if ai writing is slop, explain to me how it keeps winning literary prizes

So, you think AI writing isn't good enough yet. Fair. You've seen the slop.

But explain this one thing: a story that may have been AI-written just won a literary prize, judged by people who read for a living.

Let that sink in before we go any further.

Btw, you're not wrong. A LOT of AI writing is genuinely terrible. The tell-tale em dashes, the lists-of-three, the way every paragraph sprints toward a moral... you've scrolled past it on CT & LinkedIn ++ you've gotten it in emails, and ofc your instinct to dismiss it is not wrong, and I'm not here to argue you out of that instinct.

The slop stuff is everywhere, and you've trained yourself to spot it. That's a reasonable response to a real pattern.

Keep that skill.

What I'd like to show you today is one thing that complicates things a bit.

Here's a passage. Read it as if it just came across your feed.

"my mother did not come with me to the port. she said goodbye from the kitchen doorway, her hands still dusted with flour, a dish towel folded over her shoulder. i knew, even then, that she was already somewhere else."

So, tell me, did you flag it? Probably not, aye?!

It reads like someone who's done this before. Sure, a little restrained, but emotionally precise without overcooking it.

I generated that in ~15s with a few Claude 'prompting' tricks.

You prolly didn't flag it this time, and neither did the prize judges here:

Well, this is a first: a ChatGPT-generated story won a prestigious literary prize (The Commonwealth Prize). "Not X, not Y, but Z" sentences everywhere, the "hums" trope, and plenty of other obvious markers of AI writing. A major milestone for AI, at any rate...

Nabeel S. Qureshi · @nabeelqu · May 18

'The Serpent in the Grove' by Jamir Nazir is a story set in rural Trinidad about a struggling farmer, a silenced young wife and a grove that seems to remember what others try to bury. Awarded the Caribbean regional winner title for its lyrical precision and haunting atmosphere.

Commonwealth Foundation Creatives · @cwfcreatives · May 16

Granta published one of the flagged regional winners, then ended their partnership with the prize when the controversy broke.

Their statement acknowledged they may have handed a prize to AI-generated work, and admitted they may never know for certain.

The Commonwealth Foundation reviewed the situation and said it was satisfied that AI was not used. The institution was satisfied, meanwhile the detectors & writers were saying otherwise.

On a personal note, I don't think anybody is lying, necessarily. We've just arrived at a moment where the question "was this written by AI?" has no clean answer, because the text itself doesn't carry the proof anymore.

We've reached the stage where a reading public, a prize committee, and a major literary journal all processed a piece of writing and could not confidently tell what made it.

That's what kills "slop" as a reliable category, imho.

And you might be thinking: okay, Gabe, but I could have spotted it. I know the tells.

And sure thing, but the tells you're thinking of aren't AI tells. They're style habits that humans created, and AI were trained on them too, so spotting them doesn't do what you think it does.

Don't believe me? Sure, here's the data: people can correctly identify AI text about 50% of the time.

Which is essentially a coin flip.

Okay, Gabe, the "i can spot it" move is gone. What's left then?

If the writing is good enough to get past trained readers, past judges who read fiction for decades, past editorial teams at Granta, then "slop" was never the argument.

The argument was always something else: where did it come from?

Quality is settled. That's what the prize situation proves.

The real discomfort is about what produced them, and whether that matters, and if so why.

Now here's where all of this gets interesting, as two things are happening in parallel right now.

First, the disclosure side. An Ipsos study of over 23,000 people across 30 countries found that 79% want companies to disclose when AI is used in content.

That's a near-consensus.

But enforcement is basically zero.

So we have a near-universal preference that is, at the moment, being ignored.

Second, the fingerprint side. A University of Maryland and Google DeepMind study called StoryScope stopped trying to read AI at the word level.

They looked at structure instead: the shape of how a story develops across tens of thousands of texts.

StoryScope identified AI versus human authorship around 93% of the time, and it held up even after the surface style was edited to mask the obvious tells.

Each model has what you might call an accent.

Not in the vocabulary, but in the skeleton.

Claude tends to flatten the way tension escalates, treating the emotional arc as something to resolve cleanly.

GPT reaches for dream sequences and interior monologue.

Gemini leans on description that moves from the outside in, cataloguing the external world before it enters the character.

These patterns survive rewrites as you can change every word, but you can't easily change the handwriting.

For example, a voice coach can teach you a new accent until you sound different at dinner, but your gait doesn't change and your posture doesn't change.

Structure is gait. Vocabulary is accent.

So in short, here's my take: AI is not slop. Never was. Stop spending energy proving it is.

The question now is provenance.

Where did this come from, which model, which prompt, how much human editing, at what stage.

The structure already answers that, whether you like it or not, whether you disclose it or not.

So the only move that makes sense to me is to be the one who said it first.

So, here's my tip for the writers who use AI in their process, get ahead of the question before someone else runs the detector. Own the AI narrative as part of your workflow.

For the agents and teams shipping AI-assisted content, own the fingerprint before someone reads it as an accusation.

The moment is now, because the window where you could just... not say anything is closing faster than most people realize.