Slop Is Hard to Define.
Your Reach Isn’t.
On July 30, 2026, LinkedIn began rolling out a button that lets any reader flag a post as Seems like AI slop — and a flagged post can lose reach before a single claim in it gets checked. LinkedIn’s own chief product officer says the definition keeps changing. The reach reduction doesn’t wait for it to hold still.
A new option started appearing on LinkedIn posts on July 30, 2026: Seems like AI slop. Click it, and the report feeds a classifier LinkedIn says is built to reduce a flagged post’s reach in feed recommendations shown outside the poster’s own network — before a single fact in the post has been checked, before a single reader has vouched for it, before anyone has asked whether the person posting knows what they’re talking about.
Hari Srinivasan, LinkedIn’s chief product officer, announced the feature himself: “AI slop is a top priority for all of us,” he wrote. “We really care about this.” As Fortune later reported, he also said something a launch post rarely needs to: “Slop is hard to define and the definition changes; this lets us tune our models and make better feeds.” The company isn’t claiming to have solved the classification problem. It’s chasing a target that, by the same chief’s admission, won’t hold still.
LinkedIn already carries a meaningful share of professional recognition: the feed a peer scrolls, the profile a referral checks, the post that reaches someone deciding who to call. A reach decision made inside that room now runs on a signal its own architect calls unstable — nothing in what’s been said points to credentials, sourcing, or track record as an input. LinkedIn’s spokesperson has compared the effect to a reader marking the post “not interested.” A post can lose reach on register alone, before anyone — a reader, a hiring manager, a referral already halfway to calling — has weighed the entity’s standing. What the classifier actually asks isn’t whether the post is true — it’s whether the post reads like something a machine wrote.
What the button is actually measuring
The trigger is stylistic, not factual. The report button, and the classifier it feeds, are built around surface features associated with AI-generated prose, not around whether a claim is true, sourced, or earned. Nothing in what Srinivasan said rules out precise, disciplined writing tripping that same signal as manufactured filler.
Evaluation never precedes the suppression. Reach reduction in out-of-network feed recommendations is designed to execute on the classifier’s read, before any reader has assessed the post’s substance and before the entity’s actual authority enters any calculation about whether the room would see it.
That standard is admittedly provisional. Srinivasan’s framing concedes the definition moves as the tool gets tuned. A register that clears today’s bar isn’t guaranteed to clear next month’s, and as far as the announcement itself says, no published criteria yet exist for an entity to check itself against in the meantime.
The fair read, and the harder one
LinkedIn isn’t wrong to be fighting something real: manufactured engagement degrades a feed for everyone, and the company reports blocking hundreds of thousands of automated comment attempts daily and a further wave of automated activity in recent months. Srinivasan drew a clear line in the same breath: “AI and slop are not the same thing; many people refine thoughts with AI.” That’s the chief stating a genuinely hard classification problem in public, on the record.
Underneath the stated rationale sits a harder mechanic. A tool built to catch manufactured filler is, by construction, also a tool that scores real writing by ear. That scoring runs on a standard Srinivasan himself says is still being tuned, at exactly the platform where professional reputation increasingly gets read before it gets asked about. That’s Answer Engine Authority™’s third phase, content authority structure, meeting a variable nobody designed for: the platform doing the structuring can move its definition of authenticity without warning, and an entity finds out only after the reach is already gone.
Nobody is being asked to write worse to survive a slop filter — that would be a kind of failure in itself, and the wrong lesson to take from this. What holds up regardless is what this record keeps pointing to: corroboration and structure durable enough to survive a proxy that admits, in its launch language, that it isn’t finished being defined.
In the July 30, 2026 LinkedIn post announcing the feature, Hari Srinivasan describes the rollout in his own words as “ramping,” not a completed launch. “AI slop is a top priority for all of us. We really care about this” and “AI and slop are not the same thing; many people refine thoughts with AI” are confirmed directly against that post; “Slop is hard to define and the definition changes; this lets us tune our models and make better feeds” is Srinivasan’s own quote as reported from the same announcement by Fortune. TechCrunch (July 30, 2026) corroborates the feature and Srinivasan’s quotes; Engadget (July 30, 2026) independently corroborates the phased-rollout framing (“testing a new reporting tool”) and additionally reports the “not interested” comparison quoted above, attributed to a LinkedIn spokesperson.
LinkedIn’s own figure for automated-comment-blocking, hundreds of thousands of attempts detected daily, is corroborated by all three outlets. On broader automation blocked in recent months, the count splits two to one: Engadget and Fortune both report billions, while TechCrunch alone reports millions. Neither figure is treated as settled in the body copy above.