What is a prompt?
A prompt is the text you enter into an AI model: your question, your task or a document with a request attached. Everything you type before you press Enter is the prompt. The model reads that text and produces an answer based on it.
The word comes from the verb "to prompt", meaning to urge someone into action. You set the model to work on something: a text, a summary, an analysis or a piece of code.
Why does the prompt determine the result?
The prompt determines the result because an AI model only knows what is in the prompt and in the ongoing conversation. It knows your business and your intent only once you state them. Whatever you leave out, the model fills in with an average assumption.
Compare these two prompts:
- "Write a text for our careers page."
- "Write an 80-word introduction for our careers page. Audience: senior developers. Tone: informal and direct, without corporate language."
The first gives you a generic recruitment text. The second gives you a text that fits the reader and the length you need.
What are the parts of a good prompt?
A good prompt has four parts: context, task, constraints and an example.
1. Context Say who you are and who the output is for. "I own an installation company with twelve technicians and write to private customers." This lets the model choose the right level of knowledge and the right tone.
2. Task State concretely what needs to happen. "Help me with my presentation" is too broad. "Create an outline of ten slides for a pitch to investors about our planning software" gives direction.
3. Constraints Set boundaries: length, tone, form and what you want to avoid. "At most 200 words, without jargon, as running text."
4. Example Show what you consider good. Paste an earlier text of your own and ask for the same style. The model picks up sentence length and tone.
A fill-in template with all four parts:
"I am [role] at [type of business]. My reader is [audience]. Write [what you need] about [topic]. Length: [number of words]. Tone: [tone]. Form: [running text, list or table]. Below is an example of a text in the style I am looking for: [paste example]"
What types of prompts are there?
There are five types of prompts you come across in practice: direct prompts, instruction prompts, few-shot prompts, step-by-step prompts and system prompts.
| Type | What it is | When you use it |
|---|---|---|
| Direct prompt | A single question or task | Simple tasks, such as translating |
| Instruction prompt | Task with context and constraints | Most professional work |
| Few-shot prompt | Task with two or more examples | When style and form are fixed |
| Step-by-step prompt | Request to write out the reasoning | Trade-offs and calculations |
| System prompt | Fixed instruction sent with every conversation | Recurring work and AI agents |
Direct prompt. "Translate this text into Dutch." For simple tasks that is enough.
Instruction prompt. A task with context and constraints, like the template above. This is the form you use most often.
Few-shot prompt. You give examples and ask for more in the same style: "Below are three product descriptions I like. Write two more in the same style for these products: [products]."
Step-by-step prompt. You ask the model to write out its reasoning first: "First list the pros and cons, and only then give your advice." This is known as chain-of-thought. It shows you how the answer came about and where an error sits.
System prompt. A fixed instruction that is sent along with every conversation. It states who the AI is, how it writes and what it does and does not do.
Read how to apply these forms in the 5 prompt techniques for entrepreneurs.
Which example prompts can you copy right away?
You can copy and fill in these five prompts. Replace the text in square brackets with your own details.
1. Proposal outline "I own a [type of business]. A customer is asking for a proposal for [assignment]. Create an outline with these parts: the customer's situation, our approach, timeline and investment. Write two to four sentences per part. Leave amounts open as [AMOUNT]. Do not make up details I have not mentioned."
2. Rewriting a customer email "Rewrite the email below. Keep the content and the agreements. Make it shorter, at most 120 words, and professional in tone. Start with the core of the message. The email: [paste email]"
3. Summarizing a document for a decision "Summarize the document below for a director who has five minutes. First give the decision being requested, then the arguments for and against in a table, then the risks. Quote figures literally from the document. The document: [paste text]"
4. Few-shot: posts in your own style "Below are three LinkedIn posts I wrote myself. Write two new posts in the same style and length about [topic]. My posts: [paste three posts]"
5. Step by step: substantiating a choice "I am deciding between [option A] and [option B] for [goal]. My situation: [describe situation]. First list the pros and cons per option. Then name which information is still missing. Only then give your advice, with the main reason."
Which mistakes are easy to make when writing a prompt?
Five mistakes are common when writing a prompt.
- Too vague. "Write something nice" gives the model no direction. State the goal, the reader and the length.
- No context. The model knows nothing about you until you tell it.
- Too much at once. Split a large task. Ask for an outline first and then work it out part by part.
- Not refining. Your first prompt is a starting point. Say what needs to improve and ask for a new version.
- Copying someone else's prompt unchanged. Adapt every prompt to your own business and your own tone.
In addition, always check the facts and figures in the answer. A good prompt lowers the chance of errors. Proofreading remains necessary.
What is a system prompt for an AI agent?
A system prompt is the fixed instruction an AI agent receives with every task. It states who the agent is, which business it works for, how it communicates and what it may and may not do. You write that instruction once and adjust it when your way of working changes.
Your daily instruction becomes short as a result. "Summarize yesterday's meeting" is enough, because the agent already knows your context. Read what else sets an agent apart in what an AI agent is.
The skill stays the same. If you can write a clear prompt, you can also instruct an agent well.
Frequently asked questions
What does prompt mean? A prompt is an instruction or task for an AI model. The word comes from the verb "to prompt", meaning to urge someone into action. You prompt the model to produce a text, a summary or an analysis.
How long should a prompt be? As long as needed to give context, task and constraints. For a simple question one sentence is enough. For a proposal or analysis, a prompt of one paragraph with an attached example is common.
What is prompt engineering? Prompt engineering is the deliberate writing, testing and improving of prompts so that an AI model reliably gives the desired result. For daily work it comes down to giving context, being concrete and refining.
What is the difference between a prompt and a system prompt? A prompt is the instruction you give per message. A system prompt is a fixed instruction in the background that is sent along with every conversation, for example with an AI agent.
Does the same prompt work in ChatGPT and Claude? Yes, a prompt with context, task and constraints works in both. The answers differ per model in style and length. So test a prompt you use often in the tool you work with.
Want to keep learning?
If you are new to AI, start with AI for beginners. If you want to write prompts for your own work with guidance, see the 1-on-1 AI training.
