LinkedIn's New AI Doesn't Hate Robots. It Hates Boring.

Confession before we start: An AI helped write this post. About LinkedIn's AI that hunts posts written by AI. I'll be over here, in the front row, waiting for my reach to die.

Here's what happened while you were blaming yourself for cratering engagement numbers.

LinkedIn quietly retired its old algorithm and replaced it with something called 360Brew — a 150-billion-parameter model trained on every post, comment, and interaction on the platform. The old system counted your likes and shares like a popularity contest. This one actually reads your words. And buried in the technical documentation is something called an "authenticity score," trained to identify and suppress robotic, templated, AI-generated writing.

The result? Median reach dropped 47% since the rollout. Ninety-eight percent of users lost ground. The top two percent gained significantly. So if your numbers fell off a cliff this year, relax. You're not bad at this. You're just normal, which is now apparently a punishable offense.

The Arms Race Nobody Wanted

Now, the part I find genuinely funny.

I draft on LinkedIn constantly, and I use AI to help. Claude, specifically, helps me organize thinking and tighten structure. Then I run every draft through a second AI tool that scores whether it still sounds like a human wrote it.

So I'm using a robot to make sure I don't sound like a robot, so a third robot doesn't catch me sounding like a robot.

We built a hall of mirrors and called it content strategy.

This is not where I thought I'd be spending professional energy in 2025. But here we are, in what I'm calling the "detection theater" era — where massive computational resources are deployed on both sides of an arms race that shouldn't exist. LinkedIn trains billion-parameter models. Writers employ counter-detection tools. Everyone's exhausted, and for what? (But what do I know — I've only watched platform algorithm shifts kill entire strategies four times in my career.)

What 360Brew Actually Detects

Here's the thing 360Brew is quietly admitting, even if LinkedIn won't say it plainly: It can't actually detect AI. Nobody can, reliably.

I've tested this extensively with clients trying to maintain their voice while using AI tools. The false positive rate on "AI detection" is absurdly high. Human-written posts get flagged. AI-written posts sail through. The underlying technology just isn't there yet, and probably never will be — because the Turing test isn't about detecting machines. It's about whether machines can convincingly imitate humans.

What 360Brew can detect is generic.

Templated openers. The "I'm thrilled to announce." The "3 lessons I learned about leadership." The five tidy bullets that could belong to anyone. The thought leadership cosplay that clutters every professional feed.

The algorithm isn't punishing AI. It's punishing "nobody in particular wrote this."

This is a critical distinction. LinkedIn's AI isn't magic. It's pattern matching on a massive scale. And the pattern it's matching isn't "written by GPT-4" versus "written by human." It's "generic corporate-speak" versus "specific person with specific point of view."

The Napster Moment For Content

I've watched this movie before.

In the early 2000s, the music industry spent billions trying to detect and shut down file sharing. They built elaborate DRM systems. They sued grandmothers. They tried to make the technology itself illegal. None of it worked, because they were fighting the wrong battle.

The actual solution wasn't better detection. It was Spotify — a business model that made the legal path easier than the illegal one. The industry stopped asking "how do we detect and punish" and started asking "how do we build something people actually want to use."

LinkedIn is currently in the "sue the grandmothers" phase. They're deploying massive computational resources to detect something that's functionally undetectable, while the actual problem — oceans of generic, interchangeable content — goes unaddressed.

The real fix isn't better AI detection. It's rebuilding incentives so the platform rewards specificity over volume.

What The Algorithm Actually Wants

I've been advising clients through this shift for six months now, and here's what I'm seeing in the data:

Posts with specific numbers perform 3x better than those with generic claims. "Median reach dropped 47%" beats "engagement is down" every time.

Posts anchored in named incidents outperform abstract principles. "When 360Brew launched in November" beats "platforms are changing" by a mile.

Posts that take an actual position — that make someone disagree — crush posts designed to make everyone nod along.

The pattern is consistent: 360Brew rewards receipts. Specific companies. Actual conversations. The thing that happened Tuesday, not the timeless principle that could've been written any year.

Because here's what the AI scoring system has figured out: Generic writing is cheap to produce and cheap to ignore. If anyone could've written it, it doesn't matter that you did.

The Uncomfortable Question

So here's where we sit with this.

The moat was never "did a person physically type these words." It's whether a specific person, with specific experience and a specific opinion, is actually present in the writing.

AI can fake the format. It can mimic the structure. It can produce grammatically perfect sentences that sound vaguely professional.

It cannot fake your receipts. It cannot replicate the pattern recognition from surviving three different platform algorithm changes. It cannot inject the specific client conversation from Tuesday afternoon that crystallized your thinking.

Which raises the question nobody seems comfortable asking: What percentage of professional content had a real person in it to begin with?

Before AI, how much LinkedIn writing was just rearranged corporate talking points? How many "thought leadership" posts were already templated and interchangeable? How many of us were writing like robots before the robots arrived?

Maybe 360Brew isn't creating a new problem. Maybe it's just making visible a problem that's been there all along — that most professional content is performance, not substance. That we've been optimizing for looking like we have something to say rather than actually having something to say.

I don't have a clean answer here. I'm sitting with this tension myself.

What Actually Survives

I've tested dozens of approaches across client accounts. Here's what still works:

Lead with the specific thing. Not "authenticity matters on LinkedIn." Instead: "LinkedIn's 360Brew algorithm dropped median reach 47% in four months."

Put yourself in the frame. "I was reviewing client analytics when I noticed" beats "organizations should consider" every single time. First person. Present tense. You, the practitioner, encountering the thing.

Name the precedent. Don't predict what's coming. Show what happened last time. Napster and Spotify. Kodak and digital. The NYSE trading floor and electronic trading. Credibility comes from survived cycles, not forecasts.

Take an actual position. Hedging kills reach. "It remains to be seen" is content death. Make someone disagree with you. That's how you know you said something.

Use AI, just leave yourself in it. The draft can come from anywhere. The specific numbers, the named incident, the uncomfortable question — those have to come from you.

What To Do Monday Morning

Here's the specific action: Go look at your last five posts. Not to judge them. To audit for specificity.

Count the named companies, specific numbers, actual incidents. If there aren't any, you're writing in the generic zone where 360Brew is trained to suppress.

Then ask: Could someone else in my industry have written this exact post? If yes, that's the problem. Not AI. Interchangeability.

The algorithm is counting on you to forget that you have specific expertise. Specific scars. Specific pattern recognition from doing this work, not just commenting on it.

Go ahead, use the tools. Just make sure there's a you left in the draft.

Because the machine grading your humanity is betting you won't.

What's the most generic advice you've seen this week? Reply and let me know — I'm collecting examples of content that could've been written by anyone (or no one).