On 30 July, LinkedIn added a button. Tap the three dots on any post and you can now report it as “Seems like AI slop.” The post vanishes from your feed, and the signal goes off to train their detection models. It arrived alongside a startling number: the AI-detection firm Pangram reckons more than 40% of long-form LinkedIn posts are now fully machine-written, and LinkedIn hosts roughly 62% of all the AI content scanned across the big social platforms. So: a slop button. About time.
Except the button gives the game away, and it is worth reading the exact words. It does not say “this is AI-generated.” It says seems like AI slop. That hedge is the whole argument.
Slop was never an AI problem
Notice what LinkedIn is not doing. It is not banning AI. In the same breath it quietly retired its own “Enhance your post” writing tool and replaced it with a lighter proofreader that leaves your voice intact. A platform declaring war on machine-made content does not do that. What it is actually policing is not the tool. It is the output.
Slop is not “written by a computer.” Slop is content that positions nothing, moves no one, and exists only to feed the feed. People have been producing it by hand for years: the “humbled and thrilled to announce,” the broetry with a line break after every word, the Agree? posts engineered for the comment box. None of that needed a language model. The model did not invent slop. It just made slop free, and removed the last natural brake on it, which was that writing rubbish used to take almost as long as writing something good.
That is why people reach for “AI” as the explanation. Generic plus high-volume pattern-matches to a robot. But the AI is a red herring. Plenty of hand-typed posts are slop, and plenty of AI-assisted ones are not. The line was never human versus machine. It is substance versus nothing.
“But it got engagement”
Here is the objection that matters, because it is the one that sounds clever. If a post pulls likes and comments, how can it be slop? Surely engagement is proof it landed.
No. Engagement is not an acquittal. It is the oldest vanity metric there is, and slop is frequently engineered to earn it. Rage-bait, fake vulnerability, the deliberately wrong hot take fishing for corrections: these are not accidents that happened to go off. They are slop optimised for exactly the number you are holding up as the defence. A post can win every like in the room and still move no one a single step closer to doing anything.
Which is the same mistake this whole business runs on everywhere else. It got impressions, so it worked. It got reach, so it was a good campaign. A metric that measures attention is not measuring value, and attention you convert into nothing is just expensive noise. Likes you convert into nothing are slop with good PR.
The actual test
If you want to know whether the thing you are about to post is slop, do not ask whether a machine could have written it. Ask what job it does. Content has one: to move a real person one stage closer to a decision, by telling them something true, specific and useful that they could not get from anyone else. A post that does that is not slop, however it was made. A post that does not is slop, however many hands touched it and however many likes it collects.
So the antidote to slop is not “stop using AI.” It is “have a point.” Use the tools if they help you say the thing faster. Just make sure there is a thing. Say what only you could say, show your working, write for a buyer rather than for an algorithm, and put your name to a view you would defend.
Detection is coming, and it misses the point too
The timing this fortnight was almost too neat. The same week LinkedIn shipped its slop button, Anthropic began adding an invisible watermark to the text its Claude models generate, one built to survive a copy and paste and even some editing. Brussels asked for it: the EU’s new transparency rules want AI-made content to be detectable, and every serious model will follow.
Worth having. But read what the watermark actually certifies. It proves a machine was involved in producing the words. It says nothing about whether the words are any good, whether they are true, or whether a person shaped them into something worth reading. It marks processing, not slop. A watermarked post can be excellent; an unwatermarked one can be worthless. The detector answers “was AI here,” which was never the question that mattered. The question is still “does this do a job,” and no watermark will answer that for you.
That gap, between what we can now detect and what we actually care about, is worth a piece of its own. Another day.
A rose by any other name would smell as sweet. Slop by any other name still stinks. And a fistful of likes has never once made it a rose.
