How to train AI on your own voice (in 3 steps)

Happy Sunday,

The single most common complaint I hear about AI content is “it doesn’t sound like me.” And it’s true, by default, it sounds like the blended average of the entire internet, which is to say, like nobody.

But here’s what most people miss: you can teach it. Your voice isn’t some mystical thing, it’s a set of patterns, and patterns are exactly what AI is good at learning. Today I’ll show you the 3-step process to build a reusable “voice profile” so everything you generate from now on sounds like you.

IN THIS EMAIL

» Why your AI content sounds like everyone else’s » The 3-step voice profile » The voice-extraction prompt to steal

🗣️ Why it sounds generic

AI has never met you. When you ask it to write “in a casual tone,” it reaches for the most average version of casual it knows, and so does everyone else who types those words. That’s why so much AI content has the same smooth, lifeless texture, everyone’s pulling from the same default.

Your actual voice is made of specifics: the words you reach for, the ones you’d never use, your sentence length, your rhythm, whether you’re blunt or warm, whether you use slang or keep it clean. The model can absolutely write in that voice, but only if you show it what that voice is. Vague instructions get vague results. You have to make the implicit explicit.

📝 The 3-step voice profile

Step 1: Collect your samples. Pull together 5 to 10 pieces of writing that genuinely sound like you, captions, emails, even long voice-note transcripts. Quality over quantity. You want stuff where you sounded like yourself, not where you were stiff and “professional.”

Step 2: Have AI extract the rules. This is the magic step everyone skips. Paste your samples and ask the model to analyze your voice and write it back to you as a set of explicit rules, tone, sentence length, vocabulary you use and avoid, quirks, rhythm. The model is great at spotting patterns you can’t see in your own writing. You’ll read it and go “yeah, that’s actually how I write.”

Step 3: Save it and reuse it. Take that extracted voice profile and keep it somewhere you can paste it instantly, a note, a saved prompt, your tool’s custom instructions. From now on, every time you generate something, you lead with the profile. The output stops being generic because the model finally has a real spec to write to, instead of guessing at “casual.”

That’s the whole thing. Ten minutes of setup once, and every piece of content after gets the benefit.

🔒 Steal this prompt

Below are several samples of my writing. Analyze them and write me a detailed “voice profile” I can reuse: my tone, typical sentence length and rhythm, words and phrases I tend to use, words and styles I clearly avoid, and any quirks that make my writing recognizably mine. Write it as a clear set of rules another writer could follow to sound exactly like me. Here are the samples: [paste 5 to 10].

Run that, save what it gives you, and paste it at the top of your prompts going forward. It’s the closest thing there is to cloning your own writing hand.

This week: gather five things you’ve written that actually sound like you, and run them through this. Even just reading your own voice profile is weirdly clarifying.

• László

P.S. When my team takes over content for a brand, capturing the founder’s real voice into a profile like this is step one, it’s how we keep everything sounding human at scale instead of like a robot. If you want that built for your business, that’s Growth Lab Studios. Start here.

P.P.S. That’s the back catalog for now. Every issue lives in the newsletter archive, and I’ll see you in the next one.