“Slop” is the word we finally settled on for AI output nobody wanted, and honestly, it’s a good one: ugly, agricultural, the sound of something hitting a trough. But watch where people point when they say it. They point at the model. That’s the part I want to argue with, because models don’t make slop. Slop is output nobody claimed. A model generated it, sure. Then a person chose to ship it, without looking at it, without an opinion about it, without putting their name anywhere near it. The generating was automated. The not-caring was human.
Simon Willison, who did as much as anyone to pin the word down, defined slop in 2024 as AI content “mindlessly generated and thrust upon someone who didn’t ask for it.” The load-bearing word there is “mindlessly.” The machine can’t supply the mind. That part was always the job, and it doesn’t ship with the subscription.
Same model, same brief
If you think slop is a capability problem, here’s a test. Take one model, one brief, and run it twice. Run one: accept whatever comes back, paste it in, hit publish. Run two: same model, same brief, but somebody with an opinion sits with the output. They cut the filler sentence (there’s always a filler sentence), push the layout off the default, rewrite the headline until it says the one thing that’s provably true about the product. They notice the testimonial section is generic praise from nobody and delete it whole. They swap the sparkle icon, because a sparkle icon in 2026 is a confession. None of this takes long. Most of it takes taste and a willingness to be disagreed with. Same tool. Same words going in. Two very different objects coming out.
If the tool were the problem, both runs would be slop. They’re not, and the difference isn’t anywhere in the weights. It’s whether anybody was HOME.
The question that sorts everything
I use a one-line diagnostic for this, and you can run it on anything: a homepage, a deck, a bio, a logo. Ask the artifact, “why is it like this?”
Work that somebody made has answers, and usually fast ones. The headline leads with the number because the number is the story. The mark stays small because the brand is a whisper in a loud room. You don’t have to agree with the answers; you just have to notice they exist. Slop answers with a shrug. Why three feature cards? No reason. Why this gradient? It came with the template. Why this phrasing? That’s just… what came out. Nobody decided, so the median did.
Run it on your own work too, while nobody’s watching. It stings in a useful way. If you find a section you can’t defend, you’ve found the exact spot where you stopped being the designer and started being a conveyor belt. I find one more often than I’d like to admit, and the fix is always the same: make a decision there, any defensible decision, and the slop stops being slop.
- Why this headline?
- (no answer)
- Why three cards?
- (the template had three)
- Why this gradient?
- (it came with the neighborhood)
- Why this headline?
- It leads with the one provable claim.
- Why two cards?
- The third didn't earn its spot.
- Why no gradient?
- This brand is a whisper, not a laser show.
I wrote about the drift: what happens when work gets judged against the wrong target. Slop is the step past that. Slop is what happens when nobody’s judging at all.
Where all the sameness comes from
Which brings us to the complaint hiding inside the complaint. When people say the internet is filling with slop, half of what they mean is that everything looks and reads ‘the same.’ Same stock phrasing, same gradient, same three cards. The mechanism is almost embarrassingly simple: a model’s most likely answer is the middle of its distribution. Perfectly plausible, maximally unsurprising, and identical to the middle everyone else is being handed. A Science Advances study ran this at scale with short stories: writers given AI ideas wrote individually better stories that were collectively more alike. And nobody imposed that trade. It’s just what happens when everyone takes the first answer.
The middle isn’t a scandal, by the way. It’s a fine place to START. The first generation is a stake in the ground. The work is everything you do to it after, and every one of those moves carries a reason somebody can be asked about later. Choice leaves fingerprints. In a feed full of median output, the fingerprints are the product (building got cheap; deciding didn’t).
Generating was never the sin
One more distinction, because the word “slop” tends to smear across both halves… generating and publishing. Drafts are free now. You can generate garbage all afternoon and harm no one; I do it daily, it’s called exploring. The moment that matters is the publish: the instant a person takes output and attaches it to the world, to a brand, to (nominally) their own name. That’s where slop happens, and notice the model’s already out of the room by then. A byline used to mean somebody stood behind the thing. Slop is work with nobody behind it, even when there’s a name printed right there.
And I get why it happens. Publishing used to have a gate. Printing cost money, airtime cost money, even a bad website cost a developer’s weekend, so somebody senior looked at the thing before it went out, if only to protect the spend. The gate wasn’t quality control exactly, but quality hid inside it. Now the marginal cost of one more post, one more page, one more variant is zero, and the checking didn’t get cheap along with it. Checking still costs what it always cost: a person’s attention and a person’s neck. So the checking is what gets skipped. That’s the entire supply chain of slop, and there’s no model anywhere in the guilty part.
At my desk the rule is simple: nothing ships while “why is it like this” gets a shrug from me. The machine drafts more than I could ever draft alone, faster than I can read it… and none of it counts as mine, or as done, until I’ve chosen it on purpose. The eye still has to sign.
So the next time the feed serves you something gray and vaguely familiar, blame accurately. The model produced a plausible middle, which is its job. A person published it unclaimed, which used to be unthinkable and is now a business model. The capability didn’t fail. Somebody just walked away from the byline.
Slop is a choice. So is claiming the work.