Email Rewriting with AI
AI rewrites emails faster than you can draft them from scratch — but without a voice sample, the output sounds like every other AI-polished email: slightly formal, slightly hollow, and starting with "I hope this email finds you well." The fix is a two-sentence voice definition and two sample emails pasted at the top of every session. With those, AI output is immediately usable.
Establish Your Voice First
Build this preamble once. Paste it at the start of any email rewriting session.
You are rewriting emails in my voice. My communication style:
Voice: [3-5 CONCRETE ADJECTIVES — e.g., "direct, brief, professional without being cold, no jargon, no exclamation marks"]
Avoid: [LIST BANNED PATTERNS — e.g., "'I hope this email finds you well', 'please do not hesitate to', 'as per my last email', passive voice"]
Sample emails I have written and approved:
Email 1: "[PASTE A REAL EMAIL YOU SENT THAT REPRESENTS YOUR VOICE]"
Email 2: "[PASTE ANOTHER REAL EMAIL]"
Keep this voice for all rewrites in this session.
The sample emails are the most important part. AI matches the pattern of what you show it far more accurately than any description of style.
Three Core Rewrite Prompts
1. Tone Shift
[VOICE PREAMBLE]
Rewrite this email to be [TARGET TONE — e.g., "more direct and shorter" / "warmer without being informal" / "more assertive without being aggressive"].
Original email:
[PASTE EMAIL]
Keep: all factual content and the core request.
Change: only the tone and word choices that conflict with [TARGET TONE].
Do not change the structure or add new information.
The "keep factual content" + "do not add new information" constraints prevent AI from editorialising your message.
2. Length Reduction
[VOICE PREAMBLE]
Cut this email to [3 sentences / 50 words / one paragraph]. Keep the most important point and the specific ask. Remove everything else.
Original email:
[PASTE EMAIL]
After the rewrite: tell me what you removed and why each removed part was not essential.
The explanation of what was removed is useful — it sometimes reveals that AI cut something important, and you can add it back in one sentence.
3. Difficult Message
[VOICE PREAMBLE]
Rewrite this email. It delivers [DESCRIBE THE DIFFICULT MESSAGE — e.g., "a rejection to a vendor" / "negative performance feedback" / "declining a request from a senior stakeholder"].
My draft:
[PASTE YOUR DRAFT]
Rules:
- State the decision clearly in the first sentence. Do not bury it.
- Do not soften the message to the point where the recipient might misunderstand the decision.
- Do not apologise unnecessarily or excessively.
- Include one concrete next step or alternative if there is one.
- Maximum 4 sentences.
AI's default for difficult messages is to soften and hedge until the decision is unclear. The "do not soften" and "first sentence" rules counteract this directly.
Subject Line Generation
Write 3 subject line options for this email. Each should be under 8 words and make the recipient understand the email's purpose without opening it.
Recipient: [ROLE / RELATIONSHIP — e.g., "my manager" / "a cold outreach to a VP"]
Email content:
[PASTE EMAIL BODY]
For each option: why it works for this specific recipient.
Avoid: question marks (weak), vague openers like "Following up", all-caps.
Subject lines are higher-leverage than email body rewrites — a better subject line determines whether the email gets read at all.
Reply Generation
[VOICE PREAMBLE]
Draft a reply to this email.
Email I received:
[PASTE EMAIL]
My position: [STATE YOUR ANSWER / STANCE — e.g., "I agree to the timeline but not the budget increase" / "I cannot attend but can send a delegate"]
The reply should:
- Open with my answer directly (not with pleasantries)
- Give one sentence of context if needed
- State any conditions or next steps
- Be under 5 sentences
"My position" is the critical field. Without it, AI writes a neutral non-committal reply that does not actually answer the question.
Failure Modes and Fixes
Common AI email failures
- Softens bad news until the decision is ambiguous
- Removes specific details AI considers "unnecessary context"
- Adds corporate filler: "I look forward to hearing from you"
- Changes your specific numbers or dates to vague alternatives
- Opens with "I hope this email finds you well" unless explicitly forbidden
Always edit these yourself
- Opening line: AI default openers are generic — rewrite to match the specific context
- Sign-off: match your actual sign-off style; AI adds formal sign-offs regardless of relationship
- Numbers and dates: verify AI did not round or change specific figures
- The core ask: confirm it is stated as clearly as you intended
Checklist: Do You Understand This?
- Why are sample emails more effective than describing your voice in words?
- What two constraints prevent AI from editorialising a tone-shift rewrite?
- What rule specifically counteracts AI's tendency to soften difficult messages?
- What is the most important field in the reply generation prompt — and what happens if you omit it?
- Write a length-reduction prompt to cut a 200-word project update email to a single paragraph.
- Name three parts of an AI-rewritten email you should always check manually before sending.