The AI Email Assistant That Learns Your Writing Style

August 24, 2026 · 4 min read

Most AI email tools write the same polite, slightly generic reply for everyone. The tenth email sounds exactly like the first, and none of them sound like you. MailReply takes a different approach: it treats every reply you approve as a lesson.

How the memory fills up

You don't train anything and you don't write a "style prompt". You just use it:

1. You approve a reply: insert it or copy it. That final version, including every edit you made, is what the assistant learns from.
2. It extracts your habits: how you open, how you sign off, how long you like to keep things. If you consistently shorten what it writes, it notices and starts writing shorter.
3. The next reply starts from your style: recent approved replies guide the tone and shape of every new draft, on top of the tone you picked.

Regenerating counts too: it's the signal that a draft wasn't right, so the assistant weighs your approved versions more than its own first attempts.

You can see everything it learned

Open Settings and the "What MailReply learned" card shows the replies it's learning from and the habits it picked up, like a preference for short answers. One click on Reset and the memory is wiped: no hidden profile, no black box.

Want to go further? Instruction files

On the Max plan you can attach up to three .md or .txt files: a company tone guide, your personal writing rules, terminology to always use. Every reply follows them, on top of what the assistant has already learned naturally.

Why this beats a "style prompt" in ChatGPT

A pasted prompt is static: you write it once and it never notices that your emails changed. A memory built from real approved replies stays current by definition: it learns from what you actually send, not from what you once said you wanted.