The only way to make writing sound Human

Last month, I tried to put my own course out of business.

My Copy Hunte Field Guide for Writing Like a Persuasive Human in the Age of AI is a 52-hunt editing system. Every hunt is a rule: a specific pattern you find in your draft, the word “thing,” the hedging adverb, the AI-flavored opener, and a specific move you make when you find it.

Every hunt is documented, numbered, and extremely teachable.

Which raises an uncomfortable question, one I figured I should ask before a customer did: if the whole system is rules, why can’t an AI just run it?

So I fed the hunts to the machine and told it to edit.

What came back taught me more about writing than building the course did.

You’ve typed it. I’ve typed it.

Make this sound more human.

And the machine obliges in the way it was trained. Em dashes vanish. A contraction shows up. A little fragment gets deployed, like a garnish. And yet, somehow, the result still reads like a machine, only now it’s a machine doing an impression of a person, which is somehow worse than the machine just being itself.

Here’s why that prompt can never work.

An AI writes toward the most probable next word. Which means it is, by construction, an averaging device, the mean of everything everyone has ever written. So when you ask it to “sound human,” it produces its best statistical estimate of the average human.

The problem is that no human sounds like the average human.

Your writing sounds human for the opposite reason: because it deviates. And because behind every sentence there’s evidence of a specific person making specific choices, some of them, crucially, against the rules.

But don’t get me wrong.

Good writing follows the rules.

But great writing knows where to break them.

And yes, I know that is cliché.

Everyone nods at that line.

But nobody tries to do it intentionally.

Or think about what that means, which is that a broken rule only reads as a voice if the surrounding writing proves you know the rule. A fragment from a writer in control of their sentences is a choice. The same fragment from a writer who isn’t is a typo. The difference isn’t in the text at all. It’s in the judgment that the text is evidence of.

The problem is that you can teach a rule to anyone or anything in an afternoon. And in time, people can become good at it. But judgment about when to break it is not a pattern. It’s not even really a judgment. It’s a person just writing about interesting things the way they do. And there is no prompt for a person.

Back to my experiment.

The machine ran the hunts flawlessly. Every “thing” was replaced with the noun or idea it was hiding. Every hedge got deleted. Every em dash interrogated. It had one hundred percent compliance on every hunt, every time.

A perfect score.

And what it produced was horrible.

Because its perfect score was the problem.

SO I ran it again and told it to be less perfect.

And the result was even worse.

When a writer runs Hunt 18, the hunt for the word “thing,” they cut nine and keep one because it kinda works there. Because, you know, it was inside a dialogue or the sentance kinda just needed a shrug. For reasons that, for reasons they may not be able to articulate, belong. But the machine, on the other hand, keeps zero. Or it leaves one for a reason it could fully explain.

Perfect compliance and reasoning are the biggest AI tells of all because a draft with every rule followed reads like a form filled out.

Hand the same 52 hunts to a hundred writers, on the other hand, and you get a hundred different writers, each sharper, cleaner, or more distinct from the others, not less. Because each one, based on an infinite number of factors, breaks rules in different places. Each one keeps a different “thing,” or spares a different hedge, or leaves a different rule unfollowed in a different paragraph.

That’s what makes Hemingway different from Faulkner.

Your tiny inconsistencies are your fingerprint.

And the machine can’t produce that fingerprint because it can’t fail selectively. You can’t help but.

The system works because you can’t run it perfectly.

If you could, it wouldn’t.

Seventeen of the 52 hunts are uploaded now, with the rest releasing as I finish them. When the course is complete, it will be $250.

While it’s in pre-sale, the code PRESALE gets you in for $75 (which includes everything that’s up now, everything still to come).

If you buy it, or even if you just start hunting on your own, you’re going to do it imperfectly.

You’ll miss instances.

You’ll keep the hedge you were supposed to cut.

You’ll break a rule in a paragraph where I would have followed it.

And I want you to know that this is a good thing.

Because that uneven, judgment-riddled execution is the only part of the process a machine can’t do, which makes it the only part your reader can’t get anywhere else.

So learn to run the hunts.

Practice them until they become as intuitive as they have become for me.

And don’t worry if you miss a few.

Nobody else will.

In fact, they will probably love your writing even more for it.

-Justin

P.S Yes, I see it: a man selling an editing course announces that AI can’t replicate his editing course. Which is convenient for me. All I can tell you is I tried to automate myself out of the job first, and I’m reporting back that I failed.