Skip to main content

AI prompts for everyday writing at work

Replies that get answered, summaries that surface decisions, drafts that read like a person wrote them. Each prompt below is built around the same idea: tell the model what to leave out, not just what to put in.

10 free prompts · tested in ChatGPT, Claude & Gemini · browse the full library

Sales follow-up email after no reply

Re-engage a prospect who has gone silent without sounding desperate.

You are a B2B sales rep writing a follow-up email. The previous email was sent days ago days ago and got no reply.

Context about the prospect:
- Company: company
- Role: role
- Original ask: original ask

Write a 4–6 line follow-up that:
1. Opens with one specific, non-generic observation about company (no "hope you're doing well")
2. Acknowledges the silence without guilt-tripping
3. Offers a low-friction next step (a question, a 1-line answer, or an opt-out)
4. Signs off without a CTA stack

Tone: confident, peer-to-peer, no jargon. Output only the email body.
Why it works
Constrains length, bans the cliché openers most AIs default to, and forces a single low-friction CTA.
What to swap
Replace the highlighted fields. Set the follow-up gap to control urgency.

Politely decline a request without burning a bridge

Say no professionally while leaving the relationship intact.

You are helping me decline a request without damaging the relationship.

The request: request
Why I am declining: reason
My relationship with the asker: relationship

Write a 3–5 sentence reply that:
1. Acknowledges the request specifically (not generically)
2. Declines clearly — no "maybe later" hedging
3. Offers exactly one alternative if reasonable, or none if not
4. Closes warmly without apologizing for the no

Output the reply only.
Why it works
Banning hedging language and apologetic framing is the single highest-leverage edit in business writing.
What to swap
Be honest in the `reason` field — the AI cannot soften what it doesn't know.

Rewrite anything in a specific tone

Convert a draft to a target tone without rewriting from scratch.

Rewrite the text below in the following tone: tone description.

Rules:
- Keep all factual content and structure intact.
- Do not add new claims or examples.
- Preserve roughly the same length (±20%).
- Output only the rewritten text.

Text to rewrite:
"""
original text
"""
Why it works
The triple-quote delimiter prevents the model from interpreting your text as instructions. The length constraint stops it from doubling the wordcount.
What to swap
Be specific in tone_description — 'friendly' is weak; 'warm but pragmatic, like a senior engineer talking to a junior' is strong.

Summarize a long email or Slack thread

Get to decisions and actions when you joined a conversation late.

Summarize the conversation below into:

1. **TL;DR** — one sentence
2. **Decisions made** — bulleted, with who decided what
3. **Open questions** — bulleted, with who owns each
4. **Action items** — bulleted, with owner and (if stated) deadline

Be ruthless. If something was discussed but no decision was made, put it in "Open questions". "Decisions" is only for what was actually decided.

Conversation:
"""
conversation text
"""
Why it works
Forces a decisions-vs-open-questions separation that AIs avoid by default. Eliminates the 'everyone agreed' false consensus.
What to swap
Paste the raw thread. Don't pre-edit it — the model is better at finding signal than you are.

Structure a long-form piece when the blank page wins

Get unstuck on an essay, report, deck, or document by building a working outline in one pass.

I am stuck on a blank page. Help me build an outline.

What I am writing: thing
The audience: audience
The single takeaway I want them to leave with: takeaway
What I am NOT going to say (out of scope): out of scope

Produce:
1. **Three candidate angles** — each a one-sentence framing for the whole piece.
2. **For the strongest angle**: a 5–7 point outline with one sentence per section.
3. **The opening hook** — one sentence that earns the next paragraph.
4. **The closing line** — one sentence that lands the takeaway.

Pick the angle yourself; do not ask me to choose.
Why it works
The 'pick the angle yourself' constraint stops the model from punting the hard decision back to you.
What to swap
The takeaway is the most important field — if you can't write it, the piece isn't ready.

Request time from someone senior

Ask a busy person for a meeting in a way that makes saying yes cheap.

You are helping me ask for a meeting with someone more senior than me.

Who they are: their role
What I want from the meeting: the ask
Why they specifically: why them
How much of their time I need: duration

Write an email of 5 lines or fewer that:
1. States the ask in the first sentence — no throat-clearing
2. Gives one concrete reason it has to be them and not someone else
3. Names the exact time needed and offers two specific slots
4. Makes declining easy with one sentence, without being self-deprecating

Ban: "picking your brain", "quick chat", "touch base", "circle back", "at your earliest convenience".
Output only the email body and a subject line.
Why it works
Senior people decline on cost, not merit. Naming the duration and pre-empting the decline removes the two reasons they stall.
What to swap
Be specific in `why_them` — a reason that would apply to five people reads as a mass email.

Announce an unwelcome decision to a team

Deliver a decision people will not like without hedging or corporate fog.

You are helping me announce a decision my team will not welcome.

The decision: decision
Who it affects and how: impact
What is genuinely still open: still open
What is not open for discussion: settled

Write an announcement of 200 words or fewer that:
1. States the decision in the first two sentences — before any context
2. Gives the actual reason, however unreassuring
3. Separates clearly what is settled from what people can still influence
4. Says what happens next and by when
5. Does not thank people for their patience, ask them to be excited, or describe the change as an opportunity

Tone: plain and direct. Neither apologetic nor upbeat.
Why it works
Burying the decision under context is read as evasion. Leading with it, then separating settled from open, is what stops the speculation.
What to swap
Fill `settled` honestly. If you present a settled decision as open, you will spend the next month relitigating it.

Cut a draft in half without losing the argument

Get a long piece of writing down to length while keeping every load-bearing point.

Cut the text below to target length while preserving its argument.

Rules:
1. First, list the load-bearing claims — the ones that, if removed, break the argument. Number them.
2. Then produce the cut version, keeping every numbered claim.
3. Cut in this order: throat-clearing openers, restated points, hedges ("I think", "it seems", "somewhat"), adverbs, examples beyond the first, then background the reader already has.
4. Do not compress by making sentences denser or more abstract. Remove whole sentences instead.
5. After the cut version, list anything you removed that you think was a genuine loss.

Text:
"""
paste draft here
"""
Why it works
Asking for the load-bearing claims first stops the model from compressing prose into unreadable density — the default failure mode of 'make this shorter'.
What to swap
Give a concrete target such as '400 words' or 'one screen'. Vague targets like 'shorter' produce a 10% trim.

Explain your work to someone outside the field

Make a specialist topic land with a smart reader who has none of your background.

Explain the following to audience, who is intelligent but has no background in this field.

Topic: topic
What they need to be able to do afterwards: their goal
Jargon that must be avoided entirely: banned terms

Requirements:
1. Open with why this matters to them specifically. General interest is not the hook
2. Use one analogy, and state explicitly where the analogy breaks down
3. Define any unavoidable term the first time it appears, in six words or fewer
4. Keep to word count words
5. End with the one thing they would get wrong if they only read this

Do not simplify by omitting the part that makes it difficult. Say that part plainly instead.
Why it works
Naming where the analogy breaks prevents the confident-but-wrong understanding that simplified explanations usually produce.
What to swap
Set `their_goal` to a decision or action. 'Understand it' produces a lecture; 'decide whether to fund it' produces something useful.

Write feedback someone can act on

Turn a reaction to someone's work into specific, usable written feedback.

Help me turn my reaction to a piece of work into feedback the person can act on.

The work: what it is
My unfiltered reaction: raw reaction
Our relationship: relationship
What has to change before this is done: must change
What is genuinely good: whats good

Write feedback that:
1. Names what is working first, and names it specifically. This is not a compliment sandwich
2. Separates "this must change" from "I would have done it differently" and labels which is which
3. For each required change, says what the problem is. Do not prescribe the fix
4. Ends with one question that would resolve the biggest open disagreement

Do not use "just", "maybe consider", "it might be nice if", or any phrasing that disguises a requirement as a suggestion.
Why it works
Most written feedback fails because required changes are phrased as preferences. Forcing the two into separate buckets is the entire fix.
What to swap
Put your real reaction in `raw_reaction`, including the annoyed part. The prompt's job is to translate it, and it cannot translate what it cannot see.

Frequently asked questions

Replies that get answered, summaries that surface decisions, drafts that read like a person wrote them. Each prompt below is built around the same idea: tell the model what to leave out, not just what to put in. This page collects 10 free, copy-pasteable writing & communication prompts, each tested in ChatGPT, Claude, and Gemini with a one-line rationale and the fields you need to swap in.