Write your first useful prompt
Turn a vague AI request into a clear business brief, then improve and check the draft with a worked fictional example.

AI-generated conceptual illustration.
Fictional source notes and editorial example output; not a captured ChatGPT or Claude response. Official references checked 4 October 2026.
“Write something professional” usually produces something that sounds professional. Whether it says the right thing is a separate matter.
A useful prompt gives the tool a job, the material it needs and a clear idea of what a finished answer should look like. You can write that in ordinary English.
What you'll make
a short collection message from rough notes, plus a reusable prompt template.
What you'll need
an approved ChatGPT or Claude account and a new chat. No upload, web search or connector is needed. Use the fictional notes below before adapting the method to approved real work.
1. Choose one output
For this exercise, the job is to turn rough collection notes into a customer message.
Do not ask for a message, sales strategy, operations manual and social campaign in the same prompt. You will spend the rest of the conversation trying to identify which part has gone wrong.
Write the task in one sentence:
Turn these notes into a short message for a customer collecting a
borrowed demonstration kit.
That gives the tool a destination. Now supply the route.
2. Add the reader and the source material
The reader has already arranged the loan. They need to know what to do, not be persuaded to buy anything.
Use these fictional notes:
FICTIONAL PRACTICE NOTES
Kit: DEMO-14.
Collection: reception, Tuesday, 09:00–11:00.
Customer must bring the booking reference.
Reception will check the contents with the customer.
Return: Thursday, by 16:00.
Calendar dates and reception address: not yet supplied.
Tone: helpful and straightforward.
Useful context is information that changes the answer. The collection window matters. A paragraph announcing that your business is “passionate about excellence” probably does not.
3. Say what must stay true
AI can fill gaps with plausible details. Make the gaps part of the task rather than hoping they disappear.
Specify the facts it must preserve, the things it must not invent and the format you want. These are the five parts of a practical brief:
- Task: what to produce
- Reader: who it is for and what they need
- Source: the material it should use
- Constraints: what must remain true or stay out
- Format: length and shape of the answer
Here is the complete prompt to copy:
Turn the fictional notes below into a short message for a customer
collecting a borrowed demonstration kit. They have already arranged
the loan and need practical instructions.
Use only the notes for factual details. Preserve the kit code,
collection window, booking-reference requirement, contents check
and return deadline.
Do not invent calendar dates, an address, a contact person, charges
or a booking link. Do not add promises or sales language.
Give me:
1. A customer-facing draft of no more than 100 words.
2. A separate "Before sending" list of missing information.
Use plain, friendly English. Make any unresolved detail visible.
FICTIONAL PRACTICE NOTES
Kit: DEMO-14.
Collection: reception, Tuesday, 09:00–11:00.
Customer must bring the booking reference.
Reception will check the contents with the customer.
Return: Thursday, by 16:00.
Calendar dates and reception address: not yet supplied.
Both OpenAI’s prompting guidance and Anthropic’s guidance support clear instructions, relevant context and specific output requirements. The template and example here are our editorial application of that advice.
4. Inspect the first answer
Read it beside the notes. Check facts before polishing the wording.
A suitable structure would look like this. This is an editorial illustration, not a captured response from either product:
Customer-facing draft
Please collect kit DEMO-14 from reception on Tuesday [date],
between 09:00 and 11:00. Bring your booking reference. Reception
will check the contents with you when you collect it.
Please return the kit by 16:00 on Thursday [date].
Reception address: [confirm address].
Before sending
- Confirm both calendar dates.
- Add the reception address.
- Replace all placeholders and check the final message.
It is deliberately unfinished where the source is unfinished. The tool should not be rewarded for concealing missing information.
Look for these common failures:
- Turning “by 16:00” into “at 16:00”
- Dropping the booking reference
- Adding a deposit, fee or promise that was never supplied
- Making up an address
- Mixing internal checks into a message ready to send
If the answer contains an unsupported fact, correct it. A confident explanation of the guess does not make it a source.
5. Give a specific revision
“Make it better” starts the guessing again. Describe the change you want and the facts that must survive.
For example:
Shorten the customer-facing draft to no more than 70 words.
Keep every supplied fact and the visible placeholders.
Keep the internal checks in their separate section.
Use "by 16:00", not "at 16:00".
If something important is wrong, correct that before asking for a different tone. Change one main thing at a time so you can see whether the revision helped.
For more complex work, ask the tool to identify missing information before drafting:
Before drafting, list the information you need that is missing
from my notes. Ask up to three questions that would materially
change the answer. Do not fill those gaps yourself.
There is no need to demand its hidden reasoning. Ask for the sources, calculations, assumptions and concise explanation you need to review the work.
6. Finish the job yourself
For real use, confirm the missing information with the right source, replace the placeholders and check the final draft. Do not paste a customer’s details into an unapproved account to make the example feel realistic.
Read the message as the person receiving it. Could they collect and return the kit without having to guess? If not, it is not finished.
AI can still make mistakes even when the prompt explicitly forbids them. OpenAI’s reliability guidance explains why important facts need independent checking.
7. Save the reusable brief
Keep this template in your own notes:
TASK
Create [one specific output].
READER
It is for [audience], who needs to [purpose].
SOURCE
Use the material below for factual claims.
[Paste approved source material.]
CONSTRAINTS
Preserve [facts, figures, dates or wording].
Do not invent [missing details, promises or sources].
If information is missing or conflicting, [ask or flag it].
FORMAT
Return [length, structure and tone].
Put assumptions and unresolved questions in a separate section.
Keep the source, prompt and reviewed answer together. When the task comes round again, change the relevant inputs and check the new output. A saved prompt is a useful starting point, not permission to stop reading.
Keep this: Give a clear brief. Correct a specific problem. Check the final version.
Sources and status
Official prompting and reliability references checked on 4 October 2026. All notes and example wording are fictional and illustrative. No live comparison, measured accuracy result or time-saving claim is made. Next: use business files and check the answer.