Why millions of people are flagging AI posts on LinkedIn

LinkedIn recently added a specific option in its menu for reporting posts that appear to be written by AI. Since the feature launched in late July, over one million users have clicked it. The company is pairing this feedback with its own internal classifiers—tools designed to automatically scan and identify text generated by AI models. As a result, LinkedIn reports that users are now seeing 40 percent fewer posts identified as AI-generated compared to just a few weeks ago. The platform is also starting to notify users directly if their posts have been flagged by others as potentially automated.

A classifier is essentially a digital filter that has been trained to spot patterns unique to AI. To build one, engineers feed an algorithm—a set of instructions for solving a problem—a massive collection of both human-written and AI-written text. By analyzing thousands of examples, the software learns the subtle statistical signatures that distinguish the two, such as specific word choices, overly consistent sentence lengths, or a lack of personal nuance. When the software encounters a new post, it calculates the probability that the text originated from a human or a model. In this case, LinkedIn uses these calculations to decide whether to demote the post, meaning it is shown to fewer people in their home feeds.

This is an early attempt to address the clutter that happens when AI makes content creation instant and free. When it takes no effort to produce a post, the incentive shifts from quality to quantity, which can quickly drown out genuine human conversation. LinkedIn is essentially experimenting with a new form of digital hygiene. The move suggests that social platforms are realizing that without human intervention, automated systems risk filling the entire internet with hollow, repetitive noise. It raises an interesting question: if an AI post provides useful information, does it matter that it was written by a machine, or is our preference for human connection more important than the actual content?