Ask an AI assistant to reply to an email and you usually get something correct and lifeless. It opens with thanks, closes with an offer to help further, and uses three sentences where you would use one. The person reading it can tell.
That is not because the model writes badly. It is because it has never seen you write. Here is how to close the gap.
Why drafts sound generic
A model with no examples falls back on the average of all business email. The average is formal, hedged and long. Your own email is none of those things in the same proportions. You have a usual greeting, a usual sign-off, a typical sentence length and a few phrases you lean on. You also change register depending on who is on the other end.
Voice is mostly those small habits. Miss them and the draft reads as someone else.
Four things that carry your voice
- Length. If your replies are two lines, a five-line draft is wrong before anyone reads a word.
- Openings and sign-offs. "Hi Sam," and "Sam," are different people. So are "Best," and "Thanks, P".
- How direct you are. Some people lead with the answer. Others lead with context. Readers notice when it flips.
- Who you are writing to. You do not write to your co-founder the way you write to an investor. A single "tone" setting cannot capture that.
Doing it by hand
If you are using a general assistant, you can get most of the way with examples.
- Paste three or four emails you sent to the same kind of person and say "match the length, greeting and sign-off of these".
- Give rules as limits, not adjectives. "Under 60 words, no opening pleasantry" works. "Be concise and friendly" does not.
- Tell it what you would never say. A short banned list removes most of the tell-tale phrases.
- Edit the draft, then paste your edited version back as the next example.
This works. It is also a lot of pasting, and you have to redo it for each kind of recipient.
Letting the tool learn it
The alternative is a tool that reads your sent mail and builds the profile for you. That is what Pranan does. It analyses the email you have already sent to learn sentence length, vocabulary, greetings, sign-offs and how your tone shifts per recipient, then applies that profile each time it drafts. Voice matching starts from the mail already in your Gmail, so drafts use your voice from the start and get closer as Pranan learns from the drafts you edit or reject.
The profile is yours alone. It is not blended with other accounts, and no model is trained or fine-tuned on your email. More on that here.
You still read before you send
A draft that sounds like you is not the same as a message you have agreed to. Read every draft as if a colleague wrote it for you. Check the facts, the names and the promise you are making. Pranan leaves drafts in Gmail for exactly this reason: you review, edit if needed, and send it yourself.
If a draft is close but the register is off for this one reply, ask for a warmer, shorter or more formal version rather than rewriting it. Tone tuning covers how.