TUNE IN : APPLE PODCASTS | SPOTIFY
A little while ago I got publicly accused of being a robot.
Here’s the scene. I was sitting in an airport lounge, about to board, and realised there was a gap in the posting schedule. So I did the very rare thing and wrote a post myself. Thumbs going, boarding call in the background, fire it off, done.
A few hours later the comments were waiting. We hardly ever get comments, so a couple of DMs and comments is a big day for us. And almost every one of them said some version of “haha, we can tell this one’s AI.”
The post I’d actually written by hand, in an airport. Every other post that week had been written by our AI, and nobody blinked.
Either I’ve read so much AI content that I’ve started writing like it, or my AI is now so well trained on my voice that it sounds more like me than I do. Honestly, could be either.
But it got me thinking about the opposite problem, the one most people actually have. It doesn’t matter how good your content is. The second someone gets that little AI-ick reading it, their brain checks out and they stop trusting it, even when what’s being said is genuinely good. If you feel that reading someone else’s content, your audience feels it reading yours.
Most people either shrug that off or give up and go back to writing everything by hand. There’s a third option. Because if you’re feeding a tool a one line prompt and hoping for magic, of course it’s going to sound like everyone else’s AI content. That’s not an AI problem, it’s a setup problem. Here are the three pieces of ours.
Sweeping Out Every AI Tell
The first piece is a skill trained on everything that screams “a robot wrote this.” The em dashes. The endless bullet points. The way every piece of content ends up structured exactly the same. And the words… there’s a whole list of them that give it away the second you start noticing, and more keep turning up.
The part that actually makes it work is where it sits in the process. It’s not something to remember to run. It’s built in as one of the very last steps, sweeping through right at the end once everything else is written and approved.
It’s also a learning skill, which means it doesn’t stay static. Once a month it goes off, researches the newest batch of AI tells doing the rounds, and updates its own naughty list. From then on it’s working off the latest version, without anyone having to feed it manually.
Training It on How You Actually Speak
The second piece is the big one, and it’s the difference between AI content that’s fine and AI content that sounds like you specifically.
Most people train their AI on how they write. This one was trained on how we actually speak, using a stack of real client calls to analyse tone, rhythm, and language, and set its own rules from there.
The clever part is how it keeps learning. At the end of a content session, its job is to look back at the feedback it was given and pull out the intent behind it, not the one-off fixes. Train it on every single small correction and you end up with a pile of disjointed rules that make it sound less human, not more.
And here’s the part worth calling out: it was deliberately trained to notice the bits that are grammatically “wrong.” The imperfect, slightly-off phrasing that’s actually just how a real person talks. Instead of fixing those, it was told to protect them, because that’s exactly the kind of nuance that makes content sound human instead of scrubbed clean.
Catching the Drift Before It Shows
The third piece is the one that saves the most time day to day, because even a well trained AI drifts the longer a session runs. Write a whole month of content in one sitting and by the end it’s absorbed so many tangents and side conversations that it slowly wanders away from how it’s supposed to sound.
The fix is a set of trigger words. Instead of typing out the same feedback for the fiftieth time, dropping in a single code word tells the AI to stop, read back what it just wrote through the eyes of the ideal client, flag where it landed and where it lost her, score how likely she is to take the next step, and rewrite with all of that in mind.
It’s a fast way to catch and correct drift before it ever reaches an audience.
If you tried to build all three of these one at a time, haphazardly, it would probably slow content down rather than speed it up. But that’s not how it’s meant to work. Set it up once properly, and it runs quietly underneath everything that gets made from then on. That’s the difference between having a handful of clever prompts and actually having a content department
Click here to listen to the full episode
Want AI Content That Finally Sounds Like You?
This entire setup… the cleanup skill, the voice training, and the trigger words… is exactly what we walk you through inside Claude Like a CEO, including the step-by-step setup wizard so you’re not starting from a blank screen trying to describe your own voice.
It’s $150 off right now, for a very limited time.
