LinkedIn screens every post that leaves your network for slop. Here's what passes

That matters because the people outside your network are the ones a founder brand is for: the buyer who hasn't heard of you yet. Jurka didn't say how posts shown only to your own network are handled. Strangers, though, only see you if the post gets through.

“The filter isn't asking whether you used AI. It's asking whether there's a person in the post.”

The good news is what the filter is looking for. It isn't asking whether you used AI. It's asking whether there's a person in the post.

Why it works this way

LinkedIn has been unusually clear about this. In a June post on keeping conversations real, it described slop as "low-effort, AI-generated content that may sound polished on the surface" but lacks a real perspective. The same post said: "It's ok to use AI to help you write, but your posts and comments need to represent your voice and your perspectives." Chief Product Officer Hari Srinivasan put it more bluntly, as quoted by Pure AI: "AI and slop are not the same thing."

LinkedIn added a "Seems like AI slop" option for readers on 30 July. In August, Srinivasan said it had been used more than a million times, and that members were seeing "40% less views on what we classify as AI slop" than a few weeks earlier. According to Pure AI, LinkedIn credits that drop to its classifiers rather than the button.

The best outside data points the same way. MagicPost, which sells LinkedIn tools including an AI post generator, studied 687,000 posts from 16,000 creators, comparing each post with that creator's own average. It dates a change in LinkedIn's distribution to 27 July. Since then, posts with both concrete detail and lived experience have done 15% better than the creator's norm, up from 10% before. Posts with neither have done 23% worse, down from 13% worse.

Sounding like AI isn't free, but it's not getting worse. At equal substance, posts that read more like AI already got about 9% fewer impressions before 27 July. MagicPost couldn't detect any extra cost since.

Two caveats. A vendor that sells a post generator has a reason to like that finding. And the study itself says its AI score and its generic score move closely together, so it can't cleanly split the cost between them. But the results match what LinkedIn says it's doing, so we'd act on them.

The study found something else worth knowing. "Comment KEYWORD" posts, the ones that ask you to comment a word to get a guide, went from 10% above a creator's average impressions to 47% below since 27 July. Engagement bait now costs you reach.

How to write posts that get through

The test we'd use: could a competitor post this, word for word, and have it be true? If yes, it's slop, however well it's written. Here's a routine that fits in about 90 minutes a week.

1. Keep a running note on your phone called "things that happened". A client objection, a number from this week's pipeline, a pitch you lost and why, a price you changed. Two lines each. Ten minutes a day, mostly while you're waiting for something else.
2. Once a week, pick the one entry that taught you something. That's your post. If nothing taught you anything this week, don't post. In the MagicPost data, the first post after a break of two weeks or more gets 19% more impressions than the creator's average, so a skipped week is cheap.
3. Write the first draft yourself, badly, in the order it happened. Use AI after that if you like: to tighten it, cut it or find a better first line. Don't let it add a lesson you didn't learn.
4. Check for three things before posting: a number, a name of a thing (a tool, a sector, a city, a type of client), and a moment ("last Tuesday", "on the second call"). Each one is something a generic post can't fake.
5. End on a real question you'd like answered, or just stop. MagicPost found closing questions held steady after 27 July, while requests like "let me know in the comments" fell from 2% to 19% below average.
6. Turn off any tool that comments for you. LinkedIn's June post said it will act on comments posted at scale with automation tools, and on replies that just restate the original post.

Here's the shape, with made-up details so you can see it working:

• First line, the moment: "We lost a £40k website pitch on Thursday. Not on price."
• The detail: "The other agency brought a working prototype of the client's booking flow to the first meeting. We brought a deck."
• What you're changing: "From this month, every pitch over £20k gets two days of build time before the meeting. We'll lose money on some of them."
• The question: "Has anyone priced pre-pitch work into their proposals? I'd like to know what it did to your win rate."

No competitor could post that and have it be true. That's the whole point.

How to tell it's working

Impressions will wobble whatever you do. In MagicPost's hour-by-hour data, from posts published through its own tool, the median post has collected only 72% of its views by day three, so judge a month, not a morning. These are the signals we'd watch instead:

• Prospects mention a specific post on a first call, unprompted.
• DMs from people who match your ideal client, not other creators.
• Comments longer than one line, from people you don't know.
• Your profile views include people at companies on your target list.
• A referral arrives with "I've been reading your posts" attached.

If none of those show up after eight weeks of posts that pass the competitor test, the problem probably isn't the filter. It's the topics. Go back to the note on your phone and pick the entries that would make a buyer, rather than a peer, lean in.

LinkedIn is also testing a Network tab that shows only posts from people you're connected to or follow, which Srinivasan announced in September. If it rolls out, being followed by the right 500 people could matter more than reaching 50,000 strangers. Either way, the work is the same: write down what actually happened, and post the bits only you could know.