Vibe Writing for Emails

You send thirty emails on an ordinary day. Twenty-five of them take a minute each. The other five are the ones this guide is about: the follow-up to a client who has gone quiet, the no to a colleague who will take it badly, the request to a manager for something you have already asked for twice. These take an hour, and an assistant can bring that down to ten minutes, as long as you stop it from writing like a hotel.

The model's default email is a recognizable object. It hopes this message finds you well, it wanted to reach out, it apologizes for the inconvenience, it would greatly appreciate, it thanks you for your patience and understanding, and somewhere in the third paragraph it mentions what it wants. It is also written to a stranger, because the model has never met your recipient and does not know you asked twice already.

Vibe writing for email means you direct with the ask and the history, the assistant proposes a draft under a word ceiling, you judge with the recipient's phone in mind, and you own the send, including the decision about what never gets pasted into a chat in the first place. The whole loop takes less time than the hour you were spending, and the email is shorter.

What changes for this format

The method is the same everywhere. These are the places where this format bends it.

The ask goes first, and the model buries it

A professional email exists to get a reply, a decision, or an action. The model treats the ask as the thing you build up to, after a greeting, a pleasantry, context, and an apology for the context. Your recipient reads on a phone between meetings and wants to know in the first line what you need and by when. Direct it: the ask in sentence one or two, the deadline attached, the subject line a version of the ask. If the model keeps burying it, write the first sentence yourself and have it draft only what follows.

Relationship history the model cannot see

The assistant does not know that this is your third reminder, that the recipient was promoted last month, that the last call was tense, that she replies to short emails and ignores long ones, or that you owe her a favor. It writes to a stranger, politely, from zero. Two lines of history in the brief change the draft more than any instruction about tone: third reminder, deadline now has a real cost, she prefers one-line replies. If the history is too sensitive to type, that is a sign the email is one to write yourself.

Over-apology and over-thanks

I apologize for any inconvenience. Sorry to bother you again. Thank you so much for your patience and understanding. Please do not hesitate to reach out. The model stacks these because they are the most common sentences in its training email, and because being agreeable scores well. In a real relationship they read as nervous or insincere, and they take up the space the ask needed. Ban the stack in the brief. One thanks, if it is earned. Apologize only for something you actually did, once, specifically.

Length discipline

The model writes 180 words where 50 would do, and the hard email should be shorter than the easy one. Give a word ceiling in every email brief, and make it tight: 80 words for a follow-up, 120 for a no, 150 for a request that needs context. Then run the test: can the recipient reply from their phone with one word? If the email needs a paragraph in reply, it probably needs a call instead, and the email should say so in two lines.

Confidentiality: what never gets pasted

A chat assistant is, in most set-ups, a third-party service, and what you paste may be stored or reviewed under terms you have not read. That rules out salaries, health information, legal matters, client data, anything under NDA, and the full text of the other person's email if it contains their private details. Use placeholders: [CLIENT], [AMOUNT], [THE ISSUE]. Summarize the incoming email in a line instead of pasting it. Fill the placeholders in by hand in your mail client, after. If your organization provides an approved tool with its own terms, those terms decide. If it does not, assume the strict version.

Tone drifts to warm professional

You have an email voice. Maybe you sign off with just your initial, never use exclamation marks, and open with the person's name and a comma. The model's voice is warm professional, which is nobody's. Give it three of your own sent emails as a sample (with details removed) and tell it to match them, or run the Rewrite Without Losing Your Voice prompt on its draft. Keep your own sign-off. It is the one line the recipient has seen a hundred times and will notice if it changes.

The process

01You
Decide the ask and the reply you want
Write two lines before the chat. What you need, by when. What the ideal reply looks like, word for word if you can: approved, Thursday works, yes send it. If you cannot write these, the assistant cannot help yet, because the missing piece is a decision. Many hard emails are hard because the writer has not decided what they are asking for, and the model will cheerfully draft around that gap for 200 words.
02You
Give two lines of history, with placeholders for anything sensitive
Add what the model cannot see: how many times you have asked, what changed, how the recipient reads email, what the relationship can bear. Replace every name, figure, and confidential detail with a placeholder. Summarize the email you are replying to in one line rather than pasting it. This block is short and it is where most of the draft's quality comes from.
Send this
```
Draft one email. Ask: [THE ASK, WITH THE DEADLINE]. Reply I want: [THE IDEAL ONE-LINE REPLY]. History the draft must respect: [TWO LINES OF HISTORY]. Recipient: [ROLE AND HOW THEY READ EMAIL]. Rules: the ask in the first two sentences; no hope this finds you well, no apology for writing, one thanks at most; under [WORD LIMIT] words; use the placeholders as written, do not invent names or figures. Subject line first. Then wait.
```
03AI proposes
Draft with the Email mega prompt and a word ceiling
The Email mega prompt carries the rules above as a saved brief, so the per-email turn is just the ask, the history, and the ceiling. Ask for one draft with the subject line first. Expect the draft to be structurally right and tonally a notch too smooth. That is the right failure: it means your editing is cutting rather than rebuilding.
04You
Cut the openers and the apology stack
Delete the first sentence if it is a pleasantry. Delete every apology that is not for something you did. Reduce the thanks to one or none. Move the ask up if it slipped. Most hard emails lose a third of their length here without losing anything the recipient wanted. Then read it as the recipient, on a phone, between meetings, and ask whether they can reply with one word.
05AI proposes
Bounded edit for the one paragraph that is wrong
Usually one paragraph has the wrong tone: too stiff, too soft, sounds like a complaints department. Name the effect rather than the adjective, and ask for that paragraph only, as OLD and NEW. Keep your master copy in the mail client, not the chat, so you are pasting the model's sentences into your email and never the reverse.
Send this
```
Change only the second paragraph. The problem: it reads as a complaint about the delay, and I need it to read as a plain statement of what the delay costs and by when I need an answer. No apology, no softening words. Keep the first and last paragraphs exactly as they are. Show OLD and NEW, then wait.
```
06You
Fill the placeholders, check, send
In your mail client, replace every placeholder with the real name, figure, and date. Check the recipient's name against the address line. The model may have guessed a spelling. Check every date against a calendar. Add your own sign-off. Read it once more aloud, and send. For a reply to a hostile email, wait ten minutes first. The assistant will not have talked you out of anything, and the pause might.

Toolkit

The prompts that fit this format, and when to reach for each.

Before and after

A paragraph the way an assistant tends to return it, and the same paragraph after a writer took it back.

AI draft

Hi Daniel, I hope this email finds you well. I wanted to follow up on my previous emails regarding the approval of the homepage designs. I completely understand that you have a lot on your plate, and I apologize for reaching out again. However, we are approaching the deadline, and I wanted to ensure we stay on track. Please do not hesitate to let me know if you have any questions or concerns. I would greatly appreciate it if you could review the designs at your earliest convenience. Thank you so much for your time and understanding.

Human revision

Daniel, I need a decision on the homepage designs by Thursday, or the launch moves from the 28th to the week of November 10. The 28th is also the print date for the trade show handouts, which use the same hero image, so it is a hard date. If the designs are fine, reply approved and I will take it from there. If something about them is bothering you, tell me what, even roughly, and I will have a revision to you within a day. Either answer is fine. Silence is the only one that costs us.

What changed

The ask moved to the first sentence with a date and a consequence attached, both of which the model could not have known because the writer had not put them in the brief the first time. The history, two earlier reminders, shows up as the Thursday deadline and the last line rather than as an apology for writing again. The apology stack is gone: no hope this finds you well, no sorry for reaching out, no thank you for your understanding. The length is about the same, but every sentence does a job, and Daniel can reply with one word from his phone, which was the test the writer set before drafting.

Pitfalls

The ask is in the third paragraph.

Write the first sentence yourself and put the deadline in it. Add the ask in the first two sentences to the brief so the draft starts there.

The email reads as if to a stranger.

You gave no history. Add two lines: how many times you have asked, what changed, how the recipient reads email. If the history is too sensitive to type, write the email by hand.

Three apologies and two thank-yous in 120 words.

Cut all but one, and keep that one only if it is for something you did. Ban the stack by name in the brief.

It is 200 words and the recipient will skim it.

Set a ceiling in the brief and cut to it by deleting sentences rather than compressing them. Run the one-word-reply test.

You pasted the client's full email, with their numbers, into the chat.

Stop and check your organization's policy. From now on, summarize incoming email in a line, use placeholders for every name and figure, and fill them in by hand in the mail client.

The sign-off and the opening are not how you write.

Give the assistant three of your own sent emails to match, keep your own sign-off, and run Rewrite Without Losing Your Voice when the draft is correct but sounds like a template.

Questions

Is it safe to paste emails into an AI assistant?

Not by default. Most consumer chat tools are third-party services with their own retention terms, and your organization may prohibit pasting client, employee, financial, or health information into them. Use placeholders for every name and figure, summarize the incoming email instead of pasting it, and fill details in by hand afterward. If your organization provides an approved tool, its terms decide.

How do I get the assistant to write shorter emails?

Give it a number rather than an adjective. Under 80 words produces a short email. Concise produces 160. Then cut the pleasantries and the apology stack yourself, which usually removes another third. If the email cannot get under the ceiling, the ask may need a call rather than a message.

Should the recipient know I used AI to write the email?

For routine professional email, where the facts and the ask are yours and you edited the draft, a disclosure is not expected and would be odd. The ethical line is elsewhere: never let the assistant commit you to something you did not decide, and never send a sentence you could not defend as your own. If an email carries legal or contractual weight, it goes through whatever review your organization requires, with or without AI.

Can the assistant help me reply to an angry email?

Yes, and this is one of its better uses: it does not get angry back. Summarize the complaint in one line, state the outcome you want and what you can actually offer, and ask for a draft that addresses the specific grievance without apologizing for things you did not do. Then wait ten minutes before sending, and read it once more as the person who was angry.

Next in CommunicateVibe Writing for Reports, Proposals, and MemosVibe writing for reports, proposals, and memos: your decisions and numbers, a structure built around the reader's choice, and an executive summary last.

Keep going