LinkedIn has the most AI slop. Here's the data, and how to filter it
It isn't just you
If your LinkedIn feed has started to feel like the same post written a thousand times - a one-line hook, seven numbered lessons, a humble brag and "Agree?" at the end - there is now data behind the feeling.
Between April and June 2026, AI-detection company Pangram scanned more than a million posts across five platforms using its browser extension, and The Decoder published the results. LinkedIn came out on top by a wide margin.
The numbers
Share of posts classified as fully AI-generated:
| Platform | Short posts (50-250 words) | Long posts (250+ words) |
|---|---|---|
| 30% | 41% | |
| Medium | 28% | 31% |
| X | 9% | 29% |
| 3% | 13% | |
| Substack | 12% | 10% |
A few details make the LinkedIn figure starker:
- LinkedIn was about a third of the posts scanned but nearly two-thirds of the AI content found.
- These are posts judged fully AI-generated. Posts a person drafted and then "improved" with AI are not in the 41%.
- Pangram says its model flags conservatively, so the real share may be higher, not lower.
Across all five platforms, roughly one in four posts over 250 words was AI-generated. Reddit replies, by contrast, were 98% human - conversation is still hard to fake at scale.
Why LinkedIn in particular
None of this is surprising once you look at the incentives.
Posting is the product. On LinkedIn, visibility is career currency. Consistency gets rewarded, and "post every day" is standard advice. AI makes daily posting nearly free.
The format is already a template. The LinkedIn house style - short lines, a lesson, a call to engage - is exactly the kind of structure language models reproduce easily. The genre was formulaic before the tools arrived; the tools just removed the effort.
The platform encourages it. LinkedIn itself offers AI writing help for posts. When the composer suggests a rewrite, a lot of people accept it.
Engagement bait works. Posts ending in "Agree?" or "Comment YES and I'll send you the guide" collect comments, and comments drive reach. Whether a human wrote the post matters less to the ranking than whether people respond.
What you can do about it
LinkedIn doesn't offer a setting to filter AI-written posts. Your options today:
- Tell LinkedIn you don't want to see a post. It works per post and may nudge your feed, but you're one signal among millions and engagement bait performs well.
- Unfollow the worst offenders. Effective for a handful of accounts, exhausting for the hundreds who reach you through reactions and suggestions.
- Filter in your browser. A browser extension can read each post as it loads and hide the ones that look machine-written, without changing anything on your account.
That third option is what we built PureFeed to do. It gives each post a Likely AI score and folds away anything above the strength you choose - Light, Balanced or Strict. Every hidden post leaves a slim bar that says why ("Likely AI 92"), and one click shows it.
A note on accuracy
No tool can tell with certainty whether a person or a model wrote a paragraph, and it's worth being suspicious of anything that claims to. A filter should be honest about that in two ways:
- Err toward showing posts. Hiding a real person's post is worse than letting some slop through. PureFeed's default strength is tuned that way.
- Never delete, always explain. If you can see why something was hidden and bring it back instantly, a wrong call costs you a second, not a missed message from a colleague.
The bigger picture
The goal isn't a LinkedIn with zero AI. Plenty of people use it to tidy up a post they actually wrote, and that's fine. The problem is volume: when four in ten long posts are generated, the real updates - the job post from someone you know, the write-up of a hard-won fix - get buried.
Filtering the generated filler out is how you get those back. And if ragebait is the other half of your LinkedIn problem, the same extension scores posts for negativity and outrage too.
Data: Pangram scan of 1M+ posts, April-June 2026, via The Decoder.