A prompt template is a reusable instruction for an AI model with fixed parts that stay the same and marked places for the details that change. For a task you do repeatedly, such as summarising meeting notes, drafting replies or writing product descriptions, a template gives more consistent results than typing a fresh request each time. This page provides a general template, explains each part and shows a completed example.
Why a template improves reliability
Most disappointing AI results come from missing information: the model was not told the audience, the facts or the format, so it guessed. A template makes you supply those things every time. It also makes results comparable. When the instruction is constant, you can see whether a change to it actually helped.
The template
Replace the text in square brackets. Delete sections that do not apply.
ROLE AND TASK
You are helping with [type of work] for [organisation or context].
Your task: [one clear sentence describing what to produce].
AUDIENCE AND PURPOSE
The result is for [who will read or use it].
It will be used to [purpose].
SOURCE MATERIAL
Use only the information between the markers below.
--- START ---
[paste facts, notes, data or text here]
--- END ---
REQUIREMENTS
- [length]
- [tone and style]
- [must include]
- [must not include]
ACCURACY RULES
- Do not add facts that are not in the source material.
- If information needed for the task is missing, list what is missing
instead of guessing.
- Mark anything you are unsure about with [CHECK].
OUTPUT FORMAT
[exact structure: headings, bullet points, table columns, etc.]
BEFORE YOU ANSWER
Check the result against the requirements and the accuracy rules.
The parts explained
Role and task
One sentence stating what should be produced. A role can help set the register, but real context does more than a title. You are an expert marketer adds little. You are writing for a family-run guesthouse whose guests are mostly couples adds a lot.
Audience and purpose
The same facts are written differently for a customer, a colleague or an accountant. Say who it is for and what they will do with it.
Source material
This is the most important section for business tasks. A model that is given your facts works from them. A model that is not will produce plausible invention. The markers show clearly where your material begins and ends.
Requirements
Keep the list short and free of contradictions. Brief but comprehensive pulls in two directions. Say which matters more.
Accuracy rules
These three lines reduce invented details. Asking the model to list missing information, instead of filling gaps, is particularly effective.
Output format
Describe the structure precisely. If the output will be pasted into a spreadsheet or a web page, say so.
A completed example
ROLE AND TASK
You are helping with customer communication for a small online
shop selling ceramics.
Your task: write a reply to the customer enquiry below.
AUDIENCE AND PURPOSE
The result is for a customer who asked about delivery.
It will be sent by email by our support colleague.
SOURCE MATERIAL
--- START ---
Customer message: "Hi, can I get the blue bowl set before 20 June?
It's a gift."
Stock: blue bowl set, 4 available.
Dispatch: within 2 working days of order.
Delivery: 3-5 working days within the EU.
Gift wrapping: available, 3 EUR.
--- END ---
REQUIREMENTS
- Under 120 words
- Friendly and direct, no exclamation marks
- Mention gift wrapping
- Do not promise a delivery date; give the latest order date
that fits the stated times
ACCURACY RULES
- Do not add facts that are not in the source material.
- If information is missing, list it instead of guessing.
OUTPUT FORMAT
Plain email text with greeting and sign-off "The Shop Team".
Adapting the template
- For summaries: add what the summary is for and the maximum length, and ask for decisions and open questions as separate lists.
- For data tasks: state the columns, the units and what to do with incomplete rows. See How to Use AI to Review Messy Spreadsheet Data Safely.
- For style-sensitive writing: add one or two short examples of text you consider right.
- For multi-step work: split it into separate prompts, one per step, instead of one very long instruction.
Test before relying on it
- Run the template with three or four different real inputs.
- Note what you had to correct each time.
- Turn recurring corrections into a requirement.
- Try an awkward case, such as missing information, and check that the model reports the gap.
- Save the working version with a date, and record changes.
Using it safely
- Do not paste personal data, passwords or confidential documents into a tool your organisation has not approved for that.
- Review every result before it is sent or published. The AI Output Review Checklist for Business Content lists what to look for.
- Models change. Re-test your templates from time to time.
Common mistakes
- No source material, then surprise at invented details.
- A long list of rules that contradict each other.
- Changing several things at once, so you cannot tell what helped.
- One enormous template for every task.
- Trusting the output because the template is good.
Next step
For the method behind this template, read How to Write Better AI Prompts and How to Build a Reusable AI Prompt for a Recurring Task.
