How to keep your voice when you use AI to write LinkedIn posts

The risk with AI drafting is not that any single post comes out bad. It is drift. Every pass nudges the text a little closer to the average of everything the model has read, and the average is not you. Do it for a month and your feed reads like a competent stranger has taken over the account, which is a real cost when the whole point of posting is that people feel like they already know you before the first call.

Drift is gradual, which is why it slips through. No individual edit looks wrong. You accept a smoother sentence, a tidier transition, a more balanced ending, and each one is a small improvement in isolation. Twenty of them in a row is a different writer.

Use the model for structure, not for sentences

Asking "which paragraph is weakest and why" or "what is missing from this argument" gets you real value without touching your phrasing. Asking for a rewrite hands over the exact layer where voice lives. Structure is fairly universal, so a model is genuinely good at it. Sentence rhythm is personal, so a model is genuinely bad at it, at least at being yours.

Feed it your own writing, not a description of it

A model given ten of your real posts produces something far closer to you than a model told "write in a conversational but authoritative tone." Descriptions of voice are lossy, and adjectives especially so, because every writer thinks they are conversational. Examples carry the details you cannot articulate: how you open, how long you let a sentence run, which words you never reach for.

Keep a folder of ten to fifteen posts you would happily publish again, and reuse it every time. It costs nothing after the first hour and it is the single highest-leverage habit in the whole workflow.

Keep a list of words you never use

A negative list beats a positive one, because it is unambiguous. "Never use leverage, unlock, journey, or game-changer" is enforceable. "Sound authentic" is not. Ten banned words will remove most visible drift on their own, which is a surprisingly large return for a one-line note.

Read it the next morning

The wince test is reliable and almost free. Anything that makes you slightly uncomfortable to read under your own name is drift, and the discomfort is information rather than perfectionism. If a sentence would embarrass you slightly in a client meeting, it will do quiet damage in a feed.

Why voice is worth protecting at all

People do not buy from a LinkedIn account, they buy from the person they think is behind it. Consistent voice is what makes a stranger feel they already know how you work before they ever email you. That familiarity is the actual asset you are building, and it is the thing that quietly erodes when every post sounds like it came from the same generator as everyone else's.

Where Voice Profile fits

Voice Profile reads the posts you have already published and writes down what it observes: your sentence rhythm, your openers, the words you avoid. When an observation is wrong, you flag it, and it stops being applied. It is a way of making your voice explicit enough that a tool can respect it, instead of hoping it survives another round of editing.