Prompt for Replying to an Email

The framing of the reply is the hard part, not the sentences

Most people think writing a reply is about finding polite words. It is not. The hard part is deciding what shape the reply should take before a single sentence is written. A reply that agrees to a request, a reply that declines it, a reply that chases someone who has gone quiet, and a reply that apologizes for a mistake are four completely different documents, even when they respond to the same email. If you ask an AI to "reply to this" without saying which of those you want, it will pick the safest, blandest option — usually a lukewarm half-yes that commits you to nothing and moves nothing forward. That is why the prompt makes you state the goal explicitly: agree, decline, answer, negotiate, follow up, or apologize. Naming the goal up front forces the model to structure the whole reply around a clear outcome instead of hedging, and it is the single biggest lever you have over whether the draft is useful or just words.

Tone has to match the relationship, not a generic setting

Tone is where AI-drafted replies most often feel wrong. The model has no idea whether the person emailing you is your closest client of ten years or a cold vendor you have never met, so left to itself it defaults to a slightly stiff, corporate register that fits neither. The prompt fixes this by asking for the relationship and the tone separately, because they are separate decisions. "Firm but polite" to a colleague you trust reads warmly; the same words to a stranger can read as cold. Telling the model who the recipient is lets it calibrate — softening the edges for someone you have rapport with, keeping things crisp and neutral for someone you do not. This matters most for the difficult replies. Declining a friend's request and declining a pushy salesperson call for very different warmth, and a reply that gets the tone wrong can damage a relationship even when every fact in it is correct.

Cover every ask, and always land on a next step

Received emails often contain more than one request buried in the paragraphs — a question, a scheduling ask, and a small favour, all in the same message. A reply that answers the first and forgets the other two forces another round of email and makes you look careless. The instruction to address every question or request exists precisely to catch this: it pushes the model to work through the original email point by point rather than responding to the general gist. Just as important is the requirement to end with a concrete next step. A reply that trails off with "let me know" puts the burden back on the other person and often stalls. A reply that ends with "I can do Thursday at 2pm — does that work?" moves things forward. Whether the goal is to agree, chase, or apologize, the draft should close by making it obvious what happens next and who does it.

Review before you send — never fire off a blind draft

The last discipline is the one people skip: read the draft properly before it goes out. AI writes fluent, confident sentences, and that fluency is exactly what makes it dangerous, because a wrong commitment reads just as smoothly as a right one. Check the specifics first — dates, names, numbers, and any promise the reply makes on your behalf. If you asked it to negotiate, confirm it did not concede more than you intended. If you asked it to decline, confirm it actually said no rather than softening into a maybe. The prompt tells the model to ask you for missing information rather than invent it, which cuts down on fabricated details, but it cannot know facts you never gave it. Treat the output as a strong first draft that still carries your name, adjust anything that does not sound like you, and only then hit send.

Why this prompt works

A reply is judged on framing, not just wording — a good "no" and a good "yes" read completely differently. This prompt makes you state the goal, the relationship and the must-include points up front, so the draft lands the right decision in the right tone and asks you for anything it does not know instead of guessing.

How to customize it

• State the goal — agree, decline, chase or apologise — before anything else.
• Name the relationship so the tone fits the person.
• Read the draft for dates and commitments before you send it.

Prompt variations to try

Politely decline

You are helping me reply to an email I received. Write a reply that politely declines their request.
Email I received:
[paste the email]
Relationship: [who they are].
Reason I'm declining (for my own clarity — soften or omit as appropriate): [reason].
Requirements:
- Say no clearly and early; do not bury it or drift into a maybe.
- Be warm and appreciative of the ask so the relationship stays intact.
- Offer an alternative if there is a reasonable one (a later date, a referral, a smaller version); if not, close gracefully.
- Keep it short. Do not over-explain or over-apologise.
Sign-off: [my name].

Chase a non-response

You are helping me follow up on an email that has gone unanswered.
My original email (or a summary):
[paste it]
How long ago I sent it: [e.g. 5 days].
Relationship: [who they are].
Tone: [gentle nudge / firmer, deadline approaching].
Requirements:
- Reference the earlier message briefly so they have context, without sounding accusatory.
- Restate the one thing I need from them and by when.
- Make it effortless to reply — a yes/no or a single clear question.
- Keep it short and assume they are simply busy, not ignoring me.
Sign-off: [my name].

Negotiate on price or scope

You are helping me reply to negotiate on price or scope.
Email I received:
[paste it]
My position: [what I want to hold firm on, and where I can flex].
My walk-away / limit: [the least acceptable outcome — for my clarity, not to state outright].
Tone: warm and professional, collaborative not combative.
Requirements:
- Acknowledge their concern genuinely before responding to it.
- State my position clearly and back it with the value or reason, briefly.
- Offer one or two concrete options rather than a flat no.
- End by asking which option works, so the ball is in their court with a decision to make.
Sign-off: [my name].

Common mistakes to avoid

• Not stating the goal of the reply. If you don't say whether you're agreeing, declining, or chasing, the AI defaults to a vague half-answer that commits to nothing. Name the outcome you want first.
• Leaving out the relationship. Without knowing if the recipient is a close client or a cold stranger, the model picks a generic corporate tone that fits neither. Say who they are.
• Forgetting to paste the full email. If the AI can't see every question in the original, it will answer some and miss others, forcing another round of email. Paste the whole message.
• Letting it invent details. The prompt tells the model to ask for missing facts, but if you skip that it may fabricate dates or figures. Give it the specifics, or check that it asked rather than guessed.
• Sending without reading. Fluent AI prose hides wrong commitments as smoothly as right ones. Always check dates, names, and any promise the reply makes before it goes out under your name.

Frequently asked questions

How much of the original email should I paste?

Will the reply sound like a robot wrote it?

Can it handle a firm "no" without being rude?

What if I don't know all the details it needs?

Tip: replace the parts in [square brackets] with your own details before you send. The more specific you are — audience, tone, goal, constraints — the better the AI output.