AI Prompt Templates for Office Work (Copy-and-Edit)
These are eleven prompt templates for routine office work: replying to email, recapping meetings, drafting quotes, summarising policies, writing spreadsheet formulas, posting jobs and answering complaints. Each one is written to be copied, edited and pasted into ChatGPT, Claude or Gemini.
Every template has the same parts: what the task is, the facts the AI needs, who will read the result, and the format you want back. Words in square brackets are placeholders. Replace them with your own details before you send the prompt, and read the draft before you use it. Under each template we explain why each part is there and what a usable draft should contain, so you know what to keep when you shorten it.
How to use these templates
- Fill every bracket. A missing detail is the main reason drafts come back vague. If you do not know a detail, write “unknown” so the AI does not invent one.
- Strip what you do not need to share. Use first names or [CUSTOMER] instead of full names, emails and account numbers, unless your plan and company policy allow them. See Is it safe to put company data into ChatGPT, Claude or Gemini?.
- Check facts, not just tone. Dates, prices, quantities and promises are where drafts go wrong. Compare each one with your source before sending.
The structure follows what the vendors recommend in their own guides: clear instructions, relevant context, examples where useful and an explicit output format (OpenAI, Anthropic, Google). For the reasoning behind that, read How to write a good AI prompt.
1. Email reply
Write a reply to the email below.
My goal: [WHAT YOU WANT TO HAPPEN, e.g. agree a new delivery date]
Facts to include: [DATES, NUMBERS, DECISIONS]
Do not promise: [ANYTHING YOU CANNOT COMMIT TO]
Tone: [friendly / formal / firm]. Length: under [100] words.
Email:
"""
[PASTE EMAIL]
"""
Why each part matters: the goal stops the AI writing a polite reply that decides nothing. “Do not promise” matters because drafts tend to offer refunds, discounts or deadlines nobody approved. The triple quotes separate the email from your instructions, so the AI does not treat the sender’s requests as yours.
A usable draft answers the sender’s actual question in the first two sentences, contains only the facts you listed, and ends with one clear next step.
2. Meeting recap
Turn these meeting notes into a recap email for [AUDIENCE].
Sections: Decisions, Action items (owner, task, due date), Open questions.
Only list a decision if the notes say it was agreed.
If an owner or date is missing, write "not stated" rather than guessing.
Notes:
"""
[PASTE NOTES]
"""
Why each part matters: the three fixed sections make recaps comparable from week to week. The two rules about decisions and missing owners target the most common error, which is a discussion being reported as a decision. Our meeting notes guide shows a worked example and how to check the result.
3. Quote or estimate draft
Draft a customer quote as a short email with a line-item table.
Customer need: [WHAT THEY ASKED FOR]
Line items (use these exact prices, do not change or add any):
- [ITEM] x [QTY] at [UNIT PRICE]
- [ITEM] x [QTY] at [UNIT PRICE]
Terms: valid for [30] days, payment [TERMS], delivery [LEAD TIME].
Show each line total and a grand total. Show your arithmetic.
Why each part matters: “use these exact prices” stops the AI rounding or filling in typical prices. Asking it to show its arithmetic lets you check each line in seconds. Always recompute the total yourself or in a spreadsheet: language models can make arithmetic mistakes, which is why Google suggests code execution for maths in its prompting guide.
4. Policy or document summary
Summarise the policy below for [AUDIENCE, e.g. new warehouse staff].
List: what they must do, what they must not do, deadlines, and who to contact.
Quote the policy's exact wording for any rule with a number, date or penalty.
Do not add advice that is not in the document.
Policy:
"""
[PASTE POLICY]
"""
Why each part matters: quoting exact wording for numeric rules means a mistake shows up immediately when you compare the summary with the original. “Do not add advice” keeps general knowledge from creeping in as if it were your company’s rule.
5. Spreadsheet formula
I use [Excel 365 / Google Sheets].
Columns: A = [NAME], B = [NAME], C = [NAME]. Data is in rows 2 to [LAST ROW].
I want a formula in [CELL] that [RESULT YOU WANT].
Edge cases: [blank cells / text in number column / no match].
Give the formula, then explain each part in one line.
Why each part matters: the app matters because some functions and argument names differ between Excel and Google Sheets. The column map is the detail the AI cannot guess. Edge cases decide whether you get a formula that breaks on the first blank row. Our spreadsheet formulas guide has eight verified examples.
6. Job post
Write a job post for a [JOB TITLE] at [COMPANY TYPE] in [LOCATION / remote].
Hours: [HOURS]. Pay: [PAY RANGE]. Start: [DATE].
Main duties: [3–5 DUTIES]
Must-have: [ONLY TRUE REQUIREMENTS]
Nice-to-have: [OPTIONAL SKILLS]
Plain language, under [300] words, no buzzwords. End with how to apply: [METHOD].
Why each part matters: separating must-have from nice-to-have stops the draft inflating the requirements. Asking for plain language and no buzzwords counters the generic phrases job post drafts often fall back on. Check the wording against your local employment rules.
7. Customer complaint reply
Draft a reply to this customer complaint.
What went wrong (our side): [FACTS]
What we will do: [REMEDY — refund, replacement, call back]
What we cannot do: [LIMITS]
Acknowledge the problem in one sentence, explain the remedy, give a date.
No blame on the customer. No legal admissions beyond the facts above.
Complaint:
"""
[PASTE COMPLAINT]
"""
Why each part matters: stating the remedy and its limits up front is what keeps the draft from offering more than you can deliver. The “no legal admissions” line is there because apologetic drafts can concede fault in broader terms than the facts support.
8. Rewrite for a different audience
Rewrite the text below for [AUDIENCE, e.g. customers with no technical background].
Keep every fact and number. Remove jargon or explain it in brackets.
Target length: about [N] words. Reading level: plain English.
Text:
"""
[PASTE TEXT]
"""
Why each part matters: “keep every fact and number” is the key constraint. Rewrites that simplify tend to drop qualifications such as “up to” or “in most cases”, which can change the meaning.
9. Weekly status update
Write a status update for [MANAGER / CLIENT] from these bullet points.
Format: Done this week, Next week, Risks or blockers, Help needed.
Each item one line. Keep my numbers exactly. Under [150] words.
Points:
[PASTE YOUR ROUGH NOTES]
Why each part matters: a fixed format is quicker to read and easier to compare across weeks. “Help needed” prompts you to state the request, which is often the reason for the update.
10. Procedure (SOP) first draft
Draft a step-by-step procedure for [TASK] for [WHO WILL DO IT].
Here is how we do it now, in rough notes:
[NOTES]
Number each step. Start each step with a verb. Mark any step where my notes
are unclear with [CHECK]. Add a short "Before you start" list of what is needed.
Why each part matters: the [CHECK] markers turn gaps in your notes into a visible to-do list instead of steps the AI has made up. Have the person who actually does the task review the draft.
11. Proofread without rewriting
Proofread the text below. Fix spelling, grammar and punctuation only.
Do not change wording, tone or structure.
Return the corrected text, then a list of each change you made.
Text:
"""
[PASTE TEXT]
"""
Why each part matters: without the “do not change” line, most assistants rewrite as well as correct. The change list lets you accept or reject edits in a few seconds.
Building a team library
Keep templates in one shared document with an owner and a last-reviewed date. When a template produces a bad draft, add the missing constraint to the template instead of fixing each draft by hand.
What long templates cost on the API
On a flat subscription such as ChatGPT Plus or Claude Pro, longer prompts do not change your bill. If your team runs templates through the API, each one is billed as input tokens. A template of about 150 tokens sent 10,000 times a month is 1,500,000 input tokens. On Claude Sonnet 5.5 at $2 per million input tokens (Claude API pricing), that is 1.5 × $2 = $3.00 a month for the template text alone, before the pasted content and the output. To turn a rough idea into a fuller template, try Idea → Prompt, and see How many words is a token? for converting words to tokens.
Frequently asked questions
Do these templates work in ChatGPT, Claude and Gemini?
Yes. They are plain-text instructions with no tool-specific features, so you can paste them into any of the three. Results will differ in wording, so read every draft before you use it.
Why do the templates use square brackets?
Square brackets mark the parts you must replace with your own details. Leaving a bracket in place, or deleting it without adding the detail, is the most common reason a template gives a generic answer.
Can I paste customer names and emails into a template?
Only if your company's AI policy and your plan's data settings allow it. Where you can, use first names or placeholders like [CUSTOMER] and add the real details after the draft is written.
Do longer templates cost more?
On a flat monthly subscription, no. On the API, every input token is billed, so a 150-token template sent 10,000 times a month adds about 1.5 million input tokens.
Should I tell the AI to act as an expert?
A short role line can set tone and vocabulary, but it does not add knowledge. The context you provide, such as the facts, the audience and the constraints, matters much more.
Related tutorials
Sources
- Prompt engineering guide — OpenAI
- Prompting best practices — Anthropic
- Prompt design strategies — Google
- Claude API pricing — Anthropic