It's not that AI writes badly. It's that it finishes before you do, and now there's a word for what that leaves behind.

Merriam-Webster picked "slop" as last year's word of the year. Everyone online seemed to already know exactly what it meant. 

Cheap, fast, forgettable content, the kind that fills a feed without ever landing anywhere. I read that and thought about a moment a few months back, when I asked an AI tool to draft a section of a newsletter I'd been circling for days. It came back in about four seconds. 

Clean sentences. Reasonable structure. Nothing wrong with it that I could point to. And I remember just sitting there, looking at it, feeling something I didn't expect. Not relief. Something closer to unease. It wasn't slop, exactly. It just wasn't mine yet.

Here's the part that took me longer to admit than I'd like. I almost used it anyway.

I've spent forty-five years watching technology promise to save people time. Spreadsheets saved time. Email saved time. Neither one ever asked me to hand over the thinking, too; they just moved the paper faster. 

This is different. I don't think enough people are saying so, mostly because they're too busy pasting the output straight into their newsletter and calling it a Tuesday.

An AI tool doesn't just save you the typing. It skips the part where you figure out what you actually think, and hands you something so tidy you almost feel rude questioning it. That part used to be unavoidable. 

You'd stare at a blank page, turn an idea over, throw out three bad versions before the real one showed up. It was slow, it was a little miserable, and I'm not going to pretend the misery built character. But it was the only way the thing you wrote ever became yours. Now you can skip straight to something that reads fine. 

The catch is that "reads fine" and "is actually yours" are not the same test, and four seconds is not enough time for anybody to figure out which one they got, including the machine that wrote it.

I'll say the quiet part out loud. I'm not anti-AI. I use these tools constantly, and if you're still bragging about never touching one in 2026, congratulations, you've found a hobby. But I've also watched a lot of otherwise smart people mistake "it came out fast" for "it came out right," which is the same mistake people made with fast food, except at least fast food admits what it is.

Everyone's using "slop" to describe AI content that's obviously bad. Broken hands in a fake photo. A blog post that repeats itself twice, in case you missed it the first time. An article that confidently cites a source that has never existed and never will. 

That's real, and it's a real problem, and it's also the easy problem. Anyone can spot bad slop. The word is giving people permission to feel safe about the version that isn't bad at all, and that's the one that should actually worry you.

The dangerous version reads fine. It sounds like you, or close enough that your mother wouldn't notice. Nobody flags it, and that's exactly why it's more dangerous, not less. 

Bad content gets caught. Content that's merely empty slides right through, because nothing about it trips an alarm, least of all yours, and you're the one person who was supposed to be checking.

What's missing is you. Not your byline. Your judgment. The particular way you'd have said it if you'd sat with it another ten minutes instead of hitting publish because dinner was getting cold and the draft looked close enough.

I spent time in the Navy as a sonar technician, which, contrary to what the movies suggest, mostly meant sitting in the dark listening to noise and trying to figure out whether it was a submarine or a very confused whale. 

One thing never stopped being true down there: information alone was never enough. The system would hand you a contact, a shape on a screen, a return in the water. It never told you what the contact meant. 

That part was on you. Get it wrong, and you either wake the captain over a whale, which is a fun way to make friends, or ignore something that actually mattered, which is a much worse way to end a career.

That's the closest comparison I've found for AI-generated writing. The tool hands you a return. Clean sentences, real structure, something that looks exactly like a contact worth trusting. It cannot tell you whether it's actually the thing you meant, or just the shape of it wearing your voice like a costume it found in your closet. 

That judgment is still entirely yours. A lot of people are skipping the step where they actually check, because the shape looks close enough and checking is annoying. Annoying is apparently disqualifying now.

I'd love to tell you I built an elegant framework for this. I did not. What I actually do is read the AI draft and ask myself one unglamorous question: would I have said it this way? Not "is this good." Not "does this sound right?" Would I have said it this way if every word had to come out of my own head instead of somebody else's four-second guess at me?

Most of the time the honest answer is no. Not because it's wrong, exactly. Because it's generic in a way I wouldn't have been, even on a bad day. It skipped the argument I'd have had with myself halfway through the paragraph. It landed on the safe version of the sentence instead of the true one. 

The safe version is what a model trained on the entire internet's collective, timid, hedge-everything opinions tends to produce. Once I notice that gap, I can't unnotice it. 

So I go back in and actually write the thing, using the draft as a rough map instead of a finished trip, usually while muttering something ungrateful about the tool I chose in the first place.

That takes longer than publishing the first draft. It's supposed to. If it doesn't take you longer, you skipped the part that mattered.

If you're building something online right now, a newsletter, a product, a small business, you are drowning in tools that can generate more content, faster, than you could ever produce by hand. At least four of them emailed you this week offering a "limited time" discount that resets every Monday. 

That's not hypothetical. That's just where we are in September of 2026. And the temptation, if you're tired and behind and staring at a deadline, is to let the tool finish the thought for you, call it done, and go make dinner while pretending you don't know the difference.

I understand the temptation completely, because I've given in to it more than once. Usually around 9 PM, when the newsletter's due in the morning and my judgment has clocked out for the day. But here's what I keep coming back to. 

The whole reason anyone trusts a writer is that something real is on the other end of the words. Not a persona. Not a brand voice you paid a consultant to develop and now have to perform. An actual person who thought about this long enough to mean it. 

The moment that stops being true, even quietly, even in ways only you can see, the trust erodes faster than any of us want to admit. It doesn't come back just because you apologize in the next issue.

Readers can feel the difference between something that was finished and something that was merely completed, even when they don't have the word for it. Now, apparently, they do. It's called slop, and it earned Word of the Year, which is either very funny or very damning, depending on how honest you're willing to be about your last ten posts.

It's not that AI writes badly. Most of the time it doesn't. It's that it finishes before you do, and if you let it, you won't notice the gap between those two things until somebody hands you a single word for it, and you realize the word was already about you.

I'd rather stay a step behind the tool and mean what I write than stay ahead of it and hand you something that only sounds like me, wearing my voice to a party I wasn't invited to and hoping nobody checks ID at the door.

© Copyright David Wakeman.  All Rights Reserved