# How do I prompt an AI assistant for a first draft I can actually use?

Source: https://humaniseai.ai/forum/read/how-do-i-prompt-an-ai-assistant-for-a-first-draft-i-can-actually-use/
Site: humanise.ai Forum, category Prompting and first drafts
Published: 2026-11-20. Updated: 2026-11-20. Replies: 0.

## Question by Harriet Pell (staff), 2026-11-20

Give the assistant a clear goal, the context, your own source material and your expectations, then treat what comes back as raw material. Name the audience, purpose and limits, paste a sample of your own writing, and ask it to flag facts it is unsure about. In assessed work, only do this where your rules allow it.

## Check the rules before you prompt

Everything below assumes AI is permitted for the task. At university that depends on the assessment brief, so read it first; the thread on [whether you can use ChatGPT for university assignments](/forum/read/am-i-allowed-to-use-chatgpt-for-university-assignments-in-the-uk/) shows where to look. At work, check your employer's policy and whether you are allowed to paste in client or personal information (the thread on [AI for work emails and reports](/forum/read/is-it-okay-to-use-ai-to-write-work-emails-and-reports/) covers that). If AI is not allowed for the piece, none of this applies.

## What a good prompt contains

Microsoft's guide to [writing prompts in Copilot](https://support.microsoft.com/en-gb/microsoft-365-copilot/get-started-writing-prompts-in-microsoft-365-copilot) says a prompt can include four parts:

- **Goal:** what you want the assistant to do.
- **Context:** why you need it, and who is involved.
- **Source:** the information or samples it should use.
- **Expectations:** how it should respond.

The same page says all that is required is a clear goal, but that you will often need more than a goal to get what you want, and that you should expect some back-and-forth. Anthropic's [prompt engineering overview](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview) starts from a related idea: have a clear definition of success criteria before you refine a prompt, and its techniques begin with clarity and examples. Applied to writing, that means deciding what a usable draft looks like before you type anything.

## Decide what "usable" means

Before you prompt, write down three things in a sentence each:

1. Who will read this, and what do they already know?
2. What must the piece do (persuade, explain, report, apply for something)?
3. What are the hard limits (word count, structure, referencing style, spelling variety)?

If you cannot answer these, the assistant will guess, and its guesses are generic.

## A worked brief

Suppose you are a nurse writing a short reflective account for a placement portfolio, and your course allows AI for structuring and first-draft support. A weak prompt is "Write a reflection on communication in nursing." A usable one looks like this:

> Goal: Help me draft a 400 word reflective account for my placement portfolio.
> Audience and purpose: My clinical assessor, who wants to see that I can analyse an event, not just describe it.
> Context: On my second placement a patient refused her medication until someone explained what it was for. I explained it, and she took it. I want to reflect on what that taught me about communication.
> Constraints: UK English. First person. Gibbs' reflective cycle as the structure, using its headings. No invented facts, no references.
> Style sample: Here are two paragraphs I wrote last term. Match the plain tone and sentence length: [paste your own paragraphs].
> Before you write: list anything in my notes that is unclear, and flag any factual claim you are not sure about.

Notice what is doing the work. The details come from you, the style sample anchors the voice, and the last line asks the assistant to expose its own uncertainty rather than hide it.

## Steps to follow

1. **State the goal in one sentence.** One task per prompt works better than five.
2. **Add the context and audience.** Name the reader and the purpose.
3. **Paste your own material.** Notes, an outline, or a paragraph you wrote. This is the "source" in Microsoft's list, and it keeps the draft tied to your actual thinking.
4. **Set expectations.** Length, structure, tone, spelling variety, things to avoid.
5. **Ask it to flag uncertainty.** Microsoft's guide warns that responses can include incorrect content and should be checked against trusted sources. A request to mark uncertain facts helps you know where to look first, but it does not replace checking.
6. **Iterate.** Ask for a change to one thing at a time: shorter, plainer, a different order.
7. **Run it more than once if you need to.** The same page notes that using the same prompt multiple times can give different responses.

## What to do with the draft

Treat it as scaffolding. Check every fact and reference yourself (the thread on [checking references from ChatGPT](/forum/read/how-do-i-check-that-references-from-chatgpt-actually-exist/) shows how). Then rework it into your own voice using the methods in [how to make AI-assisted writing sound like you](/forum/read/how-do-i-make-ai-assisted-writing-sound-like-me/). If the work is assessed, keep your prompts and chat history and declare the use as your university asks.

## Common mistakes

- Asking for the finished piece in one go, with no source material from you.
- Leaving out the spelling variety, then getting American English back.
- Pasting in other people's work, personal data or confidential material without checking that you may.
- Taking confident-sounding statements as true.

---
Cite as: humanise.ai Forum, "How do I prompt an AI assistant for a first draft I can actually use?", https://humaniseai.ai/forum/read/how-do-i-prompt-an-ai-assistant-for-a-first-draft-i-can-actually-use/, accessed 2026-11-20.
