An AI writing tool does what you ask, as best it understands the request. If your instruction is vague, the draft will be generic. If it is clear and specific, the draft will usually be closer to what you wanted. Learning to write a good instruction, called a prompt, is mostly learning to be clear about what you want.

Why vague prompts give vague results

Imagine asking a new assistant to "write something about our product". They would have to guess who it is for, how long it should be, and what you want people to do. A language model has the same problem, and it fills the gaps with the most common choices, which is why vague prompts produce generic text.

Every detail you supply removes a guess.

The elements of a good prompt

1. The task. Say what you want: an email, a summary, a product description.

2. The purpose. Why are you writing it? What should the reader do or feel afterward?

3. The audience. Who will read it, and what do they already know? "Busy hiring managers" produces different text from "high school students".

4. The key details. Give the facts the text must include, such as names, dates, numbers and points to cover. The tool cannot know what you do not tell it.

5. The tone and style. Formal, friendly, plain, playful. If you have a sample of your own writing, describe its style.

6. The format and length. "Three short paragraphs", "five bullet points", "about 150 words".

7. What to avoid. "No jargon" or "don't mention pricing" saves editing later.

Before and after

Before: "Write an email about the delay."

After: "Write a short, polite email to my client Priya explaining that the project will be delayed by one week because the vendor is late with data access. Say we expect to finish by Friday 14th. Offer a 10-minute call to discuss. Friendly but professional. Under 120 words."

The second prompt tells the tool who, what, why, when and how, and the result needs far less editing.

Use tools that do the thinking for you

Good tools build many of these elements into a form. In our Email Writer, for example, the fields ask for the purpose, recipient, key points, type and tone, so you supply the details and the tool handles the structure. The more specific your answers, the better the draft.

Iterate

Treat the first draft as the start of a conversation:

  • If it is too long, ask for a shorter version.
  • If the tone is wrong, say how: "less formal", "more direct".
  • If it missed a point, add it and generate again.
  • Generate more than once and combine the best parts.

Small changes to the instructions often make a bigger difference than long rewrites by hand.

Give examples

If you want a particular style, show it. A short sample of the voice you want, or an example of a good output, helps more than an adjective like "engaging".

Don't share what you shouldn't

Remember that your prompt is sent to an AI service. Leave out passwords, financial details, identity numbers and confidential information. Use placeholders such as [Client name] and fill them in afterward.

Check the output

A good prompt improves a draft but does not guarantee it is correct. Always check facts, names and numbers, and edit the text so it sounds like you. See our guides on fact-checking AI text and editing it into your own voice.

A prompt checklist

  • What is the task and purpose?
  • Who is the audience?
  • Which details must be included?
  • What tone, format and length?
  • What should be avoided?

If you can answer those, you can write a clear prompt.