Noise Compounds. Signal Needs an Owner.

What the count showed

The count, from the middle of August: the ten most recent conversations in my LinkedIn inbox, covering about three weeks, were all pitches. Ten for ten. Not one person who wanted to talk with me about something real. Half were automated sequences, follow-ups firing on a schedule no human was watching.

One message quoted my own connection count back at me as personalization: "You have 20.1k connections. If just 15% need help, we're talking about 3018 organizations right there." A scraper read my profile, dropped my number into a template, and did math on my network like it was inventory.

Another one opened with "Hi %FIRSTNAME%," and the next line read "Just kidding, Jim," then went straight into the pitch. The broken mail merge was staged. The glitch is now so familiar that senders fake it on purpose to stand out from the rest of the automation.

Two more details from my own account. LinkedIn's spam filter caught eight conversations in four years, so everything above came through the front door. And the two clearly genuine humans who wrote to me in the past month got filed by LinkedIn's sorting into the "Other" tab, while the automation landed in my priority inbox.

The dilution has been measured

My inbox is one man's sample. The platform-wide numbers are worse.

In October 2024, the AI-detection firm Originality.ai scanned 8,795 long-form LinkedIn posts and found 54 percent were likely AI-generated. Less than two years later, in July 2026, the same firm ran the same method and got 81.2 percent. A second firm, Pangram, measured a different way: a million posts pulled from real user feeds this spring and summer. LinkedIn made up a third of everything they scanned and 62 percent of all the AI content they flagged, the highest share of any platform in the study.

The two firms count differently, one includes partly AI-written posts and the other counts only fully machine-written ones, and both sell AI detection, so hold the decimals loosely. The direction survives any discount you apply.

How a platform loses its signal

Nobody at LinkedIn decided to become a B2B Facebook. The AI writing tools were a reasonable product call. The platform shipped AI post-drafting for subscribers in 2023 and built collaborative articles on AI prompts the same year. The video push was a reasonable growth call: a TikTok-style reel feed in 2024, video billed by the platform as its fastest-growing format, video growth featured in the earnings story. Every format that worked somewhere else showed up here, one reasonable decision at a time. When Wired asked in late 2024 how much of the feed was AI-written, the company said it did not track it.

I once suggested a test for your own homepage: cover the logo and read the words. Run it on your feed. Five years ago you would know within three posts that you were on LinkedIn, because the feed itself was differentiated. Professionals talking shop. Today, with the logo covered, the reels and the engagement bait and the AI-generated content could be any platform on your phone. Nothing was filtering for the old identity anymore.

That is the law underneath this note, and it applies to a lot more than one platform. Nobody chooses noise. It accumulates wherever no one is defending the signal, one reasonable decision at a time, and it compounds while everyone is busy.

The dashboards say record engagement

While all of that was happening, the official numbers stayed great. Record engagement, quarter after quarter. Double-digit member growth, five years running. More than 1.3 billion members.

The measured numbers tell a different story. Organic reach fell by a third in 2025 in the biggest dataset I could open, closer to half in a second study I know only secondhand, and the decline hit 98 percent of users.

The inbox numbers are starker. Across 13 million tracked connection requests, acceptance held steady while replies fell by roughly a third in a year. That data comes from the automation vendors' own customer campaigns, which makes it worse, not better. People still click accept, because accepting costs nothing. Answering means believing a human is on the other end.

Both sides are telling the truth. They are just measuring different things. LinkedIn reports activity: total members, total comments, total uploads, all counted on a base that never stops growing. It has never said how many members are actually active; outside estimates put it around 350 million a month, about a quarter of the headline number. Activity can set records forever while the thing the platform exists for, professionals actually talking to each other, quietly comes apart.

The numbers nobody can source

One more finding from the source-checking. Four statistics show up in almost every deck on this topic, and not one of them traces back to a real source.

If a number cannot be sourced, it does not get to shape your strategy.

The platform started cutting

What LinkedIn did next matters most, because it started cutting.

By late 2024 it retired the badge its AI-prompted articles handed out, saying the quality was too hard to maintain. In May 2026 it published a policy that content which "feels generic or repetitive, even if it appears polished on the surface" gets demoted, with classifiers that, in LinkedIn's own initial testing, catch generic content 94 percent of the time. And on July 30, 2026, it removed its own AI rewrite feature and shipped a report button that reads "Seems like AI slop." Its chief product officer put it plainly: "AI slop is a top priority for all of us." The company says it blocks hundreds of thousands of automated comment attempts every day.

Put the timeline together. The platform handed everyone the tools in 2023, said it was not counting in 2024, and by 2026 it was pulling its own tools off the shelf and asking members to flag the output. You do not ship a slop button for a problem you do not have. Cutting noise is deliberate work, it costs real money, and even LinkedIn had to assign people to it.

What the correction pays for

The correction also changed what performs, and the changes favor anyone with something real to say. Here is what the algorithm pays for now:

• Generic gets demoted. AI-obvious posts underperformed human ones by 30 percent on reach and 55 percent on engagement in the one study that measured it, a 2024 analysis I could only verify secondhand. Hold that one loosely.
• Saves and documents win. A save is worth roughly five likes, and documents and images beat video.
• Early conversation wins bigger. Genuine back-and-forth in the first hour multiplies reach around five times, per the same secondhand study.
• People beat company pages. In every dataset I could find.

And the audience did not leave. In the Content Marketing Institute's latest annual survey, fielded in mid-2025, 76 percent of B2B marketers put LinkedIn in their top three channels for thought leadership, more than any other channel. Your buyers are still on it, quieter than ever, doing their research before you know they exist.

The demotion list is generic output and automated volume. The tool is not on it. I use AI every working day, including on this note, and every number in it either traces to a source I opened or tells you plainly when it does not.

The feed did not rot because people used AI. It rotted because people pointed AI at volume. More posts, more sequences, more everything, with no judgment anywhere in the chain. Turn the same tools toward research, verification, and getting one true thing right, and they become the strongest advantage a small team has had on this platform in years.

The part that is about your company

I wrote this note because the platform just demonstrated, in public, with a billion members watching, the thing I watch happen inside growing companies every year.

No company sets out to fill itself with noise. The tools arrive one reasonable purchase at a time, the activity numbers keep growing, and the drift never announces itself. Noise compounds on its own. Signal is a decision somebody has to keep making.

That is the whole lesson of the platform's last three years. Noise compounds. Signal needs an owner. LinkedIn eventually put someone on it, at the cost of a very public correction. In a growing business, finding the few things in your marketing, technology, and AI that actually move the number, and holding the line on them together, is the seat I fill. It is where the Fractional CMTO engagement starts.

Five moves, in order

For the platform itself, here is how I run my own account now. It is the same discipline at feed scale.

1. Write down what LinkedIn is for, for you. Mine: business networking, sharing what works, and being findable by buyers who are researching quietly. That one sentence filters everything else. Without it, the feed decides for you.
2. Post less, and post truer. One post built on something real from your work beats five prompted ones. The algorithm now enforces that trade whether you like it or not.
3. Spend the first hour in the conversation. Real replies to real comments, yours and other people's. Three genuine commenters in the first hour outperform every posting trick in the datasets above.
4. Treat your profile as the trust check. Replies collapsed because trust collapsed. The first thing a buyer does before answering a message is look at who sent it. If your team's profiles read like the feed, the answer never comes.
5. Send fewer, sharper messages, built on real research. The line here is not templates versus no templates. Every good outbound motion uses them. The line is whether a person's research and judgment sit between the template and the send. Volume with nobody's judgment in the chain is the flood. My inbox up top is the proof of what the flood earns.

Where I landed

I started by admitting I wanted to visit LinkedIn less, and for a while I assumed that made me the problem. The data says otherwise, and it changed my behavior in a direction I did not expect. I am not visiting less. I am visiting differently. Fewer posts, more conversation, nothing published that I would not say to a client across a table.

The feed will keep filling. On your feed, the owner is you.

All signal. No noise.