LinkedIn's Algorithm Just Changed Everything—And It Wants the Real You

I've always loved LinkedIn as a content platform. Not ironically. I know that's an unfashionable opinion—the site gets mocked for performative thought leadership and quasi-work-related trash. But I find the platform hugely valuable for discovering new thoughtful people and for sharing my own insights about work-related stuff. I've built genuine professional relationships through content here that never would have happened otherwise, including more than a few Linkedin connections that have turned into real friends and professional collaborations.

So I've been watching closely as reach dropped for almost everyone this year. My posts. My colleagues' posts. The "is LinkedIn broken?" complaints multiplying across my feed.

Turns out LinkedIn replaced its entire recommendation system with a 150-billion parameter AI called 360Brew. I've been digging into the research—LinkedIn's published paper on arXiv, practitioner analysis from AuthoredUp tracking millions of posts—and I wanted to share what I'm learning. The short version: I'm actually glad they made this change, because it is killing off some of the things I hated most about linkedin: the first comment links, the "comment BLORB to get my 1 million killer prompts," and shorter form content that doesn't offer new insights or learning. Even though it's requiring some adaptation, the platform is becoming more like what I've always wanted it to be.

What Actually Changed

The old algorithm tracked clicks and engagement. The new one reads text.

That's the simplest framing. 360Brew is a large language model, similar to Claude or Gemini, trained specifically on LinkedIn data over nine months by their Foundation AI Technologies team. It reads your profile, your posts, your comments the way a thoughtful colleague would. When you write about "organizational transformation in government agencies," it understands the semantic relationship to "federal modernization" and "public sector change management," even if your audience has never used those exact terms.

The cold-start problem that plagued niche experts—where the algorithm couldn't find your audience because you hadn't accumulated enough behavioral signals yet—largely disappears. If you write about something specialized, the algorithm can now find your people based on meaning, not just click-pattern overlap.

But here's the catch: it has to understand what you're actually an expert in first. And it figures that out by reading your profile.

Your Profile Is Your Classification System

Before 360Brew distributes your content, it reads your headline, About section, and skills to categorize your expertise. Then it checks whether your posts actually align with that positioning.

This isn't punitive—it's just how semantic matching works. Post about travel photography one day, crypto the next, your actual work occasionally? The system genuinely can't classify you. So your content doesn't get matched to relevant audiences, because the algorithm doesn't know who would find it valuable.

For your headline, this means clear niche beats clever wordplay. Not "Marketing Maven | Thought Leader | Coffee Enthusiast" but something that actually names what you know and who you help. Your About section should identify your 2-3 core topics in the first paragraph. Make the connection obvious between your stated expertise and what you actually post about.

When I tightened my own positioning around government AI transformation—dropping the more general "digital innovation" language—engagement from federal and defense audiences actually increased. The algorithm could finally find my people. That was counterintuitive at first. Narrowing felt like limiting. But semantic specificity is how the system works now.

The data suggests it takes about 90 days of consistent, topic-aligned posting for 360Brew to fully categorize you and start optimizing distribution. That's a longer horizon than the old system required, but the matching is more accurate once it clicks.

What Signals Actually Matter Now

Analysis of over 3 million posts by AuthoredUp found something striking: one save drives roughly 5x more reach than a like, and about 2x more than a comment. That's a significant shift in the value hierarchy.

It makes sense when you think about what a save actually signals. A like is quick validation—you saw something, you approved. A save means you found the content valuable enough to come back to later. The algorithm reads that as lasting utility, which earns broader distribution.

The implication is that your content needs to be reference-worthy. Before publishing, the question to ask is whether someone would want to return to this next week. Step-by-step frameworks with real examples. Data with non-obvious implications. Annotated templates. Contrarian perspectives backed by evidence. These are the formats that earn saves.

The other signals that matter now are more qualitative than before. Paragraph-length comments that add genuine perspective carry more weight than emoji reactions or one-word responses. When people start replying to each other in your comments—not just to you—that's particularly valuable, because it means you created an actual conversation rather than just a broadcast. Delayed engagement matters too: posts that continue earning saves and substantive comments 24-72 hours after publishing perform 4-6x better in suggested feeds. The algorithm reads that pattern as durable quality.

Research on language models also shows they weight the beginning of text more heavily—the first 1-2 sentences of your post get 3-5x more processing attention than content buried deeper. Which means leading with specifics (names, numbers, organizations, concrete claims) outperforms throat-clearing openings. "Gong's conversation intelligence helped us identify why 67% of deals stalled in legal review" lands differently than "I've been thinking about sales tools lately."

What Stopped Working

Some tactics that worked on the old ID-based system actively backfire now.

Engagement pods get spotted. Because 360Brew is a language model, it can measure lexical diversity—how similar the comments on your posts sound. Ten comments using the same phrases from the same small circle get marked as low-value signals rather than genuine engagement.

Links in the first comment, which people assumed avoided algorithmic penalty, actually get deboosted now. If you need an external link, put it in the main post and accept modestly lower reach, or minimize external links entirely.

Hashtag stuffing is irrelevant. The algorithm reads your actual text and follows meaning, not hashtags. Think of hashtags as navigation aids for human readers, not signals for ranking.

And volume over quality actively hurts you. Frequency isn't a primary signal anymore. One excellent post per week that earns saves and generates real discussion outperforms seven mediocre posts that get quick likes and nothing else. Topic consistency over 90 days matters more than posting schedule.

The Authenticity Thing

Here's what gets me excited about this change.

Multiple sources analyzing 360Brew report the same finding: the model can distinguish authentic voice from AI-generated patterns. The phrasing I keep seeing is that "text patterns written by AI vs. a credible, authentic voice will be discernible." Content that sounds like genuine human expertise performs differently than content-mill output.

This is LinkedIn telling us, through its architecture, that the path to visibility isn't sounding more polished or more corporate or more like everyone else's ChatGPT output. It's sounding more like yourself.

I've always loved authentic voices on this platform. The consultants and executives who build real influence share a common trait: they have distinct perspectives, informed by actual experience, and they're willing to share them—including the uncertainties, the lessons learned the hard way, the things they're still figuring out. That kind of professional vulnerability, grounded in real expertise, is what builds trust. It's what makes content memorable.

That's always been true for building relationships with humans. Now it's apparently true for earning algorithmic distribution too.

Why I'm Sharing This

I want more people I know to write long-form content and share it on LinkedIn.

The platform rewards it now. Depth over frequency. Substance over tactical plays to mess with the algorithm. Authentic voice over manufactured engagement. If you have genuine expertise and real perspectives—and you do—this algorithm is designed to help your content find the right audience. Clear thinking on a specific topic, expressed in your own voice, can now compete with polished content from people with massive followings. The algorithm matches based on relevance and quality, not just existing reach.

Here's what I'm doing with this information:

I've tightened my profile around government AI transformation specifically. Headline, About section, skills—all aligned so the algorithm knows exactly what I'm about.

I'm committing to 90 days of posting primarily within my core topics: AI-era organizational transformation, public sector modernization, the intersection of technology and human systems. Not because I don't have thoughts about other things, but because I want the system to clearly categorize my expertise.

I'm optimizing for saves rather than likes—WHICH IS EXACTLY WHAT SHOULD BE HAPPENING! I'm focusing more on longer content that will be of enduring value, that people want to reference again later, that makes people worry if they don't have time to read it thoroughly. Before each post, I'm asking whether this is something someone would want to reference later.

And I'm leaning further into my actual voice. Less corporate polish. More genuine perspective. If LinkedIn's AI can detect the difference, that's one more reason to sound like a human with actual opinions rather than a carefully hedged professional persona.

If you've been thinking about writing more on LinkedIn, this is a good moment to start. The algorithm is finally optimized for substance. Bring yours.

I'd genuinely like to hear what you're seeing on your end. Has your reach shifted? Are different kinds of content performing better? And if you've been thinking about writing more long-form—what's been holding you back?