12 ready-to-use workplace prompts (copy, paste, save hours)
Prompts grouped by real task — emails, meetings, reports, spreadsheets, interviews. Replace anything in [brackets].
A good prompt packs four things into one instruction: a role, the task, the required output format, and constraints. The twelve below are ready to run today. Copy the block, replace the bracketed parts, paste into any chat AI.
Email and communication
- Polite refusal.
You are an experienced office manager. Write a 3-sentence email declining [the request] while keeping the relationship. Tone: warm but firm. No over-apologizing. - Follow-up without nagging.
Draft a follow-up email to [person], who has not replied since [date]. Goal: get a yes/no by [date]. Keep it under 80 words. - De-escalate an angry customer reply.
Rewrite this customer reply so it acknowledges the problem, de-escalates, and states the next step with a deadline. Draft: [paste].
Meetings and inbox
- Meeting summary.
Below are my raw meeting notes. Produce: decisions (max 5), action items with owners and deadlines, and questions left open. Notes: [paste]. - Inbox triage.
Classify these emails into Needs my reply / FYI / Can delegate / Junk. For each Needs my reply, draft a 2-sentence response. Emails: [paste].
Reports and writing
- Weekly report.
Turn these bullets into a manager-ready weekly report: 3 achievements with numbers, 2 blockers, 3 next-week priorities. Points: [paste]. - Long article to five bullets.
Summarize this article into exactly 5 bullets a busy executive can read in 60 seconds. Article: [paste].
Data and spreadsheets
- Excel formula.
I have a spreadsheet with columns [A, B, C]. I want to [goal]. Give me the exact formula, then explain it in one sentence. - Clean a messy list.
Standardize this list of company names: consistent capitalization, remove legal suffixes, flag duplicates. Data: [paste up to 50 rows].
Interviews and career
- Pressure-test your answer.
Here is my interview answer: [paste]. Ask me the three hardest follow-up questions a real interviewer would ask. - Rewrite a resume bullet.
Rewrite this resume line so it leads with impact, uses an active verb, and adds a metric placeholder in brackets: [line]. - Cover letter opener.
Draft the first paragraph of a cover letter for [role]. Connect my background [2 lines] to the job's need [2 lines]. No clichés.
Keep a running copy of the prompts that saved you real time. After two weeks you will have a personal library of ten to twenty prompts tuned to your actual work — and that library, not the chat app, is what compounds.
How it works
- Pick the prompt closest to a task you are doing this week.
- Replace every bracketed placeholder with your real details.
- Save the prompt that worked into my-prompts.md with one note on what it was good for.
Different models weight instructions differently. If the output is weak, strengthen the role line, add an example of the style you want, and specify the exact format — these three moves fix most flat results.
Yes — translate the instruction into Chinese and keep your documents in Chinese. The structure (role, task, format, constraints) is identical; a Chinese version of this library lives on the Chinese page of this tool.
Keep a file called my-prompts.md, paste the prompt plus one line describing when it worked, and reuse it. Prompts that sit in chat history get lost; prompts in a note file compound over months.
Any mainstream browser-based chat tool with a free tier. Prompts are model-agnostic — role, task, format and constraints work the same in every tool.
Ambiguity. A bad prompt names a topic (write about onboarding) without a role, a format or a constraint. If your prompt could be asked of any intern, it is too vague — add who you are, what format, and what not to do.
Who this tool is for
This collection suits the person who already opens a chat window and then types the same paragraph of instructions again: the admin drafting a fourth refusal email this month, the HR generalist summarising another interview panel, the operations coordinator turning messy notes into a status update. When your tasks repeat in shape but differ in detail, a saved prompt removes the setup cost and leaves your attention for the judgement that a model cannot do for you.
It is the wrong starting point for three groups. If you need a pipeline that runs without you, this is text, not automation — read the agent guide instead. If the decision is regulated or high-stakes, such as a disciplinary outcome, a redundancy selection or a legal position, a prompt is not a control; a documented process with a second reviewer is. And if you have never written one, copy first by all means, but plan to edit within the week: a prompt you are afraid to adjust is a prompt you will abandon.
How to read your result
The library supplies the instruction; the result is whatever the model sends back. Read that reply against the four ingredients the prompt was built from, in this order. Role: does the vocabulary fit the person you named, or does it read like a generic assistant? Format: did you receive exactly the sections you asked for, or one smooth paragraph? Constraints: is the length limit respected, and are the things you banned actually absent?
Two checks catch most problems before anything goes onward:
- Search the reply for a square bracket. A surviving placeholder means part of the instruction was skipped or ignored.
- Pick one factual claim — a date, an owner, a number — and confirm it against your source. A plausible-sounding owner can be invented.
If two of the four ingredients failed, do not patch the output by hand. Retype the instruction with the missing part made explicit; editing the reply teaches neither you nor the model anything about why it went wrong.
A worked example
An operations coordinator ran the meeting-summary prompt on eleven lines of raw notes from a 45-minute supplier call about a packaging change. Her instruction asked for decisions capped at five, action items carrying an owner and a deadline, and the questions left open. What came back:
| Section | What the reply contained | What she changed |
|---|---|---|
| Decisions | Three, each traceable to a line in her notes | Nothing |
| Action items | Four rows with owners and dates | One owner was wrong; she reassigned it to the packaging lead |
| Open questions | Two, one of them invented | Deleted the invented one, added the costing question she had noted verbally |
That same week she used the polite-refusal prompt on a request to turn around a 40-page translation in two days. The draft came back at three sentences, warm but firm, with a counter-proposal for a partial delivery. She changed one word and sent it. The difference between the two runs was not the model: the meeting prompt had vague inputs, so its output needed repair; the refusal prompt had a clear ask, a date and a tone, so its output needed a proofread.
Common mistakes
- Leaving the brackets in. A placeholder pasted unchanged produces a reply about "a request" in general, and the prompt then gets blamed for being generic.
- Pasting material you are not cleared to paste. Salary figures, customer names and health details belong in a prompt only if your company's policy says so.
- Asking for "make it better". Better against what? Name the audience, the length, and the one thing that must be preserved.
- Judging a prompt on a single run. Run it twice on different inputs before deciding it is a keeper or a dud.
- Keeping the good ones in chat history. A prompt that is not in your note file disappears at the next account switch or cleared conversation.
- Assuming the prompt verifies facts. No prompt here guarantees a usable draft, and none of them check whether the names and numbers you pasted are correct — that remains your job.