How to Make AI Write in Your Own Voice
- Hiske van den Akker
- 4 days ago
- 4 min read
You typed out a whole paragraph explaining how you write. Warm but direct. Conversational. A bit funny. Not corporate. You pressed enter, and what came back was something a marketing department would put on a poster.
So you rewrote it yourself, at half past eleven at night, again.
Here is what went wrong, and it is not you. You described your voice. Every one of those words means something different to a language model than it means to you. "Conversational" could be a podcast host, a customer service script, or your neighbour. "Direct" could be blunt or rude. You handed it a mood board and asked for a portrait.
The fix is almost embarrassingly simple. You stop describing and you start showing.
Step one: give it your actual writing
Find three to five pieces of your own writing where you thought, that is exactly how I would say it. Emails, captions, a voice note you transcribed, the long message you sent a client at midnight. Paste them in whole. Then say one thing: match the voice in the examples above.
Pick the pieces where somebody replied and said "this is so you". Those are your best samples, because another human already confirmed it. If you have fifteen or twenty of those, even better, though five is plenty to start.
Do not clean them up first. The typos, the tangent halfway through, the place where you started a sentence and changed your mind, all of that is the texture. Polish it and you have handed over someone else's writing again.
Step two: tell it what you never say
Every writer has a list of words that make them wince. Mine has about forty on it. Journey, unlock, elevate, tribe, and any sentence built as "it's not just X, it's Y". Write yours out and hand it over as a rule, not a suggestion.
This one thing removes most of what makes writing sound generated. Not because the model is bad at writing, but because it defaults to the average of everything ever published, and the average is beige.
Step three: correct it like you would correct a person
The first draft will be close and not right. This is where nearly everyone gives up and decides AI cannot do it.
Do not rewrite it yourself. Tell it what is off. "That opener is too clipped, say the whole thought." "You made me sound apologetic, I am not." "Cut the last line, it is a slogan, not something I would say." Then have it try again.
Three or four rounds of that and something shifts. You are not prompting anymore, you are editing, and editing is a thing you already know how to do.
Step four: save it, so you never do this again
The part people miss. Once it sounds like you, do not leave it in a chat window that disappears. Put the samples, the banned words and the corrections into one document, and start every session from it.
That is the whole difference between a clever trick and a tool. You teach it to sound like you once. After that it just works, and Monday morning stops being the day you stare at a blank caption box.
The layer almost nobody adds
Everything above will get you most of the way, and it is what every guide on this topic will tell you.
Here is the piece I have not seen anywhere else. Your writing voice is not only a style. It is a rhythm, and your rhythm comes from how you are actually built.
I read Human Design charts, and once you have looked at a few hundred you cannot unsee it. A Manifestor writes in short declarative bursts and gets irritated by throat-clearing. A Projector writes to one person at a time, and her writing goes flat the second she tries to broadcast to a crowd. A Generator warms up over three paragraphs before she says the real thing, so her best line is usually the one she wrote fourth. If you have emotional authority, everything you write in a spike reads too hot the next morning, and everything you write in a low reads apologetic.
So when you build this thing, add your type and your authority to the document, and give it one instruction: never let me send something written in a spike without sitting on it first.
That is not decoration. It is the difference between a tool that writes like you on a good day and a tool that also stops you sending the thing you would regret. You do not need to be a Human Design expert to use this. You just need your chart, which takes two minutes to pull.
What this actually buys you
Not content. Your evenings.
You are the coach, the copywriter, the content person, the tech department and the one who is supposed to have a strategy. The tools that fix this got genuinely good about two years ago, and nobody sat any of us down and taught us how to use them. That is not a discipline problem. Count your job titles.
One evening of setting this up properly, and then click click click, you are done for the week and you go back to living your life.
If you want the whole thing written out
I made a free guide called Your Voice by Design that walks through this step by step, including the exact prompts and the Human Design layer. No code, nothing to install, and it works in the free version of ChatGPT or Claude.
And if you would rather build it with me, live, one build at a time, that is Lead by Design. Building your own AI team is the fastest way back to the work only you can do.
Comments