LinkedIn is now running classifiers that flag machine-made, automated content and quietly cut its reach. The company says it blocks hundreds of thousands of automated comment attempts a day and identifies generic AI writing with 94% accuracy in early testing.
2026-08-20 · by Moe Ameen
LinkedIn has moved from talking about "AI slop" to detecting and enforcing against it. In a May 2026 update, the company said it had begun limiting the reach of content that appears AI-generated and lacks a clear point of view, and building systems trained to tell posts that offer "genuine perspective" apart from ones that feel repetitive, generic, and empty. On July 30, 2026 it detailed the enforcement side: new classifiers that identify low-quality, machine-generated content and pull it out of the recommendations members see beyond their own network.
The scale figures are the news. LinkedIn says it detects and blocks hundreds of thousands of automated "slop" comment attempts every day, and that it has prevented "billions of other automation attempts (posting at scale, slop) in the last couple of months alone." In early testing, the company put its accuracy at flagging generic AI content at 94%; it did not publish a false-positive rate, which is the number that matters most to legitimate creators. LinkedIn's chief product officer Hari Srinivasan framed it plainly: "AI slop is a top priority for all of us. We really care about this. People come to LinkedIn to connect with real people and share their real perspectives, ideas, and expertise."
The control is a reach penalty, not a takedown. Flagged posts stay up and still reach the author's direct connections, but they stop circulating through the broader feed and out-of-network recommendations, so the practical effect is a quiet collapse in impressions. The stated targets are engagement bait, unoriginal "thought leadership," bot-generated comments, automation tools that post at scale, and formulaic patterns like the "it's not X, it's Y" template. LinkedIn also added a member-facing "Seems like AI slop" reporting button and a filter to view verified profiles only. The company says the rollout could take several months to fully register in the feed.
Context matters for how aggressive this is. Detection startup Pangram reported that more than 40% of the long-form LinkedIn posts in its dataset were flagged as fully AI-generated — the highest concentration among the major platforms it examined — which is the flood LinkedIn is responding to. But AI detectors are known to misfire, and a 94% accuracy claim with no disclosed false-positive rate means some genuinely human, or human-directed, content will get caught. A LinkedIn executive, Laura Lorenzetti, put the company's line this way: "When AI is overused, especially at scale and in an automated way, it dilutes the valuable insights that real human conversations can spark." Treat the specifics as an early snapshot and expect the thresholds to keep moving.
The important distinction LinkedIn is drawing is between automation and quality. What its detection control blocks is unattended machine behavior — bots posting at scale, generic templates, comment spam, content with no perspective. That is the opposite of how [Kompozy](/) works, and the difference is worth being precise about, because "I use AI to make content" and "I run a bot that sprays slop" are not the same thing to LinkedIn's classifier.
Kompozy is a generation-and-publishing engine, not an autoposter. Every output is governed by a single [Persona Brief](/glossary/persona-brief) that encodes your actual voice and a banned-word filter — the direct antidote to the formulaic, "it's not X, it's Y" tells the classifier is trained to catch. You turn one idea into a genuinely distinct LinkedIn post, a brand-exact [Carousel](/glossary/hyperframes), captioned [Persona Shorts](/glossary/persona-shorts) with your own likeness, a Blog Article, and an Email Newsletter — different native assets, not the same generic text copy-pasted everywhere, which is the pattern that gets suppressed. Crucially, nothing ships unattended: every asset passes a per-post human review gate before [Autopilot](/glossary/autopilot) schedules it and publishes to LinkedIn as one of the eight social platforms it fans out to, plus blog and email. A human signs off, the voice is yours, and the post carries a point of view — which is exactly the content LinkedIn's system is built to reward rather than bury. For the reporting-button side of the same crackdown, see our note on [LinkedIn's "Seems like AI slop" button](/news/linkedin-ai-slop-button); for a broader playbook, read [the new LinkedIn content playbook and where AI tools fit](/guides/linkedin-content-playbook-ai-creator-tools).
LinkedIn runs classifiers trained to flag content that reads as generic, automated, or machine-made — formulaic templates, engagement bait, "thought leadership" with no real perspective, and bot-generated comments. It says these systems identify generic AI content with about 94% accuracy in early testing, and separately that it blocks hundreds of thousands of automated comment attempts every day and has stopped billions of automation attempts at scale in recent months. It has not published a false-positive rate.
No. The enforcement is a reach penalty, not a takedown. A flagged post stays up and still reaches your direct connections, but it is pulled out of the broader feed and out-of-network recommendations, so impressions can drop sharply with no notification. The member-facing "Seems like AI slop" reporting button feeds the same detection system rather than removing content directly.
Not by itself. The classifier targets patterns of automated, low-perspective content — generic structure, formulaic templates, and posting at scale — not the mere fact that a tool was involved. A post written in a distinct voice with a genuine point of view is what the system is built to reward. The risk is generic output and unattended bulk posting, so the fix is voice, perspective, and a human review step, not avoiding AI.
Give every post a real point of view in a consistent voice, avoid the formulaic tells the classifier hunts for (the "it's not X, it's Y" template, engagement bait, copy-pasted generic text), and keep a human in the loop before anything publishes. A tool like Kompozy governs voice with a Persona Brief and banned-word filter, generates distinct native assets per platform rather than one recycled block of text, and runs a per-post review gate before scheduling to LinkedIn — which keeps the workflow on the quality side of the line LinkedIn is drawing.