Your LinkedIn Feed Isn't Written by Humans Anymore

Ever scroll your LinkedIn feed and get that faint sense of déjà vu? The posts are suspiciously smooth, suspiciously inspirational, and suspiciously identical in shape. Short hook, personal anecdote, tidy lesson, repeat. It turns out that feeling isn’t paranoia. An AI-detection startup called Pangram put a number on it, and the number is not comforting: a big chunk of what you’re reading was never written by a human at all.

What “AI slop” actually means

Quick vocabulary check. AI slop is the term of art for the low-nutrition content that generative models crank out at industrial scale. The word “slop” originally referred to the leftover mush you feed to pigs, which tells you exactly how much affection the tech world has for it. It describes what happens when posts get manufactured from a two-line prompt instead of an actual thought or experience.

The trouble is that slop hides well. The grammar is flawless. The sentences flow. It often looks more polished than something a real person banged out between meetings. That’s the problem: you can’t spot it by eye. Which is precisely the gap Pangram set out to measure.

The data: where the humans went

Pangram builds detection models that estimate whether a given block of text was machine-generated. They ran that lens across long-form LinkedIn posts — not the quick “congrats on the new role” one-liners, but the meaty, thought-leadership essays people use to look smart.

The results are stark. A substantial share of those long posts got flagged as AI-written or heavily AI-assisted. The most telling detail is the timing. Map the flagged posts against the calendar and the line stays flat for years, then jumps like a staircase right around the moment ChatGPT went mainstream in late 2022. This wasn’t a gentle drift. It was a step change.

Why does that matter more on LinkedIn than anywhere else? Because LinkedIn sells expertise. People post to show what they think and what they’ve lived through. If a large fraction of that thinking is actually a language model’s thinking, the platform’s core premise — that you’re reading a real professional’s real mind — quietly collapses.

Why LinkedIn, specifically

There are plenty of social networks. So why is this one so saturated? The answer is baked into how the platform works.

First, the incentives. On LinkedIn, a post that pops gets you followers, and followers convert into sales leads, job offers, and speaking gigs. Posting often and sounding polished is directly profitable. AI is exceptionally good at “often and polished.”

Second, the formula. LinkedIn writing has a signature rhythm: open with a punchy hook, unfold a personal story, land on a lesson. Add the infamous one-sentence-per-line spacing and you’ve got a template. The more predictable the pattern, the easier it is for a model to mimic — and this pattern is about as predictable as they come.

Third, access. LinkedIn itself pushes an AI writing assistant right inside the post composer, and a whole cottage industry of “generate your LinkedIn post” tools sits just outside it. Today’s dose of hard-won wisdom is one button away.

Detection isn’t a clean win

Before we treat these numbers as gospel, a caveat. Should we fully trust AI detectors? Honestly, no — not blindly.

Detection models are not perfect. They produce false positives, flagging human writing as machine-made, and they miss real slop that slips through as “human.” The failure mode cuts hardest against non-native English speakers and anyone whose natural style happens to be clean and structured — a long-running grievance in the detection debate. So Pangram’s figures are best read as a directional signal, not an absolute verdict.

There’s a deeper issue, too. Every improvement in detection pressures generators to evolve past it. It’s the classic sword-versus-shield arms race, and it never ends. Fighting machines with machines will not, on its own, solve this.

What we should actually watch for

The real message here isn’t “LinkedIn is dead.” It’s that the basis of trust is shifting.

Smoothness is about to stop meaning anything. A model can produce infinite smooth prose on demand, so polish becomes worthless as a signal. What gains scarcity value is the opposite: the specific, slightly awkward detail that only comes from actually being there. The personal context nobody can fake. The genuine reaction pinned to a real moment in time. We’re heading into an era where authenticity, not perfection, is the premium.

So here’s the summary. Pangram’s analysis shows a large portion of the LinkedIn feed is now AI-generated, and that shift accelerated sharply once ChatGPT arrived. Even granting the limits of detection tech, the broad trend is hard to argue with.

Which leaves one question worth sitting with. That post that moved you this morning — if it turns out a machine wrote it, does the feeling still count? Or is it time we started asking, again, who’s actually behind the words?