Giving your prompts a reliable shape
Good prompts tend to share the same three parts. This guide breaks down the anatomy of a well-structured prompt, the instruction, the context, and the format, so you can build strong requests every time instead of hoping for the best.
New here? Before this guide, you might want to read Writing your first prompt.
In this guide, you will learn the simple three-part shape that turns a decent prompt into a reliably good one.
The anatomy of a good prompt
Once you have written a few prompts, you start to notice that the ones that work well tend to have the same bones. A strong prompt usually has three parts: an instruction, some context, and a format.
Think of it like giving directions to a taxi driver. You say where you want to go (the instruction), you mention the useful details like “the entrance round the back” (the context), and you say how you want to get there, the scenic route or the fast one (the format). Leave any of the three out and you might still arrive, but the journey is a gamble.
Let’s take each part in turn, because getting all three into a prompt is most of what separates a frustrating answer from a genuinely useful one.
The instruction: what you want done
The instruction is the heart of the prompt: the actual task. This is the verb, the thing you want the AI to do. Write, summarise, explain, compare, brainstorm, fix.
Be direct about it. “Summarise this report” is clearer than “I was wondering if you could maybe have a look at this report and let me know your thoughts.” The AI is not offended by plain instructions, and clear ones get cleaner results.
The context: what the AI needs to know
Context is everything the AI needs to do the task well: who the answer is for, why you need it, any background, and the material itself. This is usually where a weak prompt falls down, because the AI cannot see anything you have not told it.
There is a limit to how much the AI can take in at once, a bit like a desk that can only hold so many papers before things start falling off the edge. That working space is called the context window, and for everyday prompts you will rarely bump into it. It is enough to know it exists, so you are not surprised if a very long document ever gets truncated.
The format: how you want the answer
The format tells the AI how to hand the answer back. A short paragraph? A bulleted list? A table? Three options rather than one? A formal tone or a friendly one?
This part is easy to forget and makes a surprising difference. Asking for “five bullet points, each one sentence long” gets you something you can use straight away, rather than a wall of text you then have to reshape yourself.
Try it yourself
Let’s build a prompt with all three parts on purpose. Take a task and write it out labelling each piece in your head as you go. Here is the shape:
[Instruction] Write a message to my team letting them know our Friday
meeting is moving to Monday.
[Context] It's a weekly catch-up, nothing formal. The reason is that two
people are off on Friday. I want people to feel it's no big deal.
[Format] Keep it to three or four sentences, warm and casual, ready to
paste into our chat app.
Notice you do not need to write the words “instruction” or “context” in the real prompt; they are there to remind you to include each part. Send it, then try removing the format line and sending it again. The difference in how usable the answer is will show you why all three parts earn their place.
A common first instinct
A common first instinct is to give the instruction and stop there, assuming the AI will fill in the rest sensibly. It will fill in the gaps, but it fills them with guesses, and its guesses may not match what you had in mind. You will get far more consistent results if you make a habit of checking all three parts are present: did I say what to do, give the background, and describe the answer I want?
Next steps
You now have a dependable shape for your prompts. The next step is to go deeper on the middle part, context, because it is where most of the quality lives. That is exactly what we cover in giving your AI context.
Try it yourself
Put this into practice with a ready-made prompt.
Document Summariser
Turn a long document into a clear summary at the length, focus, and reading level you need, so you can grasp the key points fast.
Use this prompt →
Blog Post Outliner
Turn a rough topic into a clear, structured blog outline with a working title, sections, and key points, ready for you to write around.
Use this prompt →
See it in action
How to summarise a long document with AI
Turn a long report, contract, or article into a clear summary you can actually use, in a few minutes. You will learn how to get the length and focus you want, and how to check the summary is faithful to the original.
Works with Claude, ChatGPT, Gemini