Guides

AI-generated LinkedIn posts: what the slop filter changes

LinkedIn now ranks down AI slop comments and flags generic posts. What changed since July, and how to keep using AI for LinkedIn without losing reach.

S

SelfAgentic Team

7 min read
A smartphone on a desk beside a keyboard, mouse and stylus, lit in teal and orange light.

On November 15, LinkedIn's chief product officer Hari Srinivasan said the platform had updated its classifiers so that suspected AI-written comments rank lower, and that testing cut views of "slop comments" by more than 20%, according to Social Media Today's report on the comment crackdown. That lands a month after LinkedIn's engineering lead said its classifiers catch AI-generated LinkedIn posts that count as slop with 94% precision.

If you use AI to write for LinkedIn, LinkedIn says it isn't penalising AI use as such. It is demoting posts and comments that read as generic or repetitive, and it now has a classifier, a report button and an analytics signal aimed at them. Fully automated posting and commenting is the riskiest setup you can run. A draft that a person edits and approves, built on something only you know, still works.

What LinkedIn changed between July and November

Each change gave automated content less room to work.

  • Late July: LinkedIn added a "Seems like AI slop" option to the menu on posts and comments, swapped its "enhance your post" button for an AI proofreader, and said posters would be told in analytics when members flag their posts, as The Register reported at the time. More than a million people used the report option in its first three weeks.
  • October 8: Tim Jurka, LinkedIn's VP of Engineering, defined AI slop as "generic and/or repetitive posts that added no real insight" and said the expanded classifiers have "94% precision at detecting AI-slop", per Social Media Today's write-up of his post. LinkedIn had earlier reported a 40% drop in views of slop.
  • November 15: suspected AI comments are now ranked lower under the default "most relevant" sort. They are re-ranked and stay up. Creator notifications were also changed to surface "high-quality comments you're likely to care about." Srinivasan said the change will reach everyone "soon."

Precision, the measure behind that 94%, tells you how often a flagged post really is slop by LinkedIn's definition. It says nothing about how much slop the system misses or how many posts it checks. LinkedIn hasn't published what the classifier looks at, so any checklist promising to "beat the filter" is guesswork.

Does LinkedIn penalise AI-generated LinkedIn posts?

For being AI-assisted, LinkedIn says no. Srinivasan put it plainly in July: "AI and slop are not the same thing," adding that "many people refine thoughts with AI, and we believe they want to know when they sound inauthentic." LinkedIn's own composer still offers AI proofreading.

The catch is that an unedited AI draft tends to be generic by default. It reaches for the safe opener and ends by asking readers what they think. The Register cited an Originality.ai scan that found 81% of 5,000 LinkedIn posts longer than 100 words were likely AI-generated. When most long posts sound alike, a classifier trained to find "generic and/or repetitive" content will catch plenty of them, whoever pressed publish.

For a founder, the useful test is whether the post contains a fact, a number or an opinion the reader couldn't have got from the previous ten posts in their feed.

Why LinkedIn AI comments are the first thing to switch off

Comment automation was already the most exposed part of the stack. In July, Srinivasan said LinkedIn catches "hundreds of thousands" of automated comment attempts a day and had blocked "billions" of other automation attempts in recent months.

The November change makes the remaining ones less useful. A bot that leaves "Great insights, thanks for sharing!" on fifty posts a day now gets ranked down in the comment thread, and the notification change means the author may never be told it arrived. You're spending effort, and some reputation, on comments few people will see.

If your tool posts through LinkedIn's official APIs, the LinkedIn Developer AI Policy points the same way. It tells developers to restrict automated posting of AI-generated content "without end user involvement" and to give users tools to edit generated content before it's published under their name. It also bars AI features that create or modify posts or comments for pages and members who aren't users of the app.

How to automate LinkedIn posts with AI without losing reach

Automation is still worth having for the slow parts, mainly research and first drafts. Keep a person on the parts LinkedIn is now measuring.

  1. Give the agent the research and the first draft, and keep publishing behind your approval.
  2. Feed it material only you have, such as a note from a sales call, a number from your own week or a decision you made and why.
  3. Edit each draft down to one specific point. Cut any line a competitor could post unchanged.
  4. Approve every post before it goes out, including scheduled ones.
  5. Watch post analytics for member flags. If posts get flagged, change the brief the agent works from instead of tweaking words.

This is the pattern we built SelfAgentic around. Its Content Lead agent drafts posts in your voice, and anything headed for LinkedIn waits in an Approvals inbox until you decide, on scheduled runs too. We wrote up which agent actions should need your approval in more detail.

A worked example: three posts a week for a small team

Say a three-person B2B startup where the founder posts on LinkedIn every Monday, Wednesday and Friday. Until now, a scheduler generated each post from a topic list and published at 9am, and a second tool dropped comments on target accounts' posts.

After the November change, a sensible rebuild looks like this:

  1. Sunday night, a research agent pulls three raw items from the week: a call note with a customer's exact objection, a competitor's pricing page change and a question a prospect asked in Slack.
  2. A writing agent turns each item into one draft post with the specific detail left in. Because it keeps memory across runs, the founder's edits from last week (shorter openers, no hashtags) carry into this week's drafts without being explained again.
  3. Monday morning, the founder spends about fifteen minutes cutting, adding one line of her own view and approving all three.
  4. Comments stay manual. The agent shortlists five posts from target accounts that are worth reading. She reads them and writes her own replies.

Posting frequency stays the same. Her time moves from writing from scratch to the part a generic draft can't supply, which is her view on something that happened this week. There's a cost: a person has to be available to approve, so a week of travel means a week with fewer posts.

The tool permissions are worth setting up deliberately too. In SelfAgentic, connecting LinkedIn and letting a particular agent use it are separate decisions, so the research agent can work without any ability to publish.

FAQ

Does LinkedIn remove AI-generated posts or comments?

Not according to its announcements. Suspected slop comments are ranked lower in the default "most relevant" view and stay on the post. Posts classified as slop get less distribution.

Is it against LinkedIn's rules to use AI to write posts?

LinkedIn offers its own AI proofreader, and its product chief has said AI and slop are different things. What it targets is generic, repetitive content and automated activity, and its developer policy asks API tools to keep users involved before AI content is posted.

How do I know if my LinkedIn posts were flagged as AI slop?

In July LinkedIn said posters would be told in their analytics when members flag a post as potentially AI. Check your post analytics regularly, and treat repeated flags as a sign your drafting process needs more of your own material.

Can I still schedule AI-written LinkedIn posts?

Scheduling a post you've read and approved is a different thing from publishing whatever a model produced. Keep a human review step for every post, and keep comments in your own words.

The practical takeaway

Switch off automated comments this week. Keep AI for research and drafts, give it material from your own work, and approve each post before it goes out. If you want agents that draft, remember your edits and wait for your sign-off before publishing, you can start building on SelfAgentic.

Keep reading

Ready to build your AI team?

See how a team of agents handles real work, and checks with you before anything goes out.

Free to start. No credit card required.