BIZPROMPT guide
How to structure a prompt for a task you repeat
Write the prompt once with all eight parts labelled: role, task, context, inputs, constraints, output format, examples and done when. Put the details that change from run to run in square brackets, so next time you fill in only those.
A task you repeat, such as a weekly status email or a summary of meeting notes, is worth one careful prompt. You write it once, keep it, and fill in the parts that change.
Start with the task line
Write one sentence that starts with a verb: write, summarize, compare, extract. Example: 'Summarize the meeting notes below for the team.' If the sentence needs two verbs, you probably have two prompts.
Label the eight parts
- Role: who the assistant should act as, such as a senior support writer.
- Task: the one sentence above.
- Context: two lines on who the result is for and why it matters now.
- Inputs: the text, data or draft to work from, with a clear start and end.
- Constraints: length, tone and what to avoid.
- Output format: the shape of the result, such as a table, bullets, or a subject line and paragraphs.
- Examples: optional, one result you liked.
- Done when: how you will judge the answer.
Put each part on its own labelled line, in the same order every time. A part you leave out on purpose is fine; delete its line instead of leaving it blank.
Mark what changes each time
Read the prompt and ask which words change from one run to the next: the date, the audience, the pasted notes. Replace each one with a square-bracket slot that says what goes there. Example: 'Context: this goes to [team name] before [meeting day].' The fixed lines stay word for word, so every run starts from the same frame.
Write a done line you can check
A done line is a test you can run on the answer. 'Make it good' cannot be checked. 'Done when the email has a subject line, three bullets and one question for the reader' can. Read it aloud; if you cannot check it, rewrite it.
Keep one output format and one length
Pick one shape and one limit. Two formats or two lengths in the same prompt leave the assistant to guess which one you meant.
A worked example
Example only. The rough prompt 'Write the weekly update.' becomes eight labelled lines:
- Role: [Who the assistant should act as, such as a project coordinator]
- Task: Write the weekly update email for my team.
- Context: The team is [team name]; the update goes out every [day].
- Inputs: The notes below, between the two lines of dashes.
- Constraints: No more than 150 words, plain English, no jargon.
- Output format: A subject line, then three bullets.
- Examples: [Optional: an update you liked]
- Done when: The reader knows what changed, what is late and what we need from them.
Only the slots in square brackets change from week to week.
What the free structurer does
The free BIZPROMPT structurer checks which of the eight parts a prompt has, gives a score from 0 to 100 for the parts present, and fills each missing part with a placeholder. It runs in your browser, no model runs on that page, and the score says nothing about the quality of any answer.
Common questions
Do I need all eight parts every time?
No. For a simple task, role, examples and done when can be short or left out. Leaving a part out on purpose is different from forgetting it, so delete the line rather than leaving an empty slot.
How long should a reusable prompt be?
Long enough to fill the parts that matter. One line per part plus the pasted inputs is a workable start. Add detail where the answers keep missing the mark.
Will a structured prompt always get a good answer?
No one can promise that. A clearer prompt usually gets a closer answer, but the structure changes the question, not the reply.