A report button is a labelling tool
On 30 July LinkedIn added an option to the three-dot menu on every post: Seems like AI slop. Reported posts get reduced reach in much the same way the platform already handles content marked as not interested. Alongside it came new classifiers intended to cut slop out of recommendations from beyond your own network, private notices to authors whose posts read as inauthentic through heavy AI use, wider access to profile and page verification, and an option to block comments from company pages you would rather not hear from.
The mechanism is easiest to see in what the company said about it. Chief product officer Hari Srinivasan described slop as hard to define, added that the definition changes, and said the reports let LinkedIn tune its models and make better feeds. Read as an engineering statement rather than a public relations one, that is a clear description of what the button is for. Human reports produce labels. The classifier trained on those labels applies the demotion at a scale no review team could reach. The button is the cheap half.
The standard is undefined on purpose
Why it matters: almost every other moderation rule you deal with is published. You can read a platform's policy on spam, on nudity, on impersonation, and predict with reasonable confidence how a given post will be treated. This one is stated as moving by design. That is honest engineering, because a fixed definition of slop would be gamed within a week, and it is a genuine problem for anyone whose distribution depends on the feed, because you cannot comply with a definition that is retuned continuously against fresh labels.
It also helps to be precise about what is being penalised. Srinivasan was explicit that AI use is not inherently the problem when it is refining an authentic thought. What draws the flag is a texture that readers recognise and report: the generic cadence, the tidy three-part structure, the closing line that could sit under any post. Texture is not a rule you can satisfy by checking a box. It is judged by an audience, and it moves as that audience gets better at spotting it.
Readers had already applied the penalty
The scale of the problem is documented outside LinkedIn. Pangram Labs, whose detection extension went out in April, scanned more than a million posts across LinkedIn, X, Reddit, Medium and Substack. On LinkedIn, more than 40 percent of long-form posts, meaning those over 250 words, came back as fully machine-generated. On X the figure was 25 percent fully AI-written with a further 23 percent AI-assisted. Reddit and Substack sat near 10 percent. By the study's count LinkedIn hosts roughly 62 percent of all the AI content it scanned across the major networks.
The bottom line sits in a different number. Those AI-written posts drew roughly 45 percent less engagement than human-written ones, and that was measured before any of this shipped. The reach cost was already being charged, quietly, by readers who scrolled past. What LinkedIn announced on 30 July adds a mechanical penalty on top of a behavioural one that was already running. If your company posts AI-drafted content, you do not need to wait to be flagged to learn what it costs you, because the comparison is already sitting in your own analytics.
In Europe the demotion has to be explained
LinkedIn was named a very large online platform under the Digital Services Act in the Commission's first designation round in April 2023, on the threshold of 45 million monthly active users in the Union. That designation carries consequences that bear directly on this announcement. Under Article 17, restricting the visibility of information a user provided, which expressly covers demoting content, obliges the platform to give that user a statement of reasons setting out the ground relied on and the facts behind it. Article 20 requires an internal complaint-handling system. Article 21 opens access to certified out-of-court dispute settlement.
That is a live asymmetry worth using. The definition changes is an acceptable answer to a reporter and a much weaker one against a statement-of-reasons obligation, which has to name a ground. A European company whose page or founder's post loses reach can request that statement rather than guess at the cause. Statements of reasons are also submitted to the Commission's public transparency database, so the aggregate behaviour of an undefined classifier becomes inspectable over time. The mechanism already exists, it costs nothing to invoke, and almost nobody does.
Native is not endorsed
One detail in the announcement deserves more attention than the button. In the same breath, LinkedIn retired Enhance your post, its own generative writing feature, and replaced it with a narrower proofreading tool that corrects grammar and spelling without rewriting the author into a generic voice. The platform that shipped the generator is now shipping the complaint button for the class of output the generator produced, and it did both inside one product update.
The durable lesson generalises well past this feed. A feature being native to a platform is not evidence that the platform will keep rewarding what the feature produces. The incentive to ship a tool is adoption of the tool; the incentive to rank its output is a separate calculation that can reverse without notice and without breaching anything you agreed to. Before you build a publishing workflow on a vendor's own AI feature, ask what your output is worth if that feature is withdrawn next quarter, and keep whatever the human wrote.
Read next: Brussels Fined the Risk Assessment, Not the Fakes | Brussels Called the Audit Weak and Approved It
