Hey Reader,

It's not a secret that I write my emails with AI.

I have a whole process around talking to Claude that turns a 15-minute brain dump into the email that you read right now.

And no one was ever able to tell (until I told them).

Because most AI content you post has a watermark on it.

And once you see it, you can never unsee it again.

AI Watermarks.

You already know about the em dash in every second sentence.

Or the "It's not about X. It's about Y" structures everywhere.

Everything also comes in set of 3s. Faster, smarter, stronger.

Recently, models started loving the word "quietly", and now it's all over.

One of my clients told me that AI ruined em dashes for her. I feel the same.

So most people do what I did for years. They take the draft and scrub the AI out of it by hand.

A coach I know spent three focused hours cleaning up one offer doc, just to make it sound like her.

I used to make this mistake too.

Where I went wrong.

Some time last year, everyone started talking about voice documents.

Just show AI what you actually sound like, and it will start writing like you.

I thought a voice document would fix it, too.

So I gave Claude about 140k words of me.

• 5 newsletters
• 16 call transcripts
• 4 YouTube videos
• 5 voice sessions
• a secret ingredient I'll explain below

this created a voice document that absolutely made AI sound more like me.

But the AI watermarks were still ALL OVER and it drove me insane.

Then I found a skill called Humanizer.

it's a public set of rules that tells AI what AI actually sounds like, so it can prevent its patterns.

Fun fact: it is built on the work that Wikipedia editors do to prevent AI slop

So I merged those rules into my setup, and... damn!!!

Suddenly, all the AI content that I thought was super slick was flagged all over.

Claude literally asked me:

"Rob, this looks like it may have been written by AI. Are you sure you wrote this?"

Fuck.

It was right.

I thought they sounded like me, and they were full of AI tells anyway.

The secret ingredient.

Here's the deal: most content that you have produced over the last year or so is AI-assisted.

You may think it sounds like you, but it really doesn't.

So I asked myself... "Where am I, my truest, rawest self?"

And for me, these were my WhatsApp chats with my wife and my closest friends.

The second I gave the export to Claude, it ran with it.

That's the last place your real voice still lives, typos and half sentences included.

So now the setup is two files working together.

• The voice doc: says what to sound like.
• The anti-AI doc: says what to cut, and it checks every draft up to three times before I ever see it.

I even ran a blind test with myself, with AI-generated and original content, and I couldn't AI from myself reliably.

That's the bar for your content.

It has to sound like you, typos, profanity and all.

Grab the Humanizer skill from here: https://github.com/blader/humanizer​

Talk soon,
Rob

P.S. I just published the full Claude video editing workflow for absolute beginners.