Write emails with AI and still sound like yourself

To make AI-written emails sound like you, give the assistant samples of your own writing and a short voice card that describes how you write, state the audience, goal, tone and length, and treat what comes back as a first draft that you edit.

Why do AI-written emails sound generic?

An assistant does not know how you write. It does not know that you open with the point, that you never write "delighted", or that you and this reader have traded short notes for two years. So it falls back on an averaged style of business writing: a warm opener, a few filler phrases, a vague closing line.

The result is a little too polite, a little too long, and full of stock phrases such as "I hope this finds you well". A cleverer one-line prompt will not fix that. The assistant is missing how you sound, who you are writing to, and what should happen next.

How do you teach an AI your writing style?

Collect three to five short emails you sent and were happy with, ideally from different situations: a quick reply, a request, a no. Strip out names, figures and anything private. Paste them in and say what they are for: "These are examples of my writing. Match the style only. Do not reuse any facts from them."

Then write a voice card: a short note on your habits that you can paste into any conversation. Cover where the main point goes, how you open and close, which phrases you use or would never write, and how formal you are. A voice card might read: "I write short, direct emails. The main point goes first. I use contractions and plain words, and I say 'happy to' instead of 'delighted'. I am friendly but not gushing: no exclamation marks, no 'I hope this finds you well'. I sign off with 'Best,' and my first name."

Keep the card in a note and paste it at the start of each conversation. Some assistants have a saved-instructions setting for this. It is the same idea as a system prompt: standing instructions for everything that follows. Whenever you catch yourself deleting the same phrase from a draft, add it to the card.

What should you tell the AI before it writes the email?

The voice card says who you are. The brief says what this email must do. Give four things every time:

• Audience. Who reads it, and what they already know.
• Goal. The one thing you want them to do or understand.
• Tone. For example friendly and direct, or formal and brief.
• Length. A sentence count or a rough word limit.

This is the better AI prompts checklist applied to one kind of message. With no stated goal, an email tends to come back as a pleasant paragraph that asks for nothing.

What do good AI email prompts look like? Three examples

A follow-up. You sent a prospective client a proposal on 3 November and have heard nothing.

• Brief: "Follow up with Marcus about the proposal I sent on 3 November. Ask whether he has questions and when he expects to decide, and offer a call if that is easier. Friendly, not pushy. Four sentences. Use only these facts."
• Draft: "Hi Marcus, I'm checking in on the proposal I sent on 3 November. Do you have any questions I can answer? When do you expect to decide? I'm happy to talk it through on a call if that is easier. Best, Sam"

A decline. You have been invited to speak on a panel and cannot.

• Brief: "Decline Dana's invitation to the 14 December panel. Say no plainly and thank her. Give no reason. Two sentences."
• Draft: "Hi Dana, thank you for thinking of me for the 14 December panel. I can't take it on this time. Best, Sam"

The "give no reason" line matters: an assistant asked to decline can soften the no until it is unclear, or add an excuse such as a full calendar.

A request. You want time with a colleague on another team.

• Brief: "Ask Priya, on another team, for 20 minutes next week to see how her team builds its monthly report. I am building a similar report for my own team. I am free Tuesday to Thursday. Goal: a yes or no and a time. Three sentences."
• Draft: "Hi Priya, could I borrow 20 minutes next week to see how your team builds its monthly report? I'm building a similar report for my own team. I'm free Tuesday to Thursday if any of those days works. Best, Sam"

Every fact in each draft, the names, dates and the 20 minutes, came from the brief. That is deliberate.

How do you edit an AI-written email?

Fix the opening and closing first, since generic phrasing hides there. Then ask for a revision with a specific instruction, such as "shorter, drop the second paragraph, less formal", not "make it better". If a few rounds of revision still miss, write the email yourself.

Before you send, read it out loud. Your ear catches what your eye skips. Stop wherever you stumble, and ask:

• Would I say this to their face? If not, change it.
• Does the first sentence sound like me? If it sounds like a form letter, rewrite it.
• Is there one clear ask? If a reader could finish and not know what to do, sharpen it.

How do you stop AI from inventing facts in an email?

An assistant does not know what you promised, when the invoice is due or whether the meeting moved. If you do not supply a fact, it may supply one for you, a habit explained in why AI makes things up. In an email that is costly: a made-up "I will have it to you by Monday" reads as something you chose to say.

Put every fact, date, amount and name into the brief yourself, and tell the assistant to add none. If it needs something you did not give it, ask it to write "TBC" in that spot instead of guessing. Then check the draft line by line against your source, and delete any promise you did not decide to make. If the email recaps a meeting, first turn the meeting notes into action items and check them, then write from that list.

What should you not paste, and which emails should you write yourself?

Do not paste anything you would not want stored. Whether a product keeps chats or uses them to improve its models varies, so check the settings. Remove names, account numbers, health details, passwords and other people's private information, and use placeholders such as "Client A" that you replace afterward. The AI assistant privacy checklist lists what to look for.

Some messages are better written by you alone: condolences, serious apologies, resignations, difficult feedback, and anything with legal, medical or financial consequences. Check workplace or school rules on AI-written text as well.

Where NAMELESS stands

NAMELESS is an AI model that is still in training and not yet available, so none of this can be tried on it today. It is built on our own architecture, with unlimited context and no rereading of the conversation. It is extremely fast, with stronger logic and reasoning, at a cost of pennies compared with today's leading assistants. To follow progress, join the waitlist.