Sentence rhythm turned out to be a stronger voice signal than vocabulary how long someone runs their sentences, where they hedge versus assert. Early drafts were accurate but felt weirdly formal. Adjusting syntax patterns made a bigger difference than swapping words.

The most interesting engineering insight from the last six months: sentence rhythm is a stronger voice signal than vocabulary. Two people can use the same words and sound completely different. Getting rhythm right is what takes a persona from "close" to "that's actually them."

Before asking a single creator to trust MePersona with their voice, I wanted to feel every rough edge myself. So for the past month I've been dogfooding our own Persona API using it to help draft and voice the marketing you're reading right now.

The idea behind MePersona is simple: reply tools fail when the fan can tell it wasn't really you. So it learns how you actually talk, then drafts across all your platforms in that voice. The fastest way to know if that works was to point it at myself first.

A month in, it's caught things I'd never have found in a demo. We're opening to a small founding group before launch follow along, and I'll share what dogfooding your own voice actually teaches you. Question for other builders here: would you trust an AI to post as you, or does that line feel uncrossable?