Helping AI drafts sound like the person sending them

I'm building Inboundy, a LinkedIn automation tool. One part I've been working on is giving people more control over how their outgoing messages sound.

The starting point is simple: describe your business and add two or three examples of messages you would actually write. Those examples give the AI a reference for tone and structure. The business context gives it something concrete to work with.

Here's an illustrative example. A freelance consultant might describe the service they offer and provide a couple of short, straightforward introductions. That gives the system a more useful starting point than a general instruction like “write a friendly message.”

Then comes the review step. You can read and edit the generated messages before sending them. If a sentence feels awkward or says something you wouldn't say yourself, you can change it directly.

I want that combination to make reviewing drafts more useful: provide context up front, then keep the individual message editable.

I recorded the review workflow here:
https://www.youtube.com/watch?v=I_gEbKF81Qg

For people building AI writing features, what information has been most useful for making a draft sound like its sender?