If you have felt the feed getting harder to sit through lately, you are not imagining it. Open LinkedIn on most days and it can feel as if the ratio has flipped. Three posts of "I'm thrilled to announce" for every one that tells you something, whether AI helped write it or not. What matters is whether there's a real person and a real point of view behind what you're reading. A decade of us posting a certain way turned the feed into this, and getting back to what made LinkedIn worth being on starts with understanding how that happened.
In 1973, sociologist Mark Granovetter found that people mostly got their jobs not through close friends or family but through casual acquaintances, people seen occasionally rather than every day. He called these weak ties, and the finding became one of the most cited in the social sciences because it showed that job leads often travel through people just outside our closest circle.
Reid Hoffman founded LinkedIn in December 2002 with Allen Blue, Konstantin Guericke, Eric Ly and Jean-Luc Vaillant, and the company became a real-world example of Granovetter's research. For a decade or so, LinkedIn delivered on that idea in practical ways. It was a place to find and stay in touch with people who mattered professionally, a leader you followed, someone you'd met at a conference. It was also a good way to pick up something useful along the way, industry news, a job opening, a piece of insight from someone doing the work.
How the Feed Changed Before Microsoft Ever Got Involved
Hoffman stepped back from day-to-day leadership by 2007. Jeff Weiner took over as chief executive in 2009 and built the revenue engine, resting on subscriptions, corporate recruiting tools and advertising, that took LinkedIn public in 2011. Microsoft bought the company outright in 2016 for $26.2 billion, still the largest acquisition Microsoft has made. It kept that business running largely unchanged and layered its own enterprise sales channels on top of it.
None of that touched the feed directly. What did was simpler and older than any acquisition. Engagement was the easiest thing for the algorithm to reward. Marketing analysts have described the algorithm as favouring posts that pick up comments and shares within the first 90 minutes of going live. Get that early reaction and a post gets pushed in front of more people. Miss it and the post disappears from view fast.
Out of that grew a mantra you'll have heard in some form at any marketing conference, post three or four times a week, stay visible, consistency builds a following. It wasn't bad advice, and it caught on widely. But it never asked whether a person had something worth saying that often, it simply assumed the answer was yes. Long before AI entered the picture, LinkedIn had built incentives that made frequency easier to reward than substance.
The billion-member milestone LinkedIn hit in 2023 wasn't down to that habit alone. Microsoft's own earnings disclosures point to the pandemic doing more of the work, sending millions of professionals onto the platform between 2020 and 2021 in place of the in-person networking, conferences and office life that had suddenly disappeared. The posting habit already in place shaped what all those new members did once they got there.
How AI Removed the Last Reason to Stop and Think
For as long as writing a post took real effort, the "post three or four times a week" rule had a natural ceiling, since not everyone had the time, the bandwidth or a fresh idea to share that often. In November 2022, that ceiling disappeared. ChatGPT exploded onto the market for free, requiring nothing more than an internet connection and an email address, and for the first time anyone struggling to keep up with the posting schedule had a tool that could fill the gap in minutes rather than hours.
Originality AI reported that long-form AI writing on LinkedIn rose sharply after ChatGPT launched, jumping 189% in the first two months alone (reported by Fast Company). Its estimate climbed to 54-58% of all long-form posts by the end of 2024 and reached 81% by the time the slop button arrived in 2026. Pangram, a separate detection firm using a different sample and a stricter word-count threshold over roughly the same period, put the figure closer to 41%. The disagreement matters. These are competing vendor estimates, not an independently verified industry number.
LinkedIn could resolve this cleanly, since it has full visibility into every post on the platform, but it hasn't published a figure of its own. That could reflect genuine difficulty in defining the category cleanly, or simple reluctance to put a hard number on something advertisers and members alike would want reassurance about.
No one can say for certain whether that reach decline comes from AI content, more creators competing for attention, or plain engagement fatigue. What's clear is how much AI-assisted writing there now is, a habit that started as an occasional shortcut and became, within about two years, common across a large share of long LinkedIn posts, though estimates of the exact scale vary.
Researchers have a name for a related risk. It's called AI search collapse, reported by Axios and studied formally as "retrieval collapse" in a 2026 ACM Web Conference paper. AI search tools are increasingly answering questions using pages that AI itself wrote, and each cycle of that narrows the pool of genuinely human sources left to draw from.
LinkedIn sits right in the middle of this. AI tools already cite it heavily when answering professional questions, and by at least one estimate, a large share of the long-form posts users publish on the platform was AI-written by 2026. Put those two facts together and LinkedIn becomes a live example of the exact feedback loop researchers are warning about.
What Has LinkedIn Done About the AI Slop Problem
Reach was already falling before any of this became a crisis. LinkedIn hasn't confirmed a figure, but third-party researchers and creators say it dropped by roughly half year over year by early 2026. One likely reason is simple crowding, since the average feed was carrying nearly three times as many posts per connection as it had in 2022, meaning any single post now competes against far more content for the same attention.
LinkedIn's answer was to ask members to flag the problem by hand. A new option appeared under the three-dot menu on any post, letting a member flag content that seems like AI slop, which reduces how widely LinkedIn shows that post afterward. There was no big launch announcement. Chief product officer Hari Srinivasan confirmed the change in a post on his own feed rather than a press release, reported the same day by Forbes and TechCrunch. For the poster, feedback stays private inside their own analytics dashboard.
LinkedIn had been part of the original problem too. A composer feature called "Enhance Your Post" let anyone run a draft through AI to polish it, and LinkedIn retired that feature the same day it launched the slop button, replacing it with a plainer proofreading tool.
Two weeks later, Srinivasan followed up with the actual numbers. He hedged deliberately rather than overselling them, noting everyone has a different network and feed, then confirmed views on content LinkedIn classifies as slop are down roughly 40% from a few weeks earlier. He was candid about the writing too, admitting he was "increasingly conscious on how to not sound like AI." A single report still can't demote a post alone, he said, since it's weighed against other signals, and LinkedIn has built safeguards to stop individual feedback being used to unfairly target other members.
LinkedIn hasn't published details on what stops coordinated flagging, or how the system tells a genuine quality problem apart from a reader who simply disagreed. Inc. raised exactly this concern. A crowd tends to be a fair judge of whether something felt real, and a far less reliable one of whether it was true.
There's a specific, documented reason to worry about that last point. A Stanford study published in the journal Patterns tested seven automated AI detectors on genuine student essays and found a 61% false positive rate for non-native English writers, against roughly 3-5% for native speakers. The plain, simple phrasing that AI models default to overlaps heavily with how people write in a second language. That study was about automated tools, not human judgment, so it doesn't prove the same bias shows up when a crowd flags posts by gut feel rather than an algorithm. But the underlying mechanism, plain phrasing reading as machine-written, plausibly applies just as easily to a person eyeballing a post as it does to a detector scoring one. That makes it a genuine risk worth LinkedIn addressing, not a settled fact either way.
Why the Fix Starts With What You Post Next
Picture the person from your conference small talk speaking to you in the same tone as one of their LinkedIn posts. You would not be reaching for their card. Yet that describes a large and rising share of long-form posts on LinkedIn now, a voice most people would never use in person. Writing in your own voice has become the exception, which is why it stands out when it happens.
None of this is an argument against AI. It has real uses, and plenty of people find it genuinely useful for specific things. There is a difference, though, between using AI to help you write something you actually think and using it to substitute for having something to say. The first can make a good post better. The second just makes it easier to produce more posts.
A business network only matters if the people on it are worth listening to. So what actually makes LinkedIn valuable, and how do we get back to it? Start with access, the fact that a network this size puts you within reach of leaders and specialists you would never otherwise get near. Access on its own does nothing, though. What matters is what people do with it, someone showing what they are building, a lesson pulled from actually running a business, a leader you can watch making something happen and explaining it honestly instead of dressing it up for the algorithm. None of that has ever been complicated, and none of it needs a machine to write it.
Fix the writing, and we fix the part of the reach problem that belongs to us. A network full of posts like that would never have needed a slop button, because the whole reason to be on it is that the people on it are bringing something real to the people connected to them. Every one of us can strengthen that network right now, one post at a time, beginning with the next one you write.