AI Slop Is Taking Over LinkedIn. Here's What's Actually Going On.

Quick answer, before the detail: if a lot of what you've been scrolling past on LinkedIn lately has felt oddly hollow technically fine, but somehow about nothing you're not imagining it. New research put a number on it, and it's higher than most people would guess. LinkedIn has also just announced it's doing something about it, though probably not what you'd expect.

What "AI slop" actually means

The term isn't just internet slang for content someone dislikes. It has a fairly specific meaning: writing (or video, or images) that's technically polished but empty no real story, no specific detail, no opinion anyone would actually argue with. It reads like it was generated by giving a topic to a model and publishing whatever came back, because in an enormous number of cases, that's exactly what happened.

It's not a new complaint. What's new is that someone finally measured it at scale, and a major platform is now building tools to actively suppress it.

Two things happened close together this year that are worth putting side by side. First, an AI detection company called Pangram Labs published data on how much of what people actually see on social media is AI-written. Second, LinkedIn announced it's going to start actively suppressing the reach of exactly that kind of content. One tells you the size of the problem. The other tells you what LinkedIn thinks the problem actually is and the two definitions don't fully overlap.

The number that made the rounds

Pangram's numbers came from a Chrome extension that scans posts people scroll past in their normal, everyday browsing not a crawl of the open web, an actual sample of what a million-plus real feed views looked like over two months. For posts longer than 250 words, the finding was that a little over 40 percent of what showed up in LinkedIn feeds was fully AI-generated, with only about 55 percent written by a human with no AI involvement at all.

That made LinkedIn the most AI-saturated of every platform in the study worse than X, worse than Medium, far worse than Reddit or Substack. LinkedIn accounted for roughly a third of everything scanned, but nearly two-thirds of all the AI-flagged content came from it alone. It wasn't just present. It was disproportionate.

The detail nobody pulled out: LinkedIn doesn't blend

Here's the part that's more interesting than the topline stat, and it's sitting right there in the numbers if you line up LinkedIn against X.

On X, the AI-assisted category posts drafted or edited with AI but not fully generated by it sat at around 23 percent. On LinkedIn, that same middle category was barely above 4 percent. Put those two numbers next to the "fully AI" and "fully human" figures for each platform, and a pattern shows up: X users blend. They draft with AI, then edit, rewrite, add their own voice back in. LinkedIn users, by contrast, mostly don't. It's either entirely written by a person, or it's handed almost completely to the model.

That's not a minor stylistic difference. It suggests two very different relationships with the tool. Somewhere on LinkedIn, "using AI to help write" quietly became "using AI instead of writing," with very little middle ground surviving in between.

What LinkedIn is actually building

LinkedIn's own announcement, delivered by VP of Product Laura Lorenzetti, is narrower than the panic suggests. Flagged posts won't be deleted. They'll simply stop being recommended beyond the poster's first-degree network visible to people who already follow you, invisible to everyone else the algorithm might otherwise have shown it to. Lorenzetti also noted that content creation on the platform is up 14 percent year over year, which is presumably part of why this became urgent enough to act on now rather than later.

The system is described as "AI solving AI" classifiers trained to catch three specific things: generic posts and comments that add nothing original, comment bots (identified partly by posting speed and volume, not just word choice), and what the company calls attention-bait video, like footage of workplace accidents paired with generic safety platitudes. The rollout is deliberately gradual. LinkedIn has said plainly that this won't be the last version of the problem people will find new ways around whatever gets built, and the company expects to keep adjusting.

The part that's a little funny

LinkedIn has offered a built-in "Write with AI" button (now relabeled "Enhance post") for years. It made posting effortless, which is part of why posting volume climbed and, almost certainly, part of why so much of that volume ended up generic. The platform spent years lowering the cost of publishing, and is now spending real engineering effort raising the cost of publishing badly.

Pangram Labs' own writeup on the data adds one more detail worth knowing: the LinkedIn executive announcement about cracking down on AI content was, by Pangram's own detector, itself flagged as likely AI-generated. Nobody needs to editorialize that one.

What "slop" actually means here and what it doesn't

The useful line in LinkedIn's own explanation is this: it's fine to use AI to help write. What gets suppressed is content that lacks a real, original point of view recycled "thought leadership," engagement bait, posts that repeat an idea everyone's already seen instead of adding to it. One outside analyst quoted in coverage of the change put it well: a thoughtful post drafted with AI can be worth more than a human-written post that says nothing at all. The target was never the tool. It's genericness.

That's a meaningfully different bar than "did AI touch this." A specific story, a real number, an opinion you'd actually defend in the comments that survives, whether AI helped you word it or not. A polished paragraph that could sit on anyone's profile with the name swapped out that's the thing losing reach now, whether a human typed every word of it or not.

What this actually changes if you use LinkedIn

If the posts you write, or the ones you actually stop to read, tend to come from something real an actual conversation, a real number, an opinion someone would push back on none of this touches you. The algorithm was never built to catch specificity. It was built to catch its absence.

If a habit has quietly formed of handing a topic to a model and publishing whatever comes back with barely a second look, this is the moment that stops being free. Not because LinkedIn can read how a post got written, but because it can now read the result and a post with nothing real in it looks the same whether a person or a model produced it. That was always the actual tell. LinkedIn just finally built something to catch it.

Which raises the more useful question, whether or not the algorithm ever notices: if a post could be about anyone, was it worth posting at all?