Method 1: A style prompt you write once
Works in ChatGPT, Gemini or Claude. It takes ten minutes and works well for occasional emails.
- 1Copy five of your own sent emails into a document: one to a colleague, one to a client, one short reply, one longer note, one where you said no.
- 2Ask the assistant to describe your writing style from them: greeting, sign-off, sentence length, formality, words you use often, things you never do.
- 3Edit that description until it is right, then save it as a custom instruction or keep it handy to paste.
- 4When you need a reply, paste the thread and your style description, and say what you want to say in one line.
Where the prompt method breaks
You still copy and paste every thread. The style drifts if you forget to paste the description. And the assistant does not know which emails need a reply in the first place, so it saves you writing time but not tracking time.
Method 2: A tool that learns from your sent mail
Several Gmail tools now build a style profile from your sent folder and draft replies in the background: Fyxer, Superhuman's Auto Drafts on its Business plan, Inbox Zero, and Momo. The draft lands in Gmail Drafts, inside the thread, and you send it from Gmail.
- Momo reads up to 25 of your sent emails from the last 180 days, strips the quoted history, and stores a profile of style only: tone, greeting, sign-off, length and habits. It is encrypted and rebuilt weekly.
- Code picks who the reply goes to. The model writes only the body.
- Momo's drafts leave [placeholders] for anything it cannot know, and are not allowed to commit you to a decision or a deadline.
- Drafts you never touch are removed after three days, so your Drafts folder stays clean.
Check every draft for these four things
Whichever method you use, an AI draft is a first draft. Before you send, check:
- 1Facts: dates, numbers, names and anything the draft says happened. Models fill gaps confidently.
- 2Commitments: did it promise a deadline, a discount or a call you did not agree to?
- 3Recipients: is anyone added or missing? A reply-all you did not intend is the most expensive AI mistake.
- 4Tone: would you say this to this person? Cut the first sentence if it is a pleasantry you would never write.