Write Emails with AI That Still Sound Like You

Turn three bullet points into a polite decline.

Level 1 · Chat Free · 30 minutes · Updated Oct 31, 2026

Most of us spend a surprising part of the day writing emails, and the hard ones are rarely long. They’re the ones where you have to say no, push back, or ask for something without sounding rude. AI can draft those in seconds. The catch: a draft can sound polished and still not sound like you, promise something you never offered, or come across as cold to someone who knows you. In the next 30 minutes you’ll turn three bullet points into an email you’d actually send, and learn when to let AI write less.

Remember the intern: smart, fast, first day on the job. They write fluently, but they’ve never met your colleague, don’t know your history, and will happily fill gaps with friendly-sounding filler and promises. You bring the facts and the relationship. They bring the wording.

What you’ll get

What you need

• Claude, any plan, including free. Just a normal chat. Other tools work too; see the end.
• Three or four emails you wrote yourself and were happy with. Any topic. They show the intern how you sound.
• A real hard email to write. We’ll use this one throughout: a colleague from another team has asked you to take over organizing the quarterly team offsite, but you’re leading a product launch the same month. You want to say no, kindly, and offer something smaller.

Don’t paste in anything you wouldn’t want outside your company: customer details, salaries, confidential plans. For a hard email you rarely need them anyway.

The three rules

Part 1: One hard email, done properly (30 minutes)

Six steps, plus a few minutes to try it on a real email. Step 2 you only do once; after that, every email takes about five minutes.

Step 1. Write your points before you open the AI (3 minutes)

On paper or in a note, write three or four short bullets. Not sentences, not polite phrasing. Just the content.

• No, I can’t organize the offsite.
• Why: I’m leading the product launch that month and can’t give it the time it deserves.
• What I can do: review the agenda draft, or suggest someone who has done it before.
• I’d like to help with the next one.

This is the most important step. Everything the email says comes from here. The AI only decides how to say it.

Step 2. Show it how you write (4 minutes)

Open a new chat in Claude and paste three or four of your own emails:

Here are a few emails I wrote. Don't reply to them. Just study how I write: sentence length, how I open and close, how direct I am, words I use and words I never use.

[paste 3-4 of your emails]

Describe my writing style in 5 bullet points.

Read the five bullets. Do they sound right? Correct anything that’s off (“I never write ‘I hope this finds you well’”). Then copy the corrected list into a note. That’s your voice sample, and you’ll reuse it for every email from now on.

Step 3. Brief the intern (3 minutes)

In the same chat:

Now write an email in my style.

To: a colleague on another team. We get along well but don't work together often.
They asked: if I can take over organizing the quarterly team offsite.
What I want to happen: they understand it's a no, don't feel brushed off, and take me up on one of my offers.
My points (use only these, don't add new reasons or promises):
- No, I can't organize the offsite.
- I'm leading the product launch that month and can't give it the time it deserves.
- I can review the agenda draft, or suggest someone who has done it before.
- I'd like to help with the next one.
Length: under 120 words.

The line “use only these, don’t add new reasons or promises” does a lot of work. Without it, the intern often adds a helpful-sounding extra offer you never agreed to.

Step 4. Ask for two different approaches (5 minutes)

One draft makes it hard to judge. Two drafts with different strategies make it easy:

Give me two versions:
A) Short and clear: the no comes in the first sentence.
B) Warmer: start by acknowledging their request, then the no, then the offers.
Keep both under 120 words.

Pick the one that fits the relationship. With a peer you like, B is often right. With someone who sends a lot of requests, A protects your time better.

Step 5. Read it as the recipient would (5 minutes)

First, ask the intern to check itself:

Compare your draft with my points. List anything in the draft that isn't in my points: extra reasons, promises, dates, offers, or claims about how I feel.

Then read it yourself, out loud if you can, and check:

• Promises. Does it commit you to anything you didn’t offer? (“I’ll send you some names by Friday” is a promise.)
• Facts. Names, dates and the request itself, described correctly.
• Filler. Lines like “I completely understand how important this is” that you’d never say.
• Voice. Would your colleague believe you wrote it?

Step 6. Make it yours, then send (5 minutes)

Edit at least one sentence by hand. Usually the opening or the closing, since that’s where AI sounds most generic. Then send it.

Now try it on a real email in your inbox: paste your voice sample, write your points, and brief it like Step 3. You’ll see it takes about five minutes.

Why you can’t skip the checking

With emails, the risk isn’t mainly wrong facts. It’s how you come across. Researchers have measured that:

• Readers notice, and judge the sender. In a 2025 survey of 1,100 professionals, 83% saw a manager’s message as sincere when it used little AI help, but only 40% to 52% did when it used a lot. Professionalism ratings also dropped, from 95% to 69% to 73%. Grammar and editing help was generally fine; heavy help on praise, congratulations or personal messages was not (Cardon & Coman, University of Florida). People were also more forgiving of their own AI use than of others’.
• Just being suspected costs you. In experiments with AI-suggested replies, conversations became faster and more positive in tone. But people who were suspected of using AI replies were rated less cooperative, whether or not they actually used them (Hohenstein et al., Cornell, 2023).
• AI nudges your tone. The same research found AI suggestions pushed people’s language toward more positive emotion. That’s nice in a thank-you note. It’s a problem when you need to say a clear no.

Both studies were about how messages are perceived, not about mistakes in them. That’s why the checking in Step 5 is half facts and promises, and half “does this sound like me?”

Checklist

Before you hit send:

• Every point in the email came from my own bullets
• It doesn’t promise anything I didn’t offer: no extra deadlines, calls, or “I’ll send you…”
• Names, dates and their request are correct
• The main message (here: the no) is clear, not buried in politeness
• I deleted filler I’d never write myself
• I edited at least one sentence by hand
• Nothing confidential went into the AI tool
• If this email is mainly about the relationship (thanks, apology, feedback), I wrote most of it myself

What you just learned

On the surface you wrote an email. Underneath, you practiced giving context, the first step of the loop: your points, your reader, your goal and your voice. The intern only chose the words. You also practiced a light version of checking: catching promises you never made and phrasing you’d never use.

There’s no Part 2 this time. At Level 1 the only setup worth keeping is your voice sample from Step 2: keep it in a note and paste it whenever you start. (If you’d like Claude to remember it automatically, the document summary tutorial shows how Projects work.)

The same method works for: replying to a complaint, asking your manager for something, chasing an overdue invoice, giving a vendor bad news, and writing a short Slack or Teams message that needs care. Change your points and your reader; keep everything else.

Using a different tool?

The prompts work anywhere. Only the buttons change.

Features and plan names change often. This table was checked in October 2026.

Sources

• Is writing with AI at work undermining your credibility? (University of Florida, Aug 2025), on Cardon & Coman, International Journal of Business Communication, 2025
• Study uncovers social cost of using AI in conversations (Cornell Chronicle, Apr 2023), on Hohenstein et al., Scientific Reports, 2023
• Learn about Gemini features in Gmail (Google Help)
• Draft an email message with Copilot in Outlook (Microsoft Support)