Two years ago we passed AI-written posts around like it was a magic trick. Look what it wrote in nine seconds.

Last week LinkedIn shipped a button so you can report those posts as “slop”.

This is a full circle moment, in under two years. Which means the most valuable writing skill right now is sounding like a human. How ironic.

People ask if I write this newsletter or if an agent does. The answer is both. My ideas never show up at a convenient time, usually when I'm driving or my hands are full, so lately I've been using Wispr Flow to catch them before they disappear. Then AI helps me shape the mess into something you can read.

AI is in every step. But It has never written an entire sentence of this for you. Fine, the Companion pages I share are a different story. It's better at those than I am.

The reason so much writing gets labelled as AI writing has nothing to do with the writer. It's just how the thing works.

If AI is supposed to be this smart, why does everything it writes end up sounding the same? Can't they fix that?

Think about the last place you moved into. New condo, new apartment, does not matter. The walls were most likely beige. Somewhere along the way, we all agreed this was the acceptable color for a room nobody has lived in yet, and the reason is that beige is safe. Nobody walks in and hates it. Nobody walks in and loves it either.

AI writing comes out of the same logic. The model was trained on close to everything ever written, the good, the bad and the truly terrible, and then it was taught to give an answer that works for anyone who might be asking. So it heads for the middle of the pack every time. It reaches for the safest word, the one most likely to come next, and when it isn't sure what you want, it adds more words instead of fewer. That's where the padding comes from.

It runs on the old motto: if you can't amaze them with your intelligence, confuse them with your BS.

So the beige is not a bug they forgot to fix. It's the default setting, and it stays that way until you tell it who you are.

Here’s the fun part: this is where you make your writing sound like you again.

Quick note before we start, because I know some of you are already thinking this is for people with a business. You have a voice, you have a brand whether or not you have a company. If you write emails, comments, posts, applications, anything at all, you have a brand and this works the same for you.

Now I'm going to be annoying about one thing. If you haven't created your Hot Start doc yet, stop reading and go do that first, this is the link. When you're done, reply to this newsletter and I'll buy you a coffee next time I see you 🙂

Then go find one piece of writing where you thought, that was sharp. An essay, an email, a long text you sent when you were fired up about something. You might have to dig, you might have to ask your mom to go through the archives and send you a picture of it, but trust me, this is the part that makes the whole thing work. AI can follow your rules all day, it can only match your style if it has seen it.

💡One setting first. In Claude, open Settings, then Capabilities, then switch on Code execution and file creation. Without that toggle the Skills menu doesn't show up at all, and this is where most people quietly give up. Detailed instructions in the companion page.

Now open a new chat, attach your Hot Start doc and your writing sample, and paste this:

Save what it gives you as a Skill, and here's the part that makes this different from keeping your rules in a doc somewhere. You never have to call it. It fires on its own, every time you ask for writing help, whether you remembered it or not.

Test it tonight. Ask Claude to write you a short email to someone you've been avoiding, and see whether it comes back sounding like you or sounding like beige.

One more thing while we're here, because you're going to start seeing Skills everywhere now. Before you install one somebody else made, paste the link into Claude and ask it to tell you exactly what the thing does. Better yet, read what it does and build your own version. Takes ten minutes and you know what's in it.

☑️ Linkedin now has a button for reporting you. On July 30 LinkedIn added a button that lets anyone flag a post as "seems like AI slop." The reports feed their models and decide how far your post travels outside your own network, and at the same time LinkedIn killed its own "enhance your post" tool and replaced it with one that proofreads without changing your voice. The platform that spent two years pushing you to write with AI decided that sounding like yourself is the feature. If you post on LinkedIn, this is the week your voice started affecting your reach.→ Read it here

🪲 Anthropic just started scanning Skills for malicious content. Enterprise accounts can now switch on automatic scanning that checks any third-party Skill or plugin for malicious content the moment someone uploads or edits one. A Skill runs instructions on your behalf, so one you downloaded from a stranger gets treated like any other software you install on your machine. Be careful where you download from. If you're on a personal plan there's no scanner watching for you, so you are the scanner. Paste the link into Claude, ask it to tell you exactly what the thing does, and if the answer is longer or stranger than the description promised, skip it and build your own. → Read it here

📇 Want to take it one step further. Ruben Hassid published a trick over the weekend that I have not tested yet, and I will this week. Type "use ASD-STE100" into Claude (or ChatGPT) and it starts writing plainly. That's an international standard for writing technical manuals, the kind aerospace uses, and the model already knows it. The catch, and he says it himself, is that it's not good for everything. He tried it on a LinkedIn post and it came back cold. So it's a fast way to strip the slop out, not a way to ensure your voice is captured, which is the whole reason you built the Skill tonight. I'll report back on how it goes.→ Read it here

I sat in on a training last week with James McAulay, through The Neuron, and one thing he said stuck with me. Only about 1 in 4 people using AI have ever tried Skills.

Not 1 in 4 people on the internet. One in four of the people already using AI every day, already paying for it. The rest have never touched the thing that makes it actually sound like them.

You get to be the one.

That's the part nobody tells you about being early. It doesn't feel like being early. It feels like being behind, because everyone online sounds like they've been doing this for years. They haven't. Most of them are still pasting answers straight into the post.

So build the Skill tonight. Then hit reply and tell me one word AI uses that you can spot from a mile away. I'm collecting them, and I'll run the list in a future issue.

See you next week. ✌🏼

Cinz.AI

Your unfair advantage, one week at a time.

Got a question, a topic request, or just want to say hi? Just reply to this email. I read everyone.