Six Platforms Came for AI Slop. None of Them Banned AI.

A pineapple with abs is cheating on a strawberry, and thirty-nine million people watched it happen.

That is Fruit Love Island, which runs daily on TikTok and has 3.3 million followers. The contestants are humanoid produce. Cherries, bananas, a strawberry, a pineapple, every one of them glossy and muscled and issued a swimsuit and an emotional wound. They betray each other in sixty-second episodes. One of those episodes crossed thirty-nine million views back in March.

I do not watch it. The 3.3 million people who follow it are a real fact about the world and not one I get to wave off. But you do not have to be one of them to end up knowing who betrayed who this week. The clips find you anyway. Mine came wedged between things I actually asked for, the way you see a billboard on a road you had to drive. That is what the volume does. Participation is optional and exposure is not.

An Indian company called Dashverse licenses this kind of thing to TikTok and ReelShort. They currently ship about a hundred fully AI-generated microseries a month. By December they intend to ship a thousand. Their exec calls 2026 the year of the vertical-first AI drama. He is probably right.

Then in the last two weeks, six platforms moved at once.

Two weeks of policy, stacked

July 20. YouTube clarified what it now calls its inauthentic content policy, which used to be called the repetitious content policy, a rename that tells you where their heads are. The target is content “made with a template with little to no variation across videos,” anything “easily replicable at scale.” Three strikes and you are out of the Partner Program permanently. Back in January they had already terminated roughly sixteen channels holding about thirty-five million subscribers between them.

July 21. Substack partnered with Pangram, an AI text detector, so readers can scan a post and see what comes back.

July 30. LinkedIn shipped a button. It sits in the three-dot menu on any post and it says Seems like AI slop. Press it and that post’s reach gets cut, and your press goes into training their detection models. Their chief product officer said something honest about it:

Slop is hard to define and the definition changes; this lets us tune our models and make better feeds.

The scale they are fighting is not small. LinkedIn says it blocks hundreds of thousands of automated comment attempts every day, and has stopped billions of automation attempts in recent months. A Pangram study found that forty percent of long-form posts on the platform read as fully machine-generated. Forty percent. That is not a contamination problem, that is the water.

Spotify has pulled more than seventy-five million spam tracks in twelve months. Seventy-five million. Songs uploaded in bulk to farm royalties, engineered to just clear the minimum length that triggers a payout, titled to hijack somebody else’s search traffic.

TikTok labels AI content, buries invisible watermarks in the metadata, and gives you a toggle to turn the stuff down in your own feed. Pinterest labels it and lets you limit it. Meta extended its AI-info labels to advertisements in June.

Six platforms. Fifteen days. Nobody coordinated this and all of them landed in the same place.

What they actually banned

Read the policies next to each other and the thing jumps out.

None of them banned AI.

YouTube punishes templates and says plainly that it wants to reward creators using AI for original storytelling. LinkedIn’s own executive went out of his way to say that using AI to refine your thinking is fine, while the enforcement lands on automation running at volume. Spotify came after duplicate uploads and payout-length filler and impersonation, not synthesis. Substack did not ban generated work at all, it built a disclosure lane and let writers opt in.

Every one of them drew the line in the same place, and the line is not the tool. It is how much of it there is, and whether any part of a person went in.

They banned volume.

Which is a strange thing for a platform to do, if you sit with it. Volume is what feeds them. Volume is the whole shape of the business, more posts feeding more watch time feeding more surface to sell against. Six companies just decided that past some threshold, more content makes the product worse, and moved against their own business model to stop it.

The detail that gave the whole thing away

Same week as the slop button, LinkedIn deleted a feature of its own.

The “enhance your post” tool is gone. That was the one that took what you typed and rewrote it into something smoother, the button that turned a sentence into a LinkedIn sentence. They pulled it.

What they put in its place is a proofreader. It fixes your writing and it is built specifically not to change your voice.

Sit with the shape of that decision. A company that sells software ran the experiment on itself, at scale, with its own product, and concluded that the machine writing for you produces something worse than the machine checking what you already wrote. So they killed the first one and kept the second.

They killed the button that writes and kept the button that checks. That is the entire argument, shipped as a product change, by the platform with the most data about it on earth.

Why the slop was going to eat itself anyway

The enforcement is the less interesting half of this. It is six companies catching up to arithmetic that had already been running for a year without them.

Fruit Love Island is a commodity. I do not mean that as an insult, I mean it as an accounting fact. Any one of us could clone the format this afternoon with a subscription and a free evening. Different produce. Same beats. It would not be as good, but it would be close enough, and close enough is the entire proposition.

When a thing is that cheap to copy, the margin on it goes to zero. Not slowly. Everyone piles into the format at once, the feed saturates, per-video returns collapse, and the only lever left is to make more. Post four a day instead of one. Then ten. Then license a thousand a month from a company whose whole business is a thousand a month.

And volume is exactly the thing that just got criminalized.

So the survival strategy and the ban are the same behavior. The escape route leads straight through the wall. Every operator scaling their way out of a saturated market is scaling toward a three-strike system and a permanent Partner Program removal, and most of them do not appear to have noticed yet.

Sixteen channels. Thirty-five million subscribers. Gone in January, before any of this was news.

I am not neutral here

Last Wednesday I published an article about the machine that runs this website. It stages my drafts, schedules them, fires the newsletter, pings the search engines. I built it so I could publish three times a week without being at the desk three times a week.

So I am not standing outside this pointing at anybody. I automate. I use AI on nearly everything I make, including the research under this piece, where it read six platform policies in the time it would have taken me to find the first one. I have made music with it. I have used it to build worlds I could see clearly in my head and could not draw with my hands.

Which is why the lesson here will not be “use AI thoughtfully.” That advice costs nothing to say, and everybody nods at it on the way to doing exactly what they were already going to do.

Effort does not draw the line. Some slop takes real effort, and some of it is made by people working hard. Disclosure does not draw it either, though disclosure is decent hygiene.

What separates the two is what you point the thing at.

The generic middle of the internet

In June I wrote a piece about second brains, which is a phrase people in tech throw around like everyone already agreed what it means. The argument was that the whole thing comes down to a folder. Plain files, named so you can find them again, holding what you decided was worth keeping. No app required, no method to buy.

And that an AI reading that folder is taking an open-book exam. Nothing is baked in, nothing is trained, it just opens your material the way you would open a book you had annotated. Change a file and the next answer changes with it.

A line from that piece is the one I keep coming back to now: what I want out of these tools is something built out of my own knowledge instead of the generic middle of the internet.

That is the whole distinction, and the platforms just wrote it into policy without using the words.

Slop is what comes out when you aim a model at the generic middle. It has to be, structurally. You gave it nothing of yours, so it returns the average of everyone, and the average of everyone is what a hundred thousand other people are also being handed this afternoon. The output is interchangeable because the input was.

Point the same model at your own folder and the arithmetic inverts.

What is in mine is not a manual. It is what I have read and what I actually thought about it. Ideas I keep circling and cannot put down. Drafts nobody has seen. Songs I wrote as somebody else, under another name, for reasons I am still working out. Whole worlds I built one object at a time because I wanted to see if I could. Arguments with books. A way of holding technology and spirit in the same hand that took me years to arrive at and that I have still not finished saying.

None of that is on the open internet. It cannot be. It only ever happened to me.

An AI reading that is not producing average work, because it is not working from average material. It is working from a body of taste. Years of deciding what was worth keeping and what was not, which is the only real definition of expertise I have ever found. Every idea I chased and dropped and came back for three months later. Every thing I refused to say because it was not true.

The model has none of that. It walks in blank every single time and borrows mine. Same tool as everyone else, same twenty dollars a month. The input is the entire difference, and the input is the one thing nobody can clone with a free evening.

Which is also, quietly, why the second brain is the durable asset and the tool is not. Models get deprecated. Mine did, in June, on a Friday, and I wrote about how much that stung. The folder survived the deprecation without noticing it happened.

What this costs

It is slower. Obviously it is slower.

Building the folder is unpaid work that pays back late, in a currency nobody can see. What you get for it is work that sounds like you. That is the entire return. It does not trend. It will not appear in any dashboard. Nobody is ever going to compliment you on the thing you did not post because it was not yours.

Meanwhile a pineapple with abs is doing thirty-nine million views and somebody is getting paid this week.

I do not have a clean answer to that. Some of this is a real choice between money now and something that compounds, and telling a person who needs money now to think about compounding is a thing you can only say comfortably from a certain distance. I have been on both sides of that distance.

What I can tell you is where the platforms just put their money, and they put it against the fast version. Not on principle. They ran the numbers on what a feed full of interchangeable work does to the people who have to sit in it, and they moved.

Nobody in this story had a problem with the tool. They had a problem with where it got aimed.

My folder is open in the other window. Somewhere in it is a note I made months ago, three lines, an idea I have not figured out how to write yet and have not been willing to throw away. I do not know what it becomes. It is worth nothing to anybody but me and it will never do thirty-nine million views, and I am going to sit with it again tonight.

Forward → Upward ↑ Onward ↗︎
Mstimaj

Sources and Further Reading

• TechCrunch: LinkedIn adds a button to report AI-generated slop, on the July 30 rollout.
• Fortune: LinkedIn’s slop button and the billions of blocked automation attempts, source of the executive quote and the daily blocking figures.
• TechTimes: LinkedIn pulls its own AI writing feature, on replacing “enhance your post” with a voice-preserving proofreader.
• TechCrunch: YouTube clarifies its policies around AI slop, on the inauthentic content policy.
• From Substack to YouTube: the platforms cracking down on AI slop, a roundup of the TikTok, Pinterest, Meta and Substack measures.
• Spotify Newsroom: Strengthening AI protections for artists, including the spam filter and the seventy-five million removed tracks.
• NBC News: The AI fruit drama on TikTok, on Fruit Love Island’s audience.
• AI-generated microdramas are real and thriving, source of the Dashverse production figures.
• AI slop, including the Merriam-Webster definition.