For the last couple years, using AI to write your posts was basically free money — faster output, similar reach, nobody could tell the difference anyway. That's quietly stopped being true, at least on the platform where it matters most for professional/creator content: LinkedIn. And the reason isn't what most people assume.
On March 12, 2026, LinkedIn's engineering team shipped a full rebuild of the feed — confirmed on LinkedIn's own Engineering Blog and covered by Search Engine Land. The old system was a chain of separate scoring models: one counted likes, one measured dwell time, one tracked click-through, one checked network proximity, then a final layer added the scores together. That structure was gameable — stack the right signals (early posting windows, engagement-bait hooks, hashtag stuffing) and reach would follow, whether or not the post said anything.
The new system throws that out. It's a single LLM-powered model that reads the post together with the author's professional profile and a given reader's history, then predicts one thing: would this specific person find this worth their time? Third-party analysts call the resulting signal the "depth score" — not an official LinkedIn metric, but a useful shorthand for what's actually being measured: sustained attention, not passive reactions.
The practical effect, per Richard van der Blom's widely-cited 2025/2026 Algorithm Insights benchmark: views down roughly 50% year-over-year, engagement down 25%, follower growth down 59% across the period around this change. That's not a platform in decline — it's a platform actively redistributing reach away from content that used to work and toward content that actually holds attention.
On May 20, 2026, LinkedIn announced a second, more targeted change — explicitly aimed at low-quality AI-generated posts, comments, and automation tools, confirmed via PYMNTS' coverage which quotes LinkedIn VP Laura Lorenzetti's own blog post directly. Flagged content isn't removed, it's suppressed so it doesn't spread beyond a user's immediate network. This builds on an already-declining baseline: company page organic reach on LinkedIn has been reported anywhere from roughly 1.6% to 2% of followers by different 2026 industry analyses — the exact figure varies by source, but the direction is consistent and severe.
Here's the important nuance, though: LinkedIn's own VP Laura Lorenzetti has said the detection system correctly flagged generic AI slop 94% of the time in early testing — but the company hasn't disclosed its false-positive rate, so how often legitimate posts get wrongly suppressed is still an open question. The bigger mechanism underneath that, though, is behavioral: generic AI content without a specific, credible insight produces near-zero dwell time and no saves regardless of whether it gets explicitly flagged — there's nothing in it worth stopping for, and the algorithm doesn't need to fingerprint the model that wrote it. Reader silence does the flagging for it either way.
That distinction matters if you're repurposing content with AI, because it means the fix isn't "make it sound more human" — it's "put something in it worth a human's time."
Most AI repurposing tools work the same basic way: paste your article in, get a caption or post back out. The tool has no memory of what you've written before, no sense of your actual voice, no idea which parts of your original piece were the interesting parts versus the connective tissue. So the output regresses toward "the platform average" — which is exactly what gets flagged as generic. Multiple 2026 breakdowns of AI content converge on the same handful of tells: predictable structure, generic emotional language, and an absence of specific, concrete detail. The exact percentages attached to each tell vary by source, but the pattern itself shows up consistently across the research.
That's the trap with naive repurposing specifically: you're not just reformatting an idea for a new platform, you're often stripping out the one specific, credible detail that made the original worth reading in the first place — because a generic repurposing pass optimizes for "fits the format," not "still says something."
A few things hold up consistently across the 2026 research on this:
This isn't a reason to stop using AI for repurposing — it's a reason to stop treating repurposing as reformatting. The platforms have gotten good enough at reading reader behavior that the old shortcut (same idea, new wrapper, ship it) doesn't survive contact with a real feed anymore. What still works is the same thing that's always worked, just enforced more precisely now: keep the one specific, true thing from your original content intact when you reshape it for somewhere new, and give a human thirty seconds with the draft before it goes out.
Sources: LinkedIn's own Engineering Blog (March 2026 feed rebuild), PYMNTS' coverage of LinkedIn VP Laura Lorenzetti's May 20, 2026 anti-AI-slop announcement (quoting her directly), Richard van der Blom's Algorithm Insights 2025/2026 Report (cross-confirmed via Dataslayer, Agorapulse, Vulse, and The State of Brand), and Dataslayer's April 2026 LinkedIn format analysis (verified directly at the source).
Turn your own content into a week of platform-ready posts.
Try the Repurposer free →