Turn a One-Off Prompt Into Template
You get a clean, reusable template with a small set of well-named placeholders and example values, ready to save and reuse. It works because the AI is made to separate what stays fixed from what changes before it rewrites anything.
Worked in our testIt followed the format: a code-block template with five correctly styled placeholders and a clear role, steps and output format, then a three-column table and no extra commentary. It falls short on step 4. All the examples describe one damaged-delivery case, so they ignore the refund and late-order mix the prompt asked for. The "her" in the {our_response} example also assumes the customer's gender, which makes it less reusable.
The prompt
Act as a prompt engineer who turns one-off requests into reusable templates for small teams.
Task: Convert the prompt below into a reusable template that {team_audience} can fill in without rewriting the wording.
Original prompt: {original_prompt}
Steps:
1. List every detail in the original that would change between uses, such as names, figures, dates, tone or audience.
2. Replace each one with a clear placeholder in curly braces, using lowercase words joined by underscores. Use no more than five.
3. Tighten the remaining wording so it states the role, the task, the steps and the exact output format.
4. Write one realistic example value for each placeholder, based on {example_context}.
Output: First the finished template inside a code block. Then a table with the columns Placeholder, Meaning and Example value. Add no other commentary.
Change the parts in curly braces. Use it whenever you notice you keep copying a good prompt and editing the same few details by hand.
What happened when we ran it
We sent
Act as a prompt engineer who turns one-off requests into reusable templates for small teams. Task: Convert the prompt below into a reusable template that the customer support staff at a small online furniture shop, none of whom have written prompts before can fill in without rewriting the wording. Original prompt: Write a polite reply to Sarah Jenkins, who ordered an oak dining table on 3 October and says one leg arrived cracked. Apologise, offer a free replacement leg shipped within five working days, and ask her to send a photo of the damage. Keep it warm but brief, under 120 words, and sign off as Tom from Oakfield Living. Steps: 1. List every detail in the original that would change between uses, such as names, figures, dates, tone or audience. 2. Replace each one with a clear placeholder in curly braces, using lowercase words joined by underscores. Use no more than five. 3. Tighten the remaining wording so it states the role, the task, the steps and the exact output format. 4. Write one realistic example value for each placeholder, based on a typical week of damaged-delivery complaints, refund requests and late-order queries at the furniture shop. Output: First the finished template inside a code block. Then a table with the columns Placeholder, Meaning and Example value. Add no other commentary.
The model replied
```
You are a customer support agent at Oakfield Living, an online furniture shop. Write a reply email to a customer.
Customer name: {customer_name}
Their order: {order_details}
Their problem: {customer_issue}
What we will do and what we need from them: {our_response}
Follow these steps:
1. Greet the customer by first name.
2. Apologise sincerely for the problem, mentioning it specifically.
3. Explain clearly what we will do and what, if anything, we need from them, using the details above. Do not add promises or offers that are not listed.
4. Close with a short, friendly line.
Format:
- Plain email text only, with no subject line and no notes before or after.
- Warm, polite and brief: under 120 words, in short paragraphs.
- Sign off with "Kind regards," then "{agent_name}" and "Oakfield Living" on separate lines.
```
| Placeholder | Meaning | Example value |
|---|---|---|
| {customer_name} | The customer's full name | Priya Nair |
| {order_details} | The item ordered and the date the order was placed | A walnut bookshelf ordered on 27 October |
| {customer_issue} | What the customer says is wrong, in plain words | The left side panel arrived with a deep scratch |
| {our_response} | The solution we are offering, any deadline, and anything we need from the customer (such as a photo or order number) | Send a free replacement panel within five working days, and ask her to email a photo of the scratch |
| {agent_name} | The first name of the staff member sending the reply | Tom |
Run on Claude Sonnet 5.5, 02 Nov 2026. Automatic check: 4 out of 5.
Mentioned for
ChatGPT, Claude, Gemini
Where it came from
Found on GitHub, shared by yarin-zhang: yarin-zhang/AI-Gist.
This is our own version of the idea, written and tested by us. The original may differ.