Practical Guides

AI Writing · 7 min

Write prompts that don't need a second pass

Most editing time on AI drafts is recovery from vague prompts. Build a small set of prompt habits that cut that work in half.

  1. 1

    Lead with the audience

    Before the task, the audience. 'For a senior buyer at a mid-market SaaS company who has not heard from us in six weeks, write...' The model produces better drafts when it knows who the writing is for. Skipping this is the single most common cause of generic output.

  2. 2

    Name the constraint that matters most

    Length, format, tone, structure — whichever is non-negotiable. 'Under 80 words.' 'Three short paragraphs.' 'No bullet points.' One sharp constraint is worth more than five soft preferences.

  3. 3

    Show one example when style matters

    When the voice is specific, paste one short example of writing in that voice before the new task. The model anchors on the example more reliably than on a description of it.

  4. 4

    Specify what must be preserved

    Names, dates, product terms, numbers, quotes. Tell the model explicitly to keep these exact. Hallucinated changes to these elements are the most damaging errors and the easiest to prevent.

  5. 5

    Ask for one thing at a time

    Multi-task prompts ('rewrite for tone, summarise, and translate') split the model's attention and produce uneven results. Break into separate calls. The total time is usually less, and the outputs are reliably better.

Common pitfalls

  • Vague tone instructions like 'make it sound better.' Specify what 'better' means.
  • Asking for length in words and expecting precision. Models hit length targets approximately; design around it.
  • Forgetting to delimit the input from the instruction. Triple backticks or XML tags prevent confusion.
  • Iterating on the prompt without naming what went wrong. Vague iteration produces vague improvement.

Frequently asked

How long should a good prompt be?

As long as the task warrants. Routine tasks: a sentence or two. Complex tasks with constraints and examples: 100-300 words is common.

Should I always use examples?

Only when style or format matters more than substance. For factual or analytic tasks, instructions alone are usually sufficient.

Does prompt language affect output?

Yes — the model writes in the language of the prompt by default and adopts some stylistic cues from it.

Run the workflow on a real draft

Open the rewrite engine and apply the steps to writing you actually have to ship today.

Start rewriting

Related guides