LinkedIn doesn't want your AI slop anymore

LinkedIn has long been on the frontlines of the AI slopidemic. Now, the company is taking new steps to reduce the reach of posts that bear the hallmarks of AI-generated drivel.

If you spend any time on LinkedIn, then you know this can't come soon enough. In a blog post, the company's VP of Product Laura Lorenzetti says the changes will target everything from outright engagement bait, to recycled "thought leadership" and other "generic" content that "lacks the authenticity and originality." The company is also taking aim at posts and comments that have obvious signs of AI construction like "it's not X, it's Y," phrasing.

LinkedIn isn't sharing a lot of detail about how it's defining or detecting AI slop, but says that its engineers collaborated with its in-house editorial team to identify "patterns in how members engage, recognizing what adds perspective, context, or expertise versus what simply repeats existing ideas without contributing anything new." When identified by LinkedIn, these posts will no longer appear in other users' recommendations, though they'll still be viewable to a person's direct connections and followers.

While undeniably welcome news, LinkedIn is also trying to walk a fine line here. The platform offers a bunch of its own generative AI tools, including a big "rewrite with AI" button in its post composer. Even as it's cracking down on AI slop, the Microsoft-owned company is careful to say that "AI-assisted" content is still welcome so long as it contains original ideas or encourages "meaningful conversation."

While LinkedIn is hardly the only platform grappling with AI slop, the professional network, which even before the rise of generative AI was overflowing with shameless self promotion and borderline spam, has been hit particularly hard by the phenomenon. Earlier this year, LinkedIn members endured weeks of what I've dubbed "em dash discourse." It began with posts about how the punctuation mark was supposedly a "tell" of AI-written posts and rapidly spiraled into weeks of mind-numbing debate over its relative merits. (Reminder: Large-language models have been trained on the mostly-stolen work of human writers and authors who — guess what — love a good, old em dash.) Since then, I feel like I see just as many LinkedIn posts bemoaning the slopified nature of the LinkedIn feed as AI slop itself.

LinkedIn, for its part, says that initial results from this work are "encouraging" and that it expects further declines in the "weeks and months ahead."