If your LinkedIn reach fell off a cliff sometime this year, there's a reason, and it isn't you.
I'm a senior copywriter at storyarb , which means I write executive LinkedIn content for a living. I've spent enough time on LinkedIn's engineering blog and research reports this summer that my own mother has started describing it to other people as a condition. She is also my most committed hater (and fan) on that website, which I've decided to read as a compliment about my reach.
Anyways, here's what I came back with.
LinkedIn stopped counting your engagement and started weighing it. So what matters right now with the new algorithm is value:
A like is applause, a save is intent, and the weights tip hard toward intent.
Every signal above a like costs the reader something, and a save is the only one where they bet on themselves, deciding a future version of them, mid-problem, six weeks out, will want this back.
When LinkedIn flags a post as low-value, it doesn't delete it or tell you. It quarantines the post to your first-degree connections, so the second and third degree (where buyers live) never see it.
One caveat, then we play: These are patterns, not physics. Posts will break every rule here and perform well anyway. Posts will run all six plays and flop. What the plays give you is control over the part you can control — staying out of the penalty zone.
( ICYMI: Last month, we covered the mechanics of Linkedin’s new algorithm, 360Brew . Highly recommend reading if you need a refresher!)
Key takeaways:
Write for the save, the only signal a reader spends something to give. It carries roughly 5X a like. LinkedIn quarantines posts it reads as low-value to first-degree connections only, which looks identical to nobody caring. Strip the 5 AI red flags LinkedIn has named, starting with contrastive construction. The penalty targets emptiness, no matter who wrote it. Play 1: Write the joke you'd tell at a work happy hour A narrow, well-timed joke generates every signal the model rewards, at the same time, in about 4 seconds.
Someone stops scrolling, tags a colleague, bookmarks it next to whatever cursed content they already keep. Best case, they DM it to a coworker, which LinkedIn now tracks.
It's also the hardest thing on the platform to fake. Machines handle generic observations fine. But a joke about the gated PDF that pulled 11 downloads, 4 of them from your own team? Well, that requires surviving that marketing meeting.
The mechanism is specificity, and it's a ladder::
Rung 1: "Meetings are too long." True, and 40 million people have posted it.Rung 2: "Marketing meetings are too long."Rung 3: "The 4th round of feedback where someone who wasn't on the kickoff call asks if we can ‘make it pop.’"Rung 3 loses the general audience and attracts the 400 people who do this for a living. Those 400 people are the only ones the model needs to see engaging to figure out who else to show you.
The test: If someone outside your function would find it funny, you're on rung 1 or 2. Go narrower. If you're slightly worried it's too inside, you've arrived.
The example: a joke that turned into a policy documentLast October I posted a list of new AI red flags in copy. It included the rule of threes, "it's not X, it's Y," capitalizing the first letter of a sentence, being grammatically correct, writing like someone who's seen sunlight, and any sign of literacy whatsoever. I closed by suggesting we'd reached the Salem Witch Trials of writing, where breathing near an em dash gets you burned at the stake.
It got 293,323 impressions, 512 comments, and 226 reposts. Not too shabby.
Seven months later, LinkedIn's VP of Product named contrastive construction as an official flag. I wasn't predicting anything, I was complaining. But only people who write for a living could tell which flags were real and which were the bit, and getting the joke is what got them commenting.
Play 2: Post the receipts, not the recap The most un-fakeable thing you can publish is proof that exists only inside your company.
Sometimes a number, just as often an artifact: a Slack thread where a decision was made, the email that changed a client's mind, the brief template your team really uses.
"We don't have proprietary data" is almost never true. Look for the test that flopped (nobody posts the autopsy), a pattern across your accounts (12 programs is plenty), the before-and-after on a process metric, the objection you hear 4 times a week on sales calls.
Then screenshot the real thing. Redact names and ask permission first (I'm funny, not brave). A screenshot of an actual process beats ten posts describing one, because a reader can steal it on sight. That's all a save really is.
How to write it: Lead with the number. "We cut revision cycles 42%" earns the read. "I want to share something we've been working on" wastes your two highest-weighted sentences.
Play 3: Delete your first sentence The 360 Brew model front-loads your opening 2 sentences when it decides what your post is about and who should see it. Most posts spend both clearing their throat.
The cut test: Delete your first sentence. If the post still makes sense, it was never a sentence. It was a warmup, and it cost you half your ranking real estate.
Before: "As marketers, we all know that content briefing is a critical part of any successful process."
After: "Our briefing doc was 9 pages long and nobody had read past page 2 in a year."
Cover the byline on the “before” option and you'd never guess whose it was. What survives the cut is usually the middle of something that happened, which is its own quality check.
3 openers that survive the cut:
A number with no preamble: "9 pages. Nobody read past page 2." Dialogue: "'Can we make it pop' is not feedback, and I say that with love." A time and a place: "Sat in a QBR last Thursday where a VP asked why reach was down 70%." Play 4: Bring a POV you could defend in a room of your peers There's a difference between disagreeing with conventional wisdom because you've watched it fail, and disagreeing with it because ragebait performs.
The model reads the second kind as engagement bait, which it is. The first earns substantive comments at 2x.
The peer-room test: Could you hold this position in a room full of people who do your job, taking follow-up questions, for 10 minutes?
If yes, post it. If it only works because it's provocative, you don't have a position. You have a costume.
A POV is usually a pattern you've seen enough times to name. Which means the build is mechanical:
Name the thing everyone in your category repeats. Name the specific instance where you watched it not work. Say what you do instead, and what it cost you to learn that. Talking about the failures (point #2) is one people often skip, and it's the entire load-bearing wall. Without it you have a contrarian opinion. With it you have evidence.
Emma Miller , our creative director, ran this play on a DoorDash for Merchants ad that had been following her around the feed. Her carousel takes one element per slide, the headline, the tagline, the AI-rendered ribs, says what's wrong, then rewrites it. "Southern food orders are more likely to lead to soap orders" becomes "People who order BBQ also buy more bar soap."
The slide that turns it from a dunk into a POV comes near the end: "What was actually good," where she credits the proprietary data point the ad was built on. She's arguing with the execution and she brought fixes. 105,000 impressions and counting.
Play 5: Use LinkedIn newsletters as a bridge, not a home LinkedIn newsletters deliver by email and push notification. They don't compete for feed real estate at all. And every new connection you make gets an automatic subscription invite, which turns ordinary networking into list growth you aren't managing.
For an executive already posting consistently on a clear topic, this is the highest-impact distribution available on the platform, and it's mostly sitting unused.
How to set it up so it works:
Pick a cadence you can survive. Monthly beats weekly-then-nothing. The subscription is permanent; your motivation is not.Name it after the topic, not the person. "The Science of Linkedin" travels. "Thoughts from Sarah" requires already caring about Sarah.Make the first edition the best thing you've written all quarter. It's the one every new connection gets invited to, indefinitely.Keep the feed and the newsletter on the same 2 or 3 themes. The coherence check reads both.For a lot of our clients, this is the working model for the motion: get found in the feed, then move the relationship somewhere an algorithm doesn't get a vote.
Play 6: Comment like it's a post, because now it is Comments you leave on other people's posts carry their own impression tracking. Your comment is content with its own reach, in front of someone else's audience, and almost nobody treats it that way.
The bare-minimum now is three sentences, because "Great post" is what you type when you want credit for reading something you didn't. Name the specific line you're responding to, add something the post didn't have, then leave a real question.
AI Slop: "Great breakdown, thanks for sharing! 🙌"
Non-AI Bop: "The point about profile alignment is the one I'd underline. We audited 12 exec accounts last quarter and the four with the worst reach all had headlines describing a job they'd left. Curious whether you're seeing that show up faster on newer accounts."
Same 15 seconds of typing. One is networking. The other is a receipt that you scrolled past.
The anatomy of a comment worth leaving:
Name the specific thing you're responding to. Not the post, the line. This proves you read it and it's the part most people skip.Add something the post didn't have. A counterexample, a number, a place where it breaks. You're not agreeing; you're extending.Leave a door open. A real question, not "thoughts?"
What LinkedIn just did, and what we tell every client On July 30, LinkedIn starting rolling out a "seems like AI slop" report button into the post menu. Detection used to be a model reading your copy. Now it's your audience, deputized. They have always wanted to tell you your posts are bad, and thanks to LinkedIn's product team, it's now a feature.
Two more details confirm the direction:
Your dashboard will privately flag when readers think your content reads as AI, even when you wrote it. LinkedIn killed its own "enhance your post" feature, replacing it with a proofreader that changes words without changing voice. When a platform retires its own AI writing tool for flattening voices, that's the clearest signal you'll get.
So here's what to strip, and none of it is a style preference of mine. Each one is a named, documented flag, straight from LinkedIn's own mouth.
Contrastive construction. The "it's not X, it's Y" pattern, named by LinkedIn VP of Product Laura Lorenzetti in May 2026 as a hallmark of machine-written posts. Search your last 20 posts. You'll find it, and you'll feel a little sick. I did.Symmetrical list openers. "Top 10 reasons I…" flags before anyone reads item one."Agree?" and "Thoughts?" endings. The model learned these produce low-quality engagement.Cross-posting from X or Threads. LinkedIn can tell, and downranks it.Engagement pods. Detection is reportedly at 97% accuracy and Lempod got pulled from the Chrome Web Store. The risk stopped being theoretical.The line worth holding: the penalty targets low-value, generic content, whoever wrote it. The trigger is emptiness, and AI produces emptiness at scale. Tightening your own sentences with a tool is fine, and LinkedIn has said so.
Handing over the thinking is what gets caught.
GRAPHIC IDEA:
A funnel diagram illustrating how the algorithm restricts the reach of posts deemed "low-value."
•Top of Funnel (Wide): "High-Value Post" - Shows reach extending to 1st, 2nd, and 3rd-degree connections (represented by different colored nodes or concentric circles).
•Bottom of Funnel (Narrow): "Low-Value Post (The Penalty Zone)" - Shows reach restricted only to 1st-degree connections.
•Label: Clearly mark the restricted area as the "Penalty Zone" to emphasize the consequence of generic content.
Our position, since you’ve read this far Reach is just delivery. What matters is what happens after the right person reads. A post's number counts everyone it passed on the way through and says nothing about whether it reached the dozen people who could buy from you, hire you, or refer you.
In your dashboard, those two results look the same.
So we judge content by what came out of it.
Take my AI red flags post as an example: It got 293,000 impressions and generated zero leads (for me or Storyarb) I'd gone viral with an audience of copywriters, and copywriters don't buy content agencies. They work at them.
Meanwhile posts that topped out near 50 likes have produced five conversations with heads of marketing who had budget and a problem, because they spoke to one person's situation and that person felt found.
Fifty of the right people beats 293,000 randos on the internet, and it isn't close.
That's the position and purpose underneath all 6 plays. This 360Brew update doesn't ask you to write differently if you were already writing for a real audience. It makes that work measurable, and it puts a price on the alternative.
So get honest about what you're chasing. If it's a big number of impressions, go make memes and jokes. If it's pipeline, these six plays are how you get there.
Write something specific enough that a stranger decides future-them will need it back. That was right in 2019, right when reach was cheap, and it will be right the next time LinkedIn rips the system out and starts over.
Which it will, probably right after you finish your first carousel.
FAQ How does LinkedIn's algorithm weight saves versus likes? LinkedIn's 360Brew algorithm weights a save at roughly 5 times a like. Substantive comments carry about twice the weight of a like, and shares sit at baseline. Saves rank highest because a reader has to decide the post will be useful to them again later.
Why did my LinkedIn reach drop in 2026? Reach dropped for most LinkedIn accounts because 360Brew scores usefulness and originality instead of counting engagement. Posts the model reads as generic get limited to first-degree connections, so second- and third-degree readers never see them. The drop looks the same as low interest.
Does LinkedIn penalize posts written with AI? LinkedIn penalizes low-value, generic posts rather than AI use itself. Editing your own writing with a tool is fine, and LinkedIn has said so. Named flags include contrastive construction, symmetrical list openers, cross-posting from X, and endings like "Agree?" or "Thoughts?"
How do you write a LinkedIn post people save? Write something a reader can use again, then make it specific enough that only your function recognizes it. Post internal proof like a screenshot of a real process, delete your opening throat-clearing sentence, and take a position you could defend to peers.
