A blog post and a Slack DM to a close colleague should feel like the same person wrote them.
One is 900 words with evidence and structure. The other is two sentences. But the thinking, the conviction, the things you’d never say — those should be identical. Because the person didn’t change. The context did.
This is where most voice personalization tools get it wrong. They extract a single profile from your message history and call it “your voice.” But your message history contains at least six distinct registers: DMs to people you’ve worked with for years, emails to executives, external partner updates, published writing, technical handoffs, and Slack threads in public channels. Averaging them produces output that’s recognizably shaped like you — and consistently wrong for the context.
The person is the constant. The mode is the variable. The medium is irrelevant.
What belongs in a single universal layer: the quality standards, the cliché guard, the conviction, the things you’d never write regardless of who’s reading. These don’t change by context. They’re who you are.
What belongs in mode-specific profiles: register, sentence length, opening patterns, vocabulary choices, what you’d never say in this context specifically. A “no worries if not” close is wrong in a leadership email but fine in a casual DM. A credential opener is wrong everywhere but for different reasons in each mode.
The architecture that works keeps the person constant and makes the mode a variable the infrastructure resolves automatically from context — recipient, channel, intent. No manual selection.
One voice. Six modes. The person is always the same.