
| Summarise it. | Summarise the article in three bullet points. |
| Explain AI. | Explain what a token is to a non-technical reader. |
| Write an email. | Write a polite email refusing the meeting and suggesting next week. |
Each strong version names the object precisely and adds a constraint. The constraint is where most prompts go wrong — it is easy to say write an email and hard to say polite, refusing, suggesting next week. But the second version is the one that gets you the email you actually wanted.
Choosing the right verb
The verb sets the shape of the answer. These four cover most everyday tasks, and each produces a different kind of output:
| Verb | Output shape |
|---|---|
| write | a text built from nothing |
| summarise | a shorter version of something given |
| explain | a text that makes something clear |
| rewrite | a new version of something given |
Swapping the verb changes the job. Write a summary of this and Summarise this both ask for a summary, but write implies composing, while summarise implies condensing. For most "make this shorter" work, summarise is the honest verb.
Precision with numbers and formats
Constraints work best when they are measurable. A model cannot guess "short", but it can follow "under 100 words", "three bullet points", "in a table", "for a B1 reader".
- List five risks. — a number the model can count.
- Answer in one sentence. — a length the model can meet.
- Give it as a table. — a format the model can produce.
These are the same quantifiers and instructions as in imperatives and directions — the grammar did not change, only the listener did.
When the model gets it wrong
If the output is wrong, the fix is usually to re-read your own instruction, not to blame the model. Ask:
- Did I name the object? (Summarise what?)
- Did I give a constraint? (how long, for whom, what format?)
- Did I use the imperative, or a vague request?
Rewrite the prompt with the missing piece and run it again. This loop — instruct, read the result, refine the instruction — is exactly how a good prompt is built.
Test yourself
The quiz below checks the imperative and the three pieces of a clear instruction.