You can usually spot AI writing in a second. The endless “in today’s fast-paced world,” the tidy three-item lists, the relentless cheerfulness — it reads like a brochure written by a committee. The problem isn’t the tool. It’s that most people ask AI to write something without ever telling it how they sound. Train it properly and the output stops sounding like a robot and starts sounding like you on a good day. Here’s the exact process I use across every business I run.
Feed it samples before you ask for anything
The single biggest mistake is starting from a blank prompt. AI has no idea what your voice is, so it defaults to the bland average of everything it has ever read. Fix that by giving it real examples of your writing first — three or four pieces you’re happy with: an email, a couple of social posts, a paragraph from something longer. Don’t ask it to write yet. Just hand it the samples and ask it to study them.
Turn your voice into a reusable style guide
Once the AI describes your voice, get it to compress that into a short, named style guide you can paste into any future chat. This is the asset that saves you from re-explaining yourself every single time. Ask for concrete rules, not vague adjectives — “short, punchy sentences with the occasional long one for rhythm” beats “engaging tone.” Save the result somewhere you can grab it in seconds.
Show it what you hate, not just what you like
Voice is defined as much by what you avoid as what you do. If you can’t stand em-dash-heavy hype, exclamation marks, or words like “leverage” and “unlock,” say so explicitly. A short banned-words list does more to fix the “AI smell” than any amount of positive instruction, because those tics are exactly what the model reaches for by default. Add them straight into your style guide so they carry forward.
Edit out loud and feed the corrections back
The first draft after all this will be close but not perfect. Instead of silently rewriting it, mark up what feels off and tell the AI why. “This intro is too formal,” “you’d never use the word vibrant,” “break this run-on into two.” Each correction teaches it more about you, and within a few rounds the gap closes. This is the same iterative loop I use everywhere — I broke down the full method in my prompt engineering framework for better AI results.
Keep a living voice file you reuse everywhere
Don’t let that style guide die in a forgotten chat. Save it as a note, a saved prompt, or a Custom GPT / Claude Project so it’s loaded automatically every time. Once it’s permanent, every email, caption, and article starts from your voice instead of the generic default — which is exactly how I keep a consistent tone across 20+ platforms without writing each one from scratch. If you want to systematise the rest of your workflow too, see how I automate repetitive tasks with AI and the AI tool stack I use to run my businesses solo.
Use your voice, not the AI’s, as the final filter
Here’s the part no prompt can do for you: read the finished piece aloud. If a sentence makes you wince or you’d never actually say it to a friend, cut it — no matter how polished it looks. AI gives you a fast, on-brand first draft; you give it the final 10% of judgement that makes it unmistakably yours. Get that loop right and you’ll produce more, sound like yourself, and never again publish something that reads like a robot wrote it.