How to Get ChatGPT to Write Emails That Sound Like You

Why the first draft sounds like everyone

Ask ChatGPT to "write an email to my client about the delay" and it has nothing to go on except those words. So it writes the most likely email: polite, padded, evenly paced, and the same email it would write for anyone else who asked.

There is no magic prompt that fixes this. What works is giving the model the two things it does not have, how you write and what actually happened, then telling it what to leave out and finishing the last bit yourself.

1. Show it how you write

Describing your style rarely works. "Friendly but professional" means nothing to a model, and everyone describes themselves that way. Examples work much better.

Open your sent folder and pick three to five emails you wrote yourself, ideally to the same kind of person you are writing to now. Paste them in and say they are examples of your style. The model will copy your length, your greetings, whether you use contractions, how you sign off and the words you actually use.

Take out anything you would not want an AI service to see first: client names, prices, personal details. Swap them for placeholders before you paste.

2. Give it the facts

Most of what makes an email sound real is detail. The model cannot know about the call last Tuesday or why the files are late, so it fills the gap with general phrases. Tell it:

• Who the email is to, and how well you know them
• What happened, in a sentence or two
• What you want them to do, and by when
• Anything they are likely to push back on

3. Tell it what not to do

Models have strong habits. Name the ones you do not want:

• No "I hope this email finds you well" or any other opener before the point
• No closing line offering to answer questions
• No em dashes
• No bullet points unless you ask
• Use contractions
• Keep it under 120 words

A list of banned habits works better than a list of adjectives, because the model can check each one.

4. Put it in one prompt

Here is a prompt you can copy and fill in.

The last rule matters. Asked for detail it does not have, a model will sometimes make it up: a date, a figure, a promise you never agreed to.

5. Save the rules so you write them once

ChatGPT has custom instructions, under Personalization in its settings, that apply to every chat. Put your banned habits there: no em dashes, no openers, contractions, short emails. Claude and Gemini have similar settings.

This matters more than it sounds. In November 2025, OpenAI's CEO Sam Altman said ChatGPT now follows a no-em-dash rule when it is set in custom instructions. A request in the middle of a chat is easier for it to forget.

6. Ask for shorter, then edit

If the draft is still too long, say "half as long". Models almost never write too little.

Then edit it yourself. This is the step that makes the email yours. Add one detail only you know, cut any line you would not say out loud, and check every name, number and date. Five minutes of editing beats twenty minutes of re-prompting.

What a good prompt cannot fix

Even with examples and rules, a model slides back to its habits over a long chat, and it is easy to stop reading closely once the first few drafts were good. The slips are the same every time: an opener sneaks back in, a "delve" appears, the sign-off turns formal. Our list of ChatGPT words to avoid in email covers what to look for.

Check it against your own writing

Presend checks every email when you click send. Its voice check compares the draft with emails you have sent yourself and tells you when it does not sound like you, with the sentences that drifted underlined. The comparison runs in your browser, so your writing never leaves it. A separate check flags AI phrasing like the openers and closers above, so you catch them on the days you skip the edit.