Best AI Prompts for Work: 10 From Our Library You Can Copy Today

If you work with text, clients, documents, or data, the best AI prompts for work aren't the clever ones. They're the ones you use again on Monday. We're a team that builds AI systems for professionals, and the list below is drawn from the prompts we open first across client work and our own inbox. You can copy any of them today.

We wrote this because most prompt libraries we see — Zapier's blog roundups, HubSpot's Connector Prompt Library, Atlassian's Rovo use cases, Notion's template gallery, Product Hunt collections — are long and generic. A good library isn't about volume; it's about ten prompts you'd miss if they disappeared.

What makes a prompt "best for work"

The good prompts share three traits:

• They solve a task you actually repeat (reply triage, meeting notes, status reports, pricing emails).
• They name the role, the inputs, the format, and the constraints — not just "write a professional email."
• They have blanks. A prompt without placeholders is a template; a working prompt is a form with slots for your context.

Not sure your prompts are the problem? Why ChatGPT Gives Generic Answers walks through the usual suspects before you blame the model.

10 AI prompts for work (copy and adapt)

Each prompt below names a role, a task, a format, and a constraint. Replace the bracketed inputs with your own text.

1. Reply to a long email in two paragraphs

You are my executive assistant. Below is an email thread. Draft a reply in US English, under 150 words, in two short paragraphs: first acknowledge their point, then state what I will do next and by when. Warm and specific tone, no corporate filler.

Thread:
[paste]

What I want to happen next: [one sentence]

2. Follow-up to a prospect who went quiet

You are a B2B sales rep writing to a prospect who hasn't replied in 10 days. Reference the last message, add one new piece of value (a question, a short case, or a resource), restate the single next step, and keep it under 90 words. No "just checking in." No "circling back."

Last message I sent:
[paste]

Prospect title and company: [paste]

3. Decline a request without burning the relationship

You are a calm head of operations. Draft a reply that declines the request below. Offer one alternative, acknowledge the trade-off, and keep the door open for the next quarter. Under 120 words, no apology longer than one line.

Request:
[paste]

4. Turn messy meeting notes into decisions and owners

You are a project coordinator. Below are rough notes from a 45-minute meeting. Return three things:
1) A table with columns: Decision, Owner, Due date.
2) A list of open questions (max 5), each tagged with who should answer.
3) One-paragraph summary for people who weren't in the room.

Notes:
[paste]

5. Weekly status report your manager will actually read

You are a chief of staff. Turn the raw updates below into a weekly status report with four sections: Shipped, In progress, Blocked, Decisions needed. Each item is one line. Flag any date slip. Audience is a department head who has 90 seconds.

Updates:
[paste]

6. Research brief from three articles

You are a research analyst. Read the three articles pasted below. Produce: (a) a 150-word synthesis, (b) three claims where the authors disagree with each other, (c) two unanswered questions worth more research. Cite each article by headline, not by URL.

Articles:
[paste]

Verify every number the model gives you back. See our hallucination checklist.

7. Rewrite a draft so it sounds like a person

Rewrite the paragraph below so it sounds like a mid-career professional, not a chatbot. Keep every fact. Cut adjectives. Vary sentence length. Replace empty words like "robust," "best-in-class," "innovative" with specifics. Target: eighth-grade reading level, US English.

Draft:
[paste]

8. First-pass contract review for a non-lawyer

You are a contracts manager reviewing a vendor agreement for a mid-sized company. For the contract below, return a table with columns: Clause, Risk level (Low/Med/High), Why it matters, Question to ask the vendor. Flag anything about termination, IP, liability caps, and auto-renewal. End with a one-paragraph plain-English summary.

Contract:
[paste]

Don't paste text that identifies the counterparty or a real deal unless your workspace is private. What not to upload to AI covers the lines not to cross.

9. Spreadsheet question you'd rather not write a formula for

You are a data analyst. I will paste rows from a spreadsheet. Treat the first row as headers. Answer the question below using the data only; don't guess at anything missing. Return the answer, the method you used, and any rows you ignored and why.

Question: [paste]

Data:
[paste]

10. First draft of a proposal section

You are a senior proposal writer for a boutique agency. Draft the "Approach" section for the proposal below. Three subsections, 80–120 words each: Discovery, Delivery, Measurement. Reference the client's stated goals verbatim. No marketing adjectives. US English.

Client brief:
[paste]

How to adapt any prompt (so these keep working)

A prompt from any library — ours, HubSpot's, Atlassian's Rovo use cases, Zapier's roundups — is a starting point, not a finished tool. Four edits land almost every prompt harder:

• Role. Replace "You are an assistant" with the specific role you'd hire: "You are a B2B product marketer who has launched three SaaS tools."
• Audience. Say who reads the output. "Written for a CFO who has 60 seconds" produces different text than "written for your team's Slack."
• Format. Ask for a table, a numbered list, or a word count. Models follow format instructions more reliably than tone instructions.
• Constraints. Add what to avoid: "no bullet points," "no adjectives," "no sentences over 20 words."

The deeper walkthrough is in How to Write Better ChatGPT Prompts for Work. If the first answer is close but off, follow-up prompts usually fix it faster than rewriting from scratch.

Where prompt libraries break

A library is useful until it isn't. Five limits worth watching, as of October 2026:

Limit	What happens	What to do
Stale pricing or features	Prompts referencing GPT-3.5, retired plugins, or old plans no longer behave the same	Verify the model and plan the prompt was written for
Confidential data	Client names, salaries, or contracts land in a shared workspace	Strip identifying details before pasting
"Magic one-liners"	Prompts promising results with no inputs	Treat them as marketing; add your own context
Model drift	Same prompt gives different output month to month	Save the output you liked and the model/date
"Sounds AI" style	Default output reads smooth and empty	Edit the draft; don't ask the model to "humanize" it

The last one trips up more teams than any other. The fix is in How to Make AI Writing Sound Human.

Building your own prompt library (the short version)

Public libraries are a decent browse. But the prompts that save you time are the ones that match your work. A simple way to build yours:

1. Keep a running doc for one week. Every time you ask AI for help, paste the task, the prompt, and the output.
2. At the end of the week, pick the five tasks you did more than once.
3. Turn each into a template with blanks. Add a one-line note on when to use it.
4. Save the templates where you work: a Notion page, a Google Doc, Raycast or Alfred snippets. Not where you don't.
5. Revisit every six to eight weeks. Model behavior shifts, and so does your work.

If you'd rather start from a curated set, our prompt library bundle collects the ones we use across marketing, sales, HR, legal, and consulting work.

Where we take this further

The 10 prompts above cover email, meetings, reporting, research, writing style, contracts, and data. The second session of our two-week practicum rewrites prompts live with the group, using the same four-element frame (role, task, format, constraints), applied to your real inbox and documents. You leave with a 30-prompt library tagged to the tasks you actually repeat, not a generic grab-bag.