How to Write Professional Emails with AI: Templates, Prompts, and Etiquette
Stop sending robotic AI emails. Here are the exact prompts I use in ChatGPT and Claude, 10 templates for common work situations, and the etiquette rules that keep you from looking lazy.
I used to spend 11 hours a week on email. That is not an exaggeration. I tracked it for a month. Eleven hours reading, drafting, editing, and second-guessing the tone of messages that most recipients would skim in 30 seconds and reply to in 10 words.
AI cut that to about 4 hours. But the first few months were rough. The drafts sounded like a robot who learned English from LinkedIn posts. "I hope this email finds you well." "I wanted to reach out." "Let's connect and explore synergies." Deleting those lines became its own time sink.
Here is what I learned after a year of using ChatGPT and Claude for professional email. The templates are battle-tested. The prompts are specific enough to give the AI something to work with. And the etiquette rules are the ones that keep you from looking like you copy-pasted a chatbot.
Why most AI-written emails sound terrible
The default ChatGPT email has a voice. You know it when you see it. Three tells give it away every time.
First, the throat-clearing. "I hope this email finds you well" is not a greeting. It is a space-filler that signals "a machine wrote this" faster than any spelling error would.
Second, the corporate enthusiasm. "I'd love to connect and explore potential synergies." Nobody talks like this. Nobody ever talked like this. The AI learned it from LinkedIn and now it cannot unlearn it.
Third, the vagueness. The AI drafts a polite, grammatically perfect email that says nothing specific. No names. No numbers. No context. Just a friendly shape with no content inside.
The fix is not a better model. GPT-4o, Claude Opus, Gemini 2.5, they all do this. The fix is giving the AI enough context that it does not need to reach for the filler.
The context formula that actually works
Here is the prompt template I landed on after about 50 iterations. It works across ChatGPT, Claude, and Gemini. The structure matters more than the model.
Role: You are writing an email from me to [person's name and title].
Context:
- Our relationship is [never met / met once / talked last week / old colleague]
- They [specific thing they did or said recently]
- I need to [specific goal: get an answer, book time, share info, apologize]
Constraints:
- Under [X] words
- No greeting filler. Start with their name or the point.
- Write like a smart colleague, not a corporate lawyer
- One specific compliment or observation about their work goes here: [insert it]
- End with a clear ask, not "looking forward to hearing from you"
Tone: [casual / direct / warm / urgent / apologetic]
The context section is where most people get lazy. They paste the prompt, type "write a follow-up email to a client," and wonder why the result sounds like a template. The AI needs to know who you are to this person, what just happened between you, and what you actually want. Without those three things, it fills the gap with cliches.
I keep a text file of context snippets for people I email regularly. Their recent LinkedIn posts. Articles they published. Deals their company closed. It takes 30 seconds to paste one in, and the email quality difference is not subtle.
10 prompts for the emails you actually send
These are copy-paste ready. Replace the bracketed parts with your details. I wrote them for ChatGPT but they work identically in Claude, which sometimes produces a more natural tone.
1. The gentle follow-up
I emailed [Name] on [date] about [topic]. No response yet. Write a follow-up email that:
- References the original email without guilt-tripping
- Adds one new piece of value (the article/link/insight I'll paste below)
- Ends with a low-effort ask: "Worth a look or should I stop bothering you?"
- Under 70 words
- Tone: casual, not needy
New value to include: [paste link or insight here]
2. The meeting request that does not sound like a sales pitch
Write an email to [Name], [Title] at [Company], asking for a 20-minute call.
Context: I read their [article/tweet/interview] about [topic]. I have been working on something related at [my context] and think there is a genuine overlap.
Structure:
- Opening line: specific reference to their work that proves I actually read it
- One sentence on why I am writing
- One sentence on what they get out of the conversation
- Close: offer two specific time windows next week, not a calendar link
- No: "I'd love to," "reach out," "pick your brain," "synergy," "excited"
- Under 90 words
3. The bad news delivery
Write an email to a client, [Name], explaining that [Project Name] will be delayed by [X days].
Context: [Brief explanation of what happened. Own the mistake if it is ours.]
Tone: direct, no corporate deflection. Use "I" not "we."
Structure:
- State the delay in the first sentence
- Explain what happened, briefly
- What we are doing to fix it
- New delivery date
- One concrete thing we are doing to prevent this from happening again
- No: "unforeseen circumstances," "out of an abundance of caution," "we apologize for any inconvenience"
4. The cold outreach that does not get deleted
Write an email to [Name] at [Company]. We have never spoken.
Their context: [Paste recent news about them: funding round, product launch, interview quote, job change]
My context: I [build/sell/make/write about] [thing] that [specific outcome].
Goal: Start a conversation, not book a demo.
Structure:
- Opening line: reference to their news that shows I did 60 seconds of research
- The connection: here is why their situation is relevant to what I do
- One sentence on what I do, framed as an outcome not a feature list
- The ask: a genuine question about their work. Not "do you have 15 minutes?"
- Under 80 words
- Read it aloud. If it sounds like something you would delete, rewrite it.
5. The progress update to your manager
Write an email to my manager, [Name], with a status update on [Project].
Accomplishments this week: [bullet list]
Blockers: [specific thing I need help with]
Next week's plan: [one sentence]
Structure:
- First line: the most important thing that happened, good or bad
- Bullet points for the rest. Nobody reads paragraphs in status emails.
- Blockers stated as "I need [specific thing] from you by [date] to keep moving"
- Close with the decision I need them to make, if any
- Tone: direct, solutions-oriented, not performatively busy
6. The introduction between two people
Write an email introducing [Person A] to [Person B].
Why: [A needs something B can provide. Or they share an interest. Be specific.]
About Person A: [one sentence that matters. Not their bio. Why they are interesting.]
About Person B: [same thing.]
Structure:
- Subject line: "A <> B: [shared interest]"
- One sentence on why I am introducing them
- One sentence each on why they should care about the other
- I step out: "[A], [B] — over to you two"
- Under 60 words
7. The complaint response that actually helps
Write a response to [Name] who is unhappy about [specific issue].
What happened: [explain the problem honestly. No spin.]
What we are doing: [specific fix, with timeline]
Tone: apologetic without groveling. Confident without being dismissive.
Structure:
- Thank them for flagging it and apologize in the first sentence
- Summarize the issue back to them so they know we understood
- State the fix and when it will happen
- Give them a direct contact: "Reply here or reach me at [phone/email] if you want to talk directly"
- No: "We take your feedback seriously," "We strive for excellence," "Your satisfaction is our top priority"
- Under 100 words
8. The "can you help me with something" ask
Write an email to [Name], a [colleague / former coworker / acquaintance], asking for help with [specific thing].
Our relationship: [how I know them, when we last talked]
What I need: [be specific. "Advice on X" not "pick your brain." A warm intro to Y, not "help with my job search."]
What they get: nothing except gratitude. Do not fake a "mutual benefit" if there is not one.
Structure:
- Acknowledge it has been a while, sincerely
- State the ask clearly in sentence two
- Make it easy to say no: "If you are swamped or this is not your thing, no worries at all"
- Under 70 words
9. The project summary for stakeholders
Write a summary email for [Project Name], Phase [X] completion.
Audience: [executives / cross-functional team / clients]
Metrics: [2-3 key numbers: on time? on budget? key result achieved?]
Structure:
- Subject: "[Project] — Phase X done. [Key metric.]"
- First sentence: the headline result
- 2-3 bullet highlights
- One honest challenge we hit and how we solved it
- What Phase [X+1] starts on and what it needs from the reader
- No: "We are thrilled to announce," "moving forward," "leveraging synergies"
- Under 120 words
10. The resignation or goodbye
Write a goodbye email to my team at [Company].
My tenure: [X years / months]
What I am proud of: [one specific thing we did together. Not "the culture." A project, a win, a moment.]
Why I am leaving: [one sentence. Keep it positive. "An opportunity to work on [X]" is enough.]
Tone: warm, grateful, not sentimental.
Structure:
- State that I am leaving and when
- One specific memory or thank-you
- What I will miss
- Where to find me: [LinkedIn / email / whatever]
- Under 100 words
- No quotes. No "it's been a journey." No "bittersweet."
The etiquette rules nobody tells you
I learned some of these the hard way. Others I watched colleagues learn the hard way.
Never let AI send the email. This is not a purist stance. It is risk management. AI hallucinates prices, deadlines, and promises. A chatbot that offers a client a 50% discount to "make things right" is not the chatbot's fault. It did what language models do. The inbox is yours. The send button is yours.
Disclose when it matters, stay quiet when it does not. If you are emailing your manager about project status, you do not need to say "this email was drafted by Claude." If you are emailing a grieving client or a legal counterparty, the human touch is the whole point. Use judgment. The rule of thumb: if the recipient would feel deceived knowing an AI wrote it, write it yourself.
Specific compliments only. "I loved your recent article" is worse than no compliment. It signals that you did not read it. "Your point about churn prediction failing on small datasets matches something I have been seeing with our 200-user accounts" is a compliment that starts a conversation. The AI cannot invent the second one. You have to feed it in.
Read every draft aloud. If you stumble over a sentence, rewrite it. If a word feels too big for the situation, swap it. If the whole thing sounds like a press release, start over with more context and fewer constraints. The read-aloud test catches more AI weirdness than any editing checklist.
Short emails get more replies. My highest-response-rate emails are under 60 words. The AI defaults to 150 to 200 words because it thinks thoroughness is politeness. It is wrong. Busy people want the subject line to tell them whether to open it, the first sentence to tell them what you want, and the rest to be optional.
Do not use AI for anything that creates legal obligation. Pricing, contract terms, delivery commitments, refund promises. If it could end up in a deposition, write it yourself. Or at minimum, have a lawyer review the AI draft before it goes anywhere near a client.
When not to use AI for email
Some emails should come entirely from you. The condolences. The apology for a serious mistake. The response to someone who just got laid off. The message to a person you mentored for three years.
The reason is not that AI writes these badly. It is that receiving an AI-written version of these messages feels disrespectful in a way that is hard to articulate but easy to recognize. The recipient knows. They might not know they know, but they feel it.
If you are tempted to use AI for one of these anyway, write the draft yourself first. Then ask the AI to suggest improvements. Do not ask it to generate the first version. The starting point matters. An AI-polished human draft sounds like a thoughtful person. A human-polished AI draft sounds like someone outsourcing their humanity.
FAQ
Q: Claude or ChatGPT for email writing?
A: Claude produces a more natural, peer-like tone. ChatGPT has better integrations and a more flexible interface. I use Claude when the email needs to sound like a specific person. I use ChatGPT when I am drafting a sequence and want to iterate fast. Both are good enough that the prompt quality matters more than the model choice. If you already pay for one, use it. If you pay for neither, try both free tiers and pick whichever voice you edit less.
Q: Can the recipient tell I used AI?
A: If you use the prompts in this article and edit the output for at least 30 seconds, probably not. If you copy-paste whatever the chatbot generates first, yes. The default AI email voice has tells. The recipients who can spot them tend to be people who also use AI tools. That group is growing.
Q: Should I tell my boss I use AI for email?
A: Depends on the boss. Some managers will be impressed that you found a way to spend less time on email. Others will hear "I automated my communication" and wonder what else you are automating. Read the room. The safer approach: produce good emails in less time and let the results speak.
Q: Are AI email tools like Lavender or Superhuman worth it on top of ChatGPT?
A: If email is 30% or more of your job, yes. Lavender's tone coaching and reply prediction are genuinely useful. Superhuman's speed matters if you process 100+ emails a day. For everyone else, ChatGPT or Claude plus the prompts above is plenty.