AI Assistant for Email Drafting That Still Sounds Like You
You asked an AI assistant to write a quick reply to a client, and what came back opened with "I hope this email finds you well" and ran to four paragraphs.

You asked an AI assistant to write a quick reply to a client, and what came back opened with "I hope this email finds you well" and ran to four paragraphs. Using an AI assistant for email drafting can save real time, but only if the draft sounds like you and says what you meant. This guide shows a simple routine that gets usable drafts in under two minutes, the prompts that work, and the kinds of email you should still write yourself.
Why AI assistant email drafts sound robotic
The assistant is not trying to sound stiff. It is filling gaps. When you type "reply to this saying no politely", it has to guess your tone, your relationship with the reader, how long the reply should be, and what you are willing to offer instead. With no information, it falls back on the safest, most generic business letter it can produce.
That is why the fix is almost never a better tool. It is better input. Our guide to giving an AI assistant context it can actually use covers this idea in general; for email, it comes down to four details.
- Who the reader is and how well you know them
- The one outcome you want from the email
- The length, in sentences, not in vague words like "short"
- A sample of your own writing so it can copy your voice
A two minute routine with an AI assistant for email drafting
This is the routine I use for most everyday replies. It works in ChatGPT, Claude, Gemini, Copilot and the assistants built into Gmail and Outlook.
- Paste the email you are replying to. Remove anything private you do not want to share, such as account numbers or a third party's health details.
- Write your answer as messy notes. For example: "no to Thursday, can do Monday 10 or Tuesday 2, send the revised quote by Friday".
- Add the constraints. "Reply to Dana, a client I have worked with for two years. Friendly but direct. Maximum 5 sentences. No greeting cliches, no 'I hope this finds you well'."
- Read the draft out loud. If a sentence is one you would never say, cut it or rewrite it.
- Check every fact. Dates, prices, names and promises. The assistant will happily invent a delivery date that sounds plausible.
The notes in step two do most of the work. The assistant is good at turning rough points into clean sentences. It is bad at deciding what you should say.
Prompts that work for common email types
Here are prompt patterns that consistently produce drafts needing only light edits.
Declining a request
"Draft a reply declining this request. Reason: we are fully booked until March. Offer to revisit in February. Keep it warm, 4 sentences, no apology more than once."
Following up when nobody answered
"Write a short follow up to my email below, sent 6 days ago with no reply. Do not guilt the reader. Ask one clear yes or no question. Under 60 words."
Explaining a delay
"Draft an update telling the client the project slips from 12 May to 19 May because the supplier sent the wrong parts. State the new date in the first sentence. Offer a 15 minute call. No excuses beyond the one reason."
Matching your voice
Paste three emails you have sent recently and add: "Study the tone, sentence length and sign off in these emails. Write the reply below in the same style." This one change does more for natural sounding drafts than any other trick. Save the samples in a note so you can paste them in seconds.
If you want to go further, our piece on how to ask an AI assistant better questions explains why specific, constrained prompts beat open ones.
The advice I disagree with: let AI write everything
Some productivity guides suggest routing every email through an assistant, even one line replies. I think that wastes time and slowly makes your writing worse.
A reply like "Monday at 10 works, see you then" takes eight seconds to type. Opening an assistant, pasting context, prompting and editing takes a minute at best. Across 40 short replies a day, that is half an hour lost, not saved.
There is also a quieter cost. Readers notice when every email from you has the same polished rhythm. People who receive a lot of AI drafted email start skimming them, and your important messages get skimmed too. Save the assistant for emails that take you more than three or four minutes to write by hand: long explanations, delicate refusals, messages in a second language, and anything where you keep rewriting the first line.
Emails you should still write yourself
Some messages should not start as an AI draft at all, even with heavy editing.
- Apologies for a real mistake. A generic apology reads as no apology. Say what went wrong in your own words.
- Condolences and personal news. People can tell, and it matters to them.
- Anything legal or contractual. An assistant may phrase a commitment you did not intend. Get proper advice instead.
- Emails containing confidential data if your company has not approved the tool. Check your workplace policy before pasting client information into any assistant.
For everything else, treat the draft as a first version from a fast but careless colleague. Keep the structure, fix the facts, and put your own sentences back where it smoothed them away.
Checking the draft before you hit send
A short checklist catches nearly every problem I see in AI email drafts:
- Does the first sentence say the main point?
- Are all dates, times and amounts correct and in the right time zone?
- Did it add a promise you did not make, like "I will send this today"?
- Is it at or under the length you asked for?
- Would the reader recognise it as you?
If any answer is no, edit by hand instead of re-prompting. A two word fix is faster than another round trip.
Frequently asked questions
Which AI assistant is best for writing emails?
For most people, the best one is the assistant already inside their email client, because it can read the thread without copying and pasting. Gmail and Outlook both offer built in drafting tools. Standalone assistants give more control over tone when you paste in samples of your own writing.
Can people tell if an email was written by AI?
Often, yes, especially when the draft uses stock openings, long paragraphs and overly polite closings. Giving the assistant samples of your own emails and a strict sentence limit makes drafts much harder to spot. A final read and a few edits in your own words help most of all.
Is it safe to paste work emails into an AI assistant?
It depends on the tool and your employer's policy. Many companies allow approved business versions that do not train on your data, while consumer versions may store conversations. Remove sensitive details such as account numbers or health information unless the tool has been cleared for that use.
More in AI Assistants
AI Assistants
How to Get an AI Assistant to Cite Sources You Can Check
I asked an assistant for the Swedish rules on invoice retention periods and it gave me a clean answer with a reference: a named...
AI Assistants
AI Assistant Custom Instructions That Are Worth Writing
Most people start every chat with an AI assistant by retyping the same three facts.
AI Assistants
AI Assistant for Spreadsheet Formulas That Actually Work
You know what the spreadsheet should do, you just cannot remember whether it is XLOOKUP or INDEX MATCH, and the forum answer...