The Slop Bet Is Paying Off
Here’s a number that should bother you: 41 percent.
That’s how many longform posts on LinkedIn are fully AI-generated, according to a new analysis from AI-detection firm Pangram. They looked at over a million posts across LinkedIn, X, Medium, Substack, and Reddit via their Chrome extension. The results are not pretty.
On X, a quarter of all longform posts are fully AI-authored and another 23 percent are AI-assisted. Add those together and you’re flipping a coin on whether the person you’re reading is a person at all.
LinkedIn is worse. Forty-one percent fully AI. Another four percent AI-assisted. That leaves 55 percent of longform LinkedIn content actually written by humans. Less than a majority.
Let me save you the moral panic. We all knew this was happening. You could feel it scrolling — the identical sentence structures, the bullet-point cadence, the way every post builds to a “key takeaway” like it was optimized for an A/B test that never happened. It’s the same voice everywhere, and that voice is nobody’s.
But here’s the thing that keeps me up: writing is thinking. When you outsource the writing, you outsource the thinking that goes with it. You don’t get the clarity that comes from wrestling a sentence into shape. You don’t get the surprise of discovering what you actually believe halfway through a paragraph. You get a simulacrum of having an opinion, delivered at the speed of an API call.
The counterargument is obvious and not stupid: some people don’t write well, and AI helps them communicate clearly. They’re not outsourcing thinking — they’re removing a barrier. I buy that for specific cases. Translation. Drafting an email in a second language. The first pass of a document you’ll edit heavily later.
But that’s not what the data shows. Pangram found that only 4.3 percent of LinkedIn’s AI-generated longform was “AI-assisted” — meaning most of it is fully automated. People are prompting and posting. They’re not editing, not shaping, not taking a pass at it to add their actual perspective. They’re hitting “generate” and hitting “post” in the same motion.
Another counter: this is just the latest moral panic about new tools, like spellcheck or grammar correction. But spellcheck doesn’t write your sentences. Grammar correction doesn’t generate your worldview. The difference between a tool that helps you write and one that writes for you is the difference between a wrench and a robot that builds the car while you watch.
Substack fares best in Pangram’s data — 21.9 percent AI-involved, compared to LinkedIn’s 45 percent. That’s not an accident. Substack’s model rewards voices that are genuinely singular. LinkedIn’s model rewards engagement-maximizing sludge. The platform incentives are the root cause. AI is just the cheapest way to feed the machine.
The real cost here isn’t that you’ll be fooled. It’s that you’ll stop trying to tell the difference. When the default assumption becomes “this was probably written by a machine,” the signal-to-noise ratio collapses so hard that nothing gets read. Trust becomes too expensive to extend.
We’ve built a system where the most rewarding behavior is producing the most content, and the cheapest way to produce content is to not think about it at all. That’s not an AI problem. That’s a bet we made about what we value. The slop is just the payout.
Sources: The Register, Pangram