# Does LinkedIn Penalise AI-Generated Posts? A 2026 Study of 2,016 Posts

Source: https://www.flypost.io/learn/ai-generated-linkedin-posts-study
Author: Sam Cassman, Founder, Flypost
Last verified: 2026-10-21

Does LinkedIn penalise AI-written posts? We can't see inside the algorithm, but AI-written posts clearly do worse. We checked 2,016 recent posts from 336 B2B creators with a dedicated AI detector, and each person's AI-written posts earned about a quarter fewer reactions than the posts they wrote themselves.

Are you wondering whether LinkedIn punishes posts written with ChatGPT or another AI tool? It's a
fair worry, and the answers out there don't help much. Some people say LinkedIn can spot AI and
buries it. Others say nobody can tell, so it makes no difference. Hardly anyone has checked either
claim against real posts, so we did.

Most people assume one of two things: AI posts are everywhere but harmless, or LinkedIn quietly
hides them. When we ran 2,016 recent posts from 336 B2B creators through a dedicated AI detector,
we found something in between. Just over half were mostly written by AI, and those posts got
noticeably fewer reactions than the same person's own writing.

In this article we'll cover how much of LinkedIn is now AI-written, how those posts perform,
whether LinkedIn itself is behind the gap, and how to use AI without paying for it. Read on to find
out more.

## How much of LinkedIn is written by AI?

Just over half of it. In our sample of 2,016 recent posts from 336 active B2B creators, **54% were
mostly written by AI**, and a third were completely AI-written, with no human writing detected at
all. Smaller accounts lean on AI the most, and the share barely moved between April and September
2026\.

**Share of posts mostly written by AI, by the creator's follower count**

|  | % of posts mostly AI-written | Notes |
| --- | --- | --- |
| 1k–10k followers | 63 | 702 posts |
| 10k–50k followers | 55 | 882 posts |
| 50k–500k followers | 38 | 432 posts |

_Source: Flypost · 2,016 posts from 336 B2B creators, Apr–Sep 2026, scored by the Pangram AI detector_

The bigger the account, the less AI. Nearly two in three posts from people with 1,000 to 10,000
followers were mostly AI-written, against fewer than two in five above 50,000.

Most people mix it up. Of the 336 creators, **a third used AI for nearly every post** (five or six
of the six we checked), more than half used it sometimes, and only **1 in 8 never did**.

Longer posts lean on it harder: **61%** of posts over 250 words were mostly AI-written. That fits
the figure that made headlines earlier this year.
[Pangram's July 2026 scan of over a million posts](https://the-decoder.com/linkedin-is-the-undisputed-king-of-long-form-ai-slop-according-to-a-study-spanning-five-platforms/),
reported by The Decoder, found 41% of long LinkedIn posts were AI-generated, more than on any other
platform it checked. Its posts came from whatever its browser extension's users saw in their feeds,
so the two figures measure different things, but they point the same way.

What didn't change was the trend. Following the same 336 people month by month, the share stayed
between 51% and 58% from April to September, with no sign of a rise or a fall.

So if you use AI to write your posts, you're in the majority. That also means your readers have
seen a lot of AI writing, which matters for what comes next.

## Do AI-written LinkedIn posts get less engagement?

Yes. When we compared each person's AI-written posts with the posts they wrote themselves, the AI
ones got **about a quarter fewer reactions** (24%) and **about 12% fewer comments**. We held the
format, the month and the length of the post steady, and the gap showed up for small, mid-sized and
large accounts.

The comparison matters more than the number, so here is how it works. Big accounts get more
reactions whatever they post, and they also use less AI. Comparing AI posts against human posts
across different people would mostly measure who uses AI, not what AI does.

So we only compared people with themselves. **236 of the 336 creators post both ways.** For each
of them, we set their AI-written posts against their own human-written posts: same person, same
audience, same follower count.

If your own posts typically get 40 reactions, a gap that size would put an AI-written one at about
30\.

**AI-written posts compared with the same person's own writing** (Reactions, compared with the same person's own human-written posts)

| Tactic | Reactions vs their own posts | Change in reactions | Creators tested |
| --- | --- | --- | --- |
| Mostly AI-written | 0.76× | -24% | 236 |
| Completely AI-written | 0.68× | -32% | 181 |
| Image posts only | 0.72× | -28% | 156 |
| 1k–10k followers | 0.70× | -30% | 86 |
| 10k–50k followers | 0.85× | -15% | 103 |
| 50k–500k followers | 0.65× | -35% | 47 |

_Source: Flypost · 236 B2B creators who post both ways, 2,016 posts, Apr–Sep 2026. Format, month and length held steady._

> A note on where we stand. Flypost makes AI-assisted writing tools, so this isn't the result we'd
> have chosen. We're publishing it because it's what the data says, and because it changes how we
> think AI should be used for LinkedIn.

### Does it matter how much of the post AI wrote?

Yes. The more of the post AI wrote, the bigger the drop. Posts the detector judged **completely
AI-written** got about **32% fewer** reactions than the same person's human-written posts, against
24% for posts that were mostly AI.

That pattern is one of the reasons we trust the result. If the gap were a fluke, there's no reason
it would grow with the amount of AI.

So if you use AI, keep more of the words yours. The fully generated post is the one that costs the
most.

### Do AI posts do worse for big and small accounts alike?

Mostly, yes. Accounts with 1,000 to 10,000 followers saw about **30% fewer** reactions on their AI
posts. Above 50,000 followers it was about **35% fewer**, though that's a smaller group (47 people),
so treat it as a strong hint.

The middle band, 10,000 to 50,000, showed the smallest gap, about **15% fewer**. On its own that one
isn't quite enough to be sure, but it points the same way as everything else.

The one thing we can't tell you about is text-only posts. Only 29 people posted both AI and human
text-only posts, too few to say anything.

### Does every AI post do worse?

No. For **6 in 10 people**, their AI-written posts did worse than their own writing. For the other
4 in 10, they did better.

So the average is real, but it isn't a rule for every account. With only six posts per person, any
one person's result is partly luck. The pattern only shows up clearly across the whole group.

The practical move is to check your own numbers. Look at your last few AI-assisted posts next to
your last few written from scratch. If AI posts are your low ones, you're one of the six.

## Does LinkedIn penalise AI content?

LinkedIn says using AI isn't the problem, but it is going after low-quality "AI slop". In July 2026
its chief product officer said LinkedIn was ramping up new classifiers to spot slop. Our data can't
tell whether the gap we measured comes from LinkedIn showing these posts to fewer people, or from
readers scrolling past them.

LinkedIn added a "Seems like AI slop" button to posts in late July 2026. Its chief product officer,
Hari Srinivasan, said at the time that members flagging posts would help train the new
classifiers, and drew a line between the two things:
[using AI and posting slop aren't the same](https://thenextweb.com/news/linkedin-seems-like-ai-slop-button),
because plenty of people use AI to refine their own thinking (The Next Web, July 2026).

A month later he reported that members were already seeing
[40% fewer views on posts LinkedIn classifies as AI slop](https://engadget.com/2241857/linkedin-says-its-ai-slop-button-is-working)
(Engadget, August 2026).

That's the closest thing to an official answer. LinkedIn doesn't say it penalises AI. It says it
penalises slop, and it decides what counts as slop.

> **What our data can and can't see.** We measure reactions and comments, not views. If LinkedIn
> shows AI-sounding posts to fewer people, they'd get fewer reactions. If readers lose interest when a
> post sounds machine-written, they'd also get fewer reactions. Both fit what we found, and we can't
> separate them. Either way, the fix is the same: make the post sound like you.

## Does using AI cause the drop in engagement?

We can't prove it does. People might reach for AI on busy days, or for routine posts such as event
plugs and announcements, which might do worse however they were written. We ruled out the format,
the timing and the length of the post, but not its topic. So the safe claim is that AI-written
posts get fewer reactions, not that AI costs you reactions.

There is a likely explanation, though, and it comes from our main study. In our
[analysis of 116,000+ posts](https://www.flypost.io/learn/linkedin-algorithm-study-data), posts where the writer told a
personal story earned
[about a third more reactions](https://www.flypost.io/learn/linkedin-algorithm-study-data#do-personal-stories-work-on-linkedin)
than the same person's other posts. Generic advice, with no story and no concrete detail, earned
about 13% less.

AI is good at generic advice and has no stories of its own. In this sample, advice posts were more
likely to be AI-written than personal stories (60% against 48%). If AI pulls your posts towards the
generic, the gap we measured is roughly what you'd expect.

## Can you tell if a LinkedIn post was written by AI?

Not by asking a chatbot. A dedicated detector does much better. We used Pangram, which a
University of Chicago study in August 2025 found made almost no mistakes in either direction on
medium and long passages. Before we used it, we tried having an AI model rate posts instead, and it
caught only 1 in 3 of the AI-written posts the detector found.

A detector is a model trained only to tell human and machine writing apart. It isn't the same as
pasting a post into ChatGPT and asking for an opinion.

The independent check on Pangram comes from
[Artificial Writing and Automated Detection](https://bfi.uchicago.edu/wp-content/uploads/2025/09/BFI_WP_2025-116.pdf),
a working paper by Brian Jabarian and Alex Imas at the University of Chicago's Booth School of
Business, August 2025. They tested four detectors and found Pangram had a near-zero rate of false
alarms and missed cases. It was the only one that stayed accurate under the strictest limit on
false accusations. No detector is perfect, so some of our 2,016 posts will have been misjudged, but
it was the most accurate of the four in that test.

We learned the chatbot lesson the hard way: an earlier, cheaper version of this check missed two in
every three AI posts, so we threw those ratings out. Nothing on this page uses them.

## How should you use AI to write LinkedIn posts?

Use it as help, not as the writer. The posts that did worst were the ones written entirely by AI, so
keep the words, the stories and the details yours, and use AI to tidy what you've written.

1. **Write the first draft yourself, even a rough one.** Completely AI-written posts did worst of
   all, about 32% fewer reactions. Let AI shorten, fix and suggest, not start from a blank page.
2. **Put in something only you know.** A thing you did, saw or got wrong, a real number, a client's
   name (with permission). In our main study,
   [personal stories earned about a third more reactions](https://www.flypost.io/learn/linkedin-algorithm-study-data#do-personal-stories-work-on-linkedin).
3. **Write your first line yourself, about yourself.** First lines with I, my or we
   [earned about 19% more](https://www.flypost.io/learn/linkedin-algorithm-study-data#do-first-person-linkedin-posts-perform-better)
   in the main study. Our [guide to LinkedIn hooks](https://www.flypost.io/learn/linkedin-hooks) shows how.
4. **Cut the parts anyone could have written.** If a sentence would fit in any post on the topic,
   it's the sentence most likely to read as AI.
5. **Check your own numbers.** Compare your recent AI-assisted posts with your own. For 4 in 10
   people, AI posts did fine, and you may be one of them.

**Instead of:** In today's competitive landscape, building trust with your audience is more important than ever.

**Try:** I lost my biggest client last year because I stopped calling them. Here's what I do differently now.

_Illustrative rewrite: generic advice → something that happened to you · Illustrative example, written by us. Not a real post from the study._

The second line is harder to write, which is exactly why it works. No tool can write it for you,
because only you know what happened.

## How we tested AI-written LinkedIn posts

**Who we looked at.** 336 English-speaking B2B creators with 1,000 to 500,000 followers, picked at
random from the 948 people in our [main study panel](https://www.flypost.io/learn/linkedin-algorithm-study-data#methodology)
who posted at least once a month from April to September 2026.

**What we collected.** One random original post per person per month, any format, at least 50
words: **2,016 posts**. Reactions and comments were counted at least 14 days after posting, once
the numbers had settled. Reposts were excluded.

**How we spotted AI.** Every post went through Pangram's AI detector in October 2026. "Mostly
AI-written" means the detector judged more than half the text to be AI-generated or AI-assisted.
"Completely AI-written" is its strictest label: no human writing detected.

**How we compared engagement.** We compared each person's AI-written posts with their own
human-written posts, never with other people's. A statistical model held the post's format, month
and length steady, and treated each person as their own baseline. 236 people had both kinds of
post. Every range was checked by re-running the analysis on a thousand random resamples of the
people.

**What we're less sure about.** Six posts per person is enough for the group but not for any one
person, which is why we don't quote individual results. Some groups were small (47 accounts above
50,000 followers; 29 people for text-only posts). The detector will have misjudged some posts. We
measured reactions and comments, not views. And we couldn't hold the topic of a post steady, which
is the main reason we can't say AI causes the drop.

**Who paid.** Flypost, which sells AI-assisted LinkedIn writing tools, paid for the detector
(about $250) and ran the analysis. The wider dataset and method are in our
[LinkedIn algorithm study](https://www.flypost.io/learn/linkedin-algorithm-study-data). Spot a flaw? Tell us and we'll
correct it here.

## Frequently asked questions

### Does LinkedIn penalise AI-generated content?

LinkedIn says using AI isn't the problem, but it is cracking down on 'AI slop', low-quality posts it now tries to spot with automated classifiers. Whatever the cause, AI-written posts do worse. In our study of 2,016 posts, the same person's AI-written posts got about a quarter fewer reactions than their own writing.

### How much of LinkedIn is AI-generated?

Just over half of posts, in our sample. A dedicated AI detector judged 54% of 2,016 recent posts from 336 active B2B creators to be mostly written by AI, and a third to be completely AI-written. Smaller accounts used AI the most: 63% of posts from people with 1,000 to 10,000 followers.

### Do AI-written LinkedIn posts get less engagement?

Yes. When we compared each person's AI-written posts with their own human-written posts, in the same format, month and length, the AI ones got about 24% fewer reactions and about 12% fewer comments. Posts judged completely AI-written did worse still, at about 32% fewer reactions.

### Can LinkedIn detect AI-written posts?

LinkedIn says it is rolling out new classifiers to spot AI slop and low-quality posts, and that members flagging posts helps train them. It hasn't said how they work. Asking a chatbot is a poor test: in our checks, an AI model caught only 1 in 3 of the posts a dedicated detector flagged.

### Is it OK to use AI to write LinkedIn posts?

Using AI for help is fine. Letting it write the whole post is where engagement dropped most in our data. Write the first draft and the first line yourself, keep your own stories and details in, and use AI to tidy rather than to write.

---

Published by Flypost (https://www.flypost.io), the LinkedIn content studio. More guides: https://www.flypost.io/learn