LinkedIn's AI Slop Button: What It Actually Flags

In July 2026, LinkedIn added a new option to the three-dot menu on every post: "Seems like AI slop." Click it, and the post gets hidden from your feed while your report feeds into the classifiers LinkedIn is training to detect low-quality AI content automatically. It's a small button with a blunt name, and it says a lot about where the platform thinks the last two years of AI writing tools have landed it.

Short answer up front: a single AI-drafted comment that you read before posting doesn't meet LinkedIn's own definition of slop. What does is content nobody checked before it went out, published at a volume no person could review. The rest of this post walks through why that distinction holds up.

What does LinkedIn's "seems like AI slop" button actually do?

The mechanic is simple. Open the options menu on any post, tap "Seems like AI slop," and that post disappears from your view. LinkedIn uses the reports as training signal for a classifier built to catch the same pattern automatically going forward. Posts that get flagged often enough lose reach beyond the original poster's own network, and the people posting them get a quiet note in their own analytics dashboard.

The scale behind the decision is what makes it notable. A Pangram study cited by Fortune found that around 40% of long-form LinkedIn posts and 30% of short-form posts were fully AI-generated. LinkedIn itself said it now blocks hundreds of thousands of automated spam comments a day and has stopped billions of other automation attempts in just the last couple of months. Hari Srinivasan, LinkedIn's chief product officer, called AI slop "a top priority for all of us."

The company also pulled its own "enhance your post" AI writing tool out of the compose box and replaced it with a proofreader, one that fixes grammar and phrasing without rewriting your voice into something generic. That swap is the clearest signal of what LinkedIn is actually trying to kill: not AI involvement in writing, but AI output that reads like nobody was involved at all.

Is an AI-drafted comment automatically AI slop?

No, not by that definition. It's tempting to read "seems like AI slop" as a proxy for "written by AI," full stop. That's not what the button is built to catch, and it's not what LinkedIn is describing when it explains the problem. Slop, as LinkedIn and everyone else using the term mean it, has three specific traits stacked together: it's generated with no reference to what it's replying to, published with no human check before it goes live, and produced at a volume no person could review one by one. Strip out any one of those three and the label stops fitting.

A comment an AI drafted from the actual caption in front of you, in your own tone, that you then read and posted yourself, fails all three tests. It's built from that specific post, not a generic angle. You read it before it went anywhere. And it's one comment on one post you're actually looking at, not thousands fired off overnight. Calling that slop because AI touched the draft stage is like calling a spell-checked email machine-written. The tool helped with the sentence. You decided what got published.

Automation is the line, not the model

Look again at what LinkedIn is actually reporting: billions of "automation attempts," hundreds of thousands of spam comments blocked daily. Those numbers describe comments posted by bots and scripts running unattended, hitting hundreds or thousands of posts an hour with output nobody at the other end ever looked at. That's the target, and it's a different animal from a person who used AI to get past a blank comment box faster, then read the draft before publishing it.

This is the same distinction we've argued before: the risk in AI-assisted commenting comes from skipping the human check at the end, not from the AI drafting stage itself. We wrote about that at length in why AI-drafted comments still need a human in the loop, and LinkedIn's new button is, in effect, the platform-level version of the same argument. If nobody reads a comment before it posts, at any volume, it's going to look like slop sooner or later, because eventually one of them will be wrong in a way a human would have caught in five seconds.

Where Comment Generator sits in this

Comment Generator was built around that human-in-the-loop line on purpose, not as a response to LinkedIn's announcement. The extension reads the caption of the specific post you're looking at, drafts a comment from that content rather than a template, and fills it into the platform's native comment box. It never clicks post for you. You choose the stance (agree, disagree, add insight), the tone, the language, and you can add your own standing instructions so drafts sound like your voice instead of a generic AI one. What lands in the box is a starting point you read, edit if needed, and publish yourself.

That also means Comment Generator for LinkedIn doesn't post at scale on its own and doesn't need your account credentials to run. It's a layer on top of the platform you're already logged into, not a script acting independently while you're not looking. Every one of the traits that make something slop, no reference to the actual post, no review, no cap on volume, is a case the product is deliberately built to avoid, not a feature it happens to lack.

What this means if you comment with AI help

Nothing about the button tells you to stop drafting with AI. It measures whether a human was part of the process before something went out under a name. If you're using AI to speed up writing comments, the practical takeaway is the same one that mattered before LinkedIn shipped this feature: read every draft before you post it, keep the volume to what you can actually review, and let the tool draft from the specific post rather than a generic angle that could sit under anything. That's also the structure behind how to write genuine LinkedIn comments that reference something specific instead of reading like they were pasted in from somewhere else.

What actually gets flagged as slop is volume with nobody checking it, not the fact that a model helped write a first draft. Keep a person reading before publish, and the label stops applying, no matter how that first draft got written.