LinkedInLinkedIn downgrades posts that sound like AI: what it considers 'generic content' and how to write without losing reach
Team at AiToolsForAppsWritten with AI assistance15 min readIn summary
- As of October 8, 2026, LinkedIn reviews with classifiers every post that is distributed beyond the author's immediate network, according to its VP of Engineering, Tim Jurka (Social Media Today, 8-10-2026).
- The misleading data: 94% is accuracy, not coverage. Out of every 100 posts that the system marks as slop, about 94 are according to its criteria. LinkedIn has not published how many it misses.
- What LinkedIn punishes is not using AI, but the “generic and/or repetitive that does not contribute any real idea” (literal definition from Jurka). AI as a corrector is explicitly allowed.
- The impact is already measurable: content classified as slop recorded 40% fewer views, according to Chief Product Officer Hari Srinivasan (Social Media Today, 20-8-2026).
- What survives the filter is concrete: real hook in the first two lines (before the cut of “see more”), a fact or a personal case and first-person criteria.

If you post on LinkedIn and notice that correct, well-written, and well-formatted posts are stuck at 200 impressions, it’s probably not about the time of posting. It's a new filter, it's explicit, and its author has explained it publicly. This article breaks down what that filter looks for, what the numbers being repeated incorrectly out there truly mean, and how to write without getting stuck in the gutter. At AiToolsForApps, we maintain LinkedIn, a set of AI tools for those who create, sell, or compete in the professional network, and what follows is the context that explains why we set up tools for headline writing, carousels, and LinkedIn posts with counters and validators instead of a “generate me a post” button.
What exactly is “AI slop” for LinkedIn
For LinkedIn, AI slop refers to “generic and/or repetitive posts that do not contribute any real idea”. This is the textual definition given by Tim Jurka, VP of Engineering, on October 8, 2026. Notice what it does not say: it does not say “written with AI.” The criterion is the absence of substance, not the origin of the text.
This distinction has been reiterated by the company from the beginning. Hari Srinivasan, Chief Product Officer, summarized it with a phrase worth memorizing: “AI and slop are not the same”. And in May 2026, Laura Lorenzetti, VP Global Editorial, had already drawn the line: “It’s fine to use AI to help you write, but your posts and comments must represent your voice and perspectives” (Social Media Today, 21-5-2026).
The practical consequence: a post co-written with a model, but with your own data, a decision you made, and a debatable opinion, goes on the good side of the filter. A post entirely written by hand, but with three tips that anyone could have signed, goes on the bad side.
“When AI is abused, especially at scale and in an automated way, it dilutes the valuable ideas that can arise from real human conversations.” — Laura Lorenzetti, VP Global Editorial at LinkedIn, May 21, 2026.
What the 94% accuracy means (and what it does not mean)
The 94% is accuracy, not exhaustiveness. It means that, out of every 100 posts that the classifier labels as slop, approximately 94 fit LinkedIn's criteria. It does not mean that it detects 94% of posts written with AI: LinkedIn has not published its recall rate, so how many it misses is unknown as of today.
The difference is not academicism: it completely changes what you can deduce from the number.
| What is read out there | What LinkedIn said | What can be concluded |
|---|---|---|
| “LinkedIn detects 94% of posts made with AI” | 94% accuracy in marking slop | Nothing about how many go unnoticed |
| “94% of AI posts lose reach” | Marked posts lose distribution | It only affects the labeled ones, not everything generated with AI |
| “There is a 6% margin of error that protects me” | ~6 out of every 100 marked would be false positives | The margin plays against you, not in your favor: if you are marked, it’s most likely that they are correct |
Two more nuances that matter to decide what to do:
- The reach that is cut is that of recommendation, beyond the first degree. The marked content is not deleted and remains visible to those who follow you directly.
- The filter is applied at the distribution point: every post that is going to be distributed outside the author's immediate network is reviewed. That is, the screening occurs just before the post has the opportunity to grow.
Where the system comes from: a timeline of 2026
LinkedIn did not activate this filter suddenly. There is a sequence of six months, and each step explains the next one.
| Date | What happened | Source |
|---|---|---|
| May 21, 2026 | LinkedIn announces it will reduce the reach of AI-generated content without clear human perspective, limit automated comments, and add verified profile filters | Social Media Today |
| Late July / August 2026 | The option to report a post as "looks like AI slop" appears in the three-dot menu of each post | Social Media Today |
| August 2026 | LinkedIn removes its own "Enhance your post" feature (complete rewrite of the draft) and replaces it with a corrector that, in Srinivasan's words, "corrects your words but does not change your voice" | IBTimes UK |
| August 20, 2026 | More than 1 million people have used the button in less than three weeks; content classified as slop receives 40% fewer views | Social Media Today |
| October 8, 2026 | Jurka details the architecture and confirms that the screening covers every post that comes from the immediate network, with 94% accuracy | Social Media Today |
The most revealing detail of that table is not the 94%: it is that the million reports from August became the training set for the classifiers in October. The filter does not encode the opinion of a product team about what makes a good post; it encodes what a million professionals marked as filler.
How the classifier works internally
LinkedIn uses a teacher-student scheme: large models that keep up with new patterns of slop generate the material with which smaller models are trained, and it is those small models that classify at scale. On top of that, there are agents with specific policies and human reviewers for doubtful cases.

In Jurka's words: "we use what is known as a teacher-student setup. The "master" models, larger, are designed to keep up with the new patterns of AI slop and identify them accurately". And about the agents: "each agent is guided by a specific policy, which defines the criteria with which it evaluates the content: for example, whether a post is promotional, whether it celebrates an achievement, or whether it is current".
Three practical implications of this architecture:
- Changing the words is not enough. The system evaluates pattern and purpose, not vocabulary. Rewriting the same empty post with different synonyms does not save it.
- Patterns are updated. The "master" part exists precisely to pursue new formulas. The hook template that works this month is a candidate to be flagged next month.
- There is category signal. That an agent evaluates whether the post is "promotional" or "celebrates an achievement" means that the same text can be valued differently depending on its purpose.
The signals that betray an empty post
A post falls on the wrong side of the filter when it contains nothing that only you could have written: no dated facts, no measured figures, no named person or tool, no opinion that someone could dispute. The format signals are just the symptom; the cause is the absence of substance.

These are the signals that repeat in the marked content, as they are gathered from the analyses of the phenomenon:
- Template opening. The start that sets a scene ("Yesterday, in a meeting...") without stating anything.
- Lists without a single original data point. Three tips, three mistakes, three keys: impeccable structure, content interchangeable with that of any other author in the field.
- The formula "it's not X, it's Y". The rhetorical twist that sounds like a revelation and does not provide new information.
- Engagement bait. Phrases whose only goal is to provoke a comment or a "yes" ("Does this sound familiar? Comment YES").
- Ending with a generic question. "And what do you think?" as a final point, without a specific question about something concrete.
- Neat lesson at the end. The moral paragraph that wraps everything up with a bow and leaves nothing open.
- Chat-like format. Asterisks, headings, and bullet points copied exactly from a conversational tool: the most literal footprint of all.
None of these signals is illegal nor does it expel you from the platform. But their accumulation is exactly the pattern that a million reports taught the classifier.
What survives the filter: hook, data, and criterion
The post that survives is one that, in its first two lines, says something concrete and verifiable; that in the body provides a figure, a case, or a personal decision; and that takes a stance in the first person. The three elements are cumulative: with the hook, you only get the click on "see more," and without personal data, the classifier has nothing to distinguish.

1. The hook lives before the "see more" cutoff
The limit of a LinkedIn post is 3,000 characters, but the reader only sees the first two or three lines before the cutoff. This space varies depending on the device and whether the post includes an image: on mobile, it hovers around 140 characters, and on desktop, a bit more, with industry guidelines placing the practical cutoff around 210 characters (Commenti, character limits 2026). There’s no room for a narrative digression and a thesis: only one of the two fits.
The reason this matters is dwell time: if no one expands the post, the dwell time stays below a second and the distribution fades. Important: what is documented by industry sources is the role of dwell time as a signal for the algorithm; LinkedIn has not published the exact weight it assigns nor the thresholds. Treat it as what it is, a reasonable heuristic, not as an official figure.
2. A piece of data or a case that is yours
This is what truly makes a difference. What counts as substance:
| Type of substance | Example of what counts | What doesn't count |
|---|---|---|
| Dated fact | "The week of November 3rd, we changed the follow-up email subject" | "Lately, we have been trying things" |
| Measured figure | "The response rate went from 11% to 19% in 40 sends" | "The results improved significantly" |
| Named person or tool | "Marta detected it while reviewing the CRM" | "A team member noticed it" |
| Disputable opinion | "Qualifying by budget on the first call cost us two clients" | "It's important to listen to the customer" |
| Personal method | "What we did, in what order, and why it worked" | "The keys to success in sales" |
The litmus test is simple: if your post could literally have been published by someone else in your field by changing the name, it has no substance of its own.
3. Criterion in the first person
"In my experience, this doesn't work, and I'll tell you why" is a statement that someone can dispute. "It's essential to provide value" is not. What is disputable is, in practice, the most reliable marker that there is a human criterion behind it, and it is also what generates comments that are not bait.
Two distinct layers: what degrades and who is taught
The anti-slop filter and the profile relevance ranking are different layers of the same funnel, and confusing them leads to optimizing what is not necessary. The slop filter acts before distribution: it decides whether your post deserves to be distributed beyond your immediate network. The profile relevance ranking acts during distribution: it decides who is shown what has already passed the first cut.
| Anti-slop filter | Profile relevance | |
|---|---|---|
| What does it decide? | If the post is distributed outside the first degree | Who is shown it |
| When does it act? | Before distribution | When assigning audience |
| What does it depend on? | On the substance of the text of that post | On the fit between your profile and the interest of the reader |
| What fixes the problem? | Rewriting with data and criterion | Working on profile coherence and niche |
Order of priorities, therefore: first, make sure to pass the filter, because an impeccable profile does not distribute a post that is not distributed. And then work on the fit, because an excellent post shown to the wrong audience also does not perform. We have a specific article about the profile relevance ranking on this same blog; this section only situates the two layers, not repeating its content.
60-second checklist before publishing
Run each draft through these seven checks. If you fail three or more, do not publish: rewrite.
- Do the first two lines assert something, or just set a scene?
- Is there at least one figure, one date, or one proper name in the body?
- Is there a sentence that someone could dispute with arguments?
- Have I removed any sentence whose sole purpose is to provoke a comment?
- Does the closing open something, or does it close with a bow and a "what do you think?"?
- Is there any copied format from a conversational tool (asterisks, headings, automatic bullets)?
- Could another person in my field publish this text by only changing the name?
That last point is the summary of all the others. And it is, probably, the best homegrown approximation to what the classifier does.
Frequently Asked Questions
Does LinkedIn penalize posts written with AI?
Not simply for being written with AI. What LinkedIn degrades is generic and/or repetitive content that does not provide any real ideas, according to the definition by their VP of Engineering Tim Jurka on October 8, 2026. The company itself maintains an integrated AI checker, and its CPO has stated that "AI and slop are not the same thing."
What does it mean that LinkedIn detects AI slop with 94% accuracy?
It means that, out of every 100 posts that the system marks as slop, about 94 meet its criteria. It does not mean that it detects 94% of AI-generated posts: LinkedIn has not published how many it misses, so the actual coverage of the filter is unknown.
If a post is marked, does it disappear from LinkedIn?
No. The marked content loses distribution in recommendations beyond your first-degree connections, but it is not removed and remains visible to those who follow you directly. It is a loss of reach, not a removal.
How much reach is lost if you are marked as AI slop?
The public data is that content classified as slop recorded 40% fewer views than a few weeks earlier, as stated by Chief Product Officer Hari Srinivasan on August 20, 2026. This is an aggregated platform figure, not a fixed penalty per post, and LinkedIn has not published the breakdown on a case-by-case basis.
How many characters are visible before "see more" on LinkedIn?
It depends on the device and whether the post contains an image: only the first two or three lines are shown, around 140 characters on mobile and slightly more on desktop, with a practical cutoff that industry guides place around 210 characters out of a maximum total of 3,000. That’s why the thesis needs to be at the top, not in the third paragraph.
What happens when someone reports my post as AI slop?
An individual report primarily affects what that person sees in their own feed; LinkedIn has described the feature as informative rather than punitive and not as a channel for policy violations. The effect on distribution appears when many people mark the same thing, because those reports feed the training of the classifiers.
Can I use AI to write my LinkedIn posts without losing reach?
Yes, as long as the result contains something only you can provide: your own data, a specific case, a decision, and a debatable opinion. Using AI to structure, shorten, or correct is compatible with the filter; delegating the criteria to it is what activates it.
Does rewriting a post with different words avoid the filter?
It is not reliable. The system evaluates purpose and pattern more than vocabulary, so changing synonyms without adding perspective does not change the assessment. The master-student architecture exists precisely to track new formulas as they emerge.
How Your Writing Will Change from Now On
LinkedIn's screening is not an AI detector: it is a detector of emptiness. That is, in reality, good news for anyone with something to say, because the cost of publishing filler has just increased, while the cost of publishing your own figures has not. The only thing that becomes mandatory is a step that was previously optional: reviewing the draft to see what is in it that no one else could have written.
That step is what we try to make mechanical in LinkedIn, the AI tool app from AiToolsForApps for those who create, sell, or compete in the professional network: character counters with the cutoff line for "see more" marked, headline and "About" formulas, generation of carousels in PDF and posts with their alternative hooks, and a cleaner that alerts when it detects an engagement bait phrase and removes it. If you want to see it applied, here are the AI tools for covers, headlines, carousels, and LinkedIn posts, no prior registration needed.
Write your next post with the 'see more' cut in view, the measured headline, and the interaction bait out: try the AI tools for LinkedIn, no registration required.
Write your next post with the 'see more' cut in view, the measured headline, and the interaction bait out: try the AI tools for LinkedIn, no registration required.Keep reading
View all articles →
Why Your Reach on LinkedIn Decreased in 2026 (and What the Algorithm is Looking at Now)

Broken Geode in Geometry Dash 2.2091: which mods are not working and how to keep creating

DLS 27 in December 2026: what is confirmed, what is not, and what happens to your club, your squad and your kits when you update
