I Can't Prove My Writing Is Mine Anymore. Here's What I Can Prove.
You don't lose authorship in one dramatic sale. You lose it one un-read sentence at a time.
The email is two sentences. I’d have called it one, which tells you how fast I read it. I’m standing in the kitchen, phone in one hand, and I say it out loud, to nobody in particular:
OH, HELL NAW.
Here it is:
is everything you write ai generated these days?
if not, just fyi, reads like slop.
The lowercase is his. So is the fyi, which – and I need you to appreciate this – is doing a lot of work for three letters. “oh hey, I can be a jerk but i’m coming from a place that looks helpful” (I would know. I have an fyi of my own I owe somebody. We’ll get there.)
This isn’t a stranger off the internet. I’ve known him since the blog days – back when you paid actual money for a domain name and posted about your feelings into Moveable Type and a small, weird community of people you’d never meet IRL actually gave a shit. He works in AI now. The actual models part, not the people who say they’re part of the AI movement because they write listicles called “10 Ways AI Will Change Everything.”
So when my face works again after the temporary but seething blinding rage, I copy his one sentence into my Obsidian journal – where I keep the stuff I’m not ready to be publicly embarrassing about yet – and I write underneath it: obviously this has me on the defensive. instead of being reactive, how can I be better?
Look at me. So measured. So grown.
Narrator, in British accent: “He would then spend the next three hours being neither of those things.”
The thing you do
The thing you do, of course, is question whether everything you’ve published HAS been AI slop. You can’t just be offended. You have to find out if he’s right first. Which means you can’t sit there and stew – you have to go back and reread everything you’ve published in the last two weeks.
Look. I’ve been publishing on the web since 1999. I have a whole bit about AI being the Adderall of the writing world: useful for getting something on the table, addictive as hell, and real easy to convince yourself what it wrote for you is you. I also have ADHD bad enough I once went a full year without writing anything, because the blank page is a terrible place to start a story when your brain refuses to give you a single drop of dopamine for building a paragraph. So I use AI as a tool. I’ve made my peace with using the tool. The line I thought I was walking was simple: AI does the boring scaffolding, I write the actual piece.
I know what slop looks like. I can feel it when I’m writing it.
So I reread. The old stuff I’d already flagged – that’s the easy category. There was a phase early on where you can practically see the novelty oozing off the prose; that specific brand of AI enthusiasm that reads like someone describing what a good blog post should contain instead of just writing one. Fine. Caught those a while ago. Embarrassing, but caught.
What I wasn’t ready for was the other category.
The words
Not “delve.” Not the em-dashes. Not “it’s not X, it’s Y,” which, yes, fine, the entire internet has agreed to be very annoying about.
These were words. Plural. A cluster of them.
I found phrases scattered through several posts – confident, technical-sounding vocabulary – that I cannot prove I ever used before there was an AI in the room. I’m not going to list them. The second I do, you’ll either nod like you’ve always known them or you’ll clock them as tells, and I don’t fully trust either of us right now. They were the kind of words you absorb from a smart magazine and just assume belong to you. You nod along while you’re editing. You think: maybe I know these words. Maybe this is just me.
Maybe this is me.
I sat with that one for a while.
The half he was right about
Here’s the fyi I owe you.
The email he got wasn’t a blog post. It was outreach – I’d been emailing a loose circle of people about freelance and consulting opportunities, and apparently I had gotten a little aggressive about it. The message he got, the one that read like slop: I didn’t write it. Not exactly. I drafted it with an AI agent I’ve been experimenting with, and then, per my own instructions, the agent sent it. On my behalf. Probably without any of the anti-slop rules I so carefully apply to everything else.
I handed the whole thing over. The drafting, the voice, the send button. I was juggling a lot of caregiving stuff at the time, I went hard on the automation, and I told myself that was just efficiency. Uhmm… oops.
So he’s half right. The half he’s right about isn’t my blog. It’s the email. It’s the fact that I let a machine talk as me to someone who’s known me for twenty years, and the machine got close enough that he couldn’t tell – except for the part of him that could, the part that fired off one lowercase sentence and then cause that teeny, tiny, low-key mental spiral you’re witnessing right now.
What you can actually prove
What I keep turning over is whether it even matters that I can tell the difference.
You can’t prove your writing is yours. Not really. Same rhythm, same vocabulary, same indefensible fondness for the en-dash – there’s no test you pass that a good model can’t also pass.
But you can prove it isn’t.
When the machine invents a word and you don’t catch it. When the agent hits send and you never read it. When you’re three paragraphs into defending a sentence and you can’t remember if you wrote it or just approved it.
That’s the part nobody tells you. You don’t lose authorship in one big dramatic sale. You lose it one un-read sentence at a time, and every single one felt like efficiency when you did it.
And it didn’t get me where I was watching for it. Not in the posts I reread four times. In the email that wasn’t even a post.
I wrote him back. Tried to be all casual. hey. you mean the email you recieved? or everything i’ve written lately?
Lowercase. Matched his. Didn’t even notice I’d done it until just now.
He hasn’t written back.
And here’s the part I can’t shake: I read that little message of mine four times before I sent it. To make sure it sounded like me. To make sure it sounded casual, which is a thing you only have to engineer when it isn’t.
Same trick the agent pulled. I just did it by hand.
There’s a companion piece coming in Late to the Future about the words. The mechanics of how it gets past you, and how you actually catch it. But Ernie, you say, aren’t YOU one of those people who write about AI through an AI Substack? Hey – what’s that thing in the sky?