Pull to refresh
Logo
Study finds over 40% of long-form LinkedIn posts are AI-generated

Study finds over 40% of long-form LinkedIn posts are AI-generated

New Capabilities

Pangram's scan of a million posts shows AI flooding professional feeds; LinkedIn says it will demote and flag it

Yesterday: LinkedIn says it will demote AI-generated posts

Overview

Updated 58 minutes ago

AI detection firm Pangram scanned more than a million social posts across five platforms and found that roughly 41% of long-form LinkedIn posts were fully machine-written. LinkedIn alone produced 62% of all AI content the company flagged.

The flood shows up in reader behavior. Posts that read as AI get about 21% less engagement than the same person's human-written posts. LinkedIn says it will quietly demote such content and privately warn users whose posts read as inauthentic.

Why it matters

If much of LinkedIn's writing is machine-made, the professional advice and expertise you read there quietly loses its basis in real experience.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

41%
Long-form LinkedIn posts fully AI-generated
Highest rate among the five platforms Pangram scanned, based on 250-word-plus posts.
25.7%
Long-form posts fully AI-generated across all platforms
One in four long-form items across LinkedIn, X, Medium, Substack, and Reddit.
62%
Share of flagged AI content appearing on LinkedIn
LinkedIn made up a third of scanned posts but two-thirds of all AI content flagged.
1,002,627
Posts scanned by Pangram, April 24–June 30, 2026
Opt-in data from users of Pangram's detection browser extension.
21%
Engagement penalty for AI-sounding LinkedIn posts
TypeSafe's study compared posts that read as AI to the same authors' human posts.

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

Sign up to generate historical perspectives on this story.

People Involved

Organizations Involved

Timeline

April 2026 September 2026

5 events Latest: Yesterday
Tap a bar to jump to that date
  1. LinkedIn says it will demote AI-generated posts

    Latest Policy

    LinkedIn announces in-house detection that demotes AI-generated posts, privately warns heavy AI users, and retires its 'Enhance your post' assistant. Reported in September 2026; exact date unconfirmed.

  2. Business Insider examines trust fallout of AI-saturated feeds

    Analysis

    Business Insider publishes analysis of the Pangram finding and the AI disclosure penalty, arguing readers sour on writers once they learn AI was involved. The piece circulates widely on TrueReddit.

  3. Pangram reports 41% of long-form LinkedIn posts are AI

    Study

    Pangram publishes findings that 25.7% of long-form posts on all platforms are fully AI-generated, with LinkedIn highest at 41%. LinkedIn accounts for 62% of all flagged AI content.

  4. Data collection closes at 1,002,627 posts

    Data

    Pangram closes its ten-week collection window having scanned more than a million long-form social posts.

  5. Pangram launches detector extension

    Launch

    Pangram releases a browser extension that flags AI content and begins collecting opt-in scan data across LinkedIn, X, Medium, Substack, and Reddit.

Scenarios

1

LinkedIn demotion cuts AI posts below 30%

Likely Resolves by Q2 2027

Discussed by: PYMNTS and Business Insider coverage of LinkedIn's announced algorithm change

LinkedIn's in-house detection demotes AI-sounding posts, and the retired 'Enhance your post' assistant removes the platform's built-in nudge toward AI writing. If the algorithm works and private warnings push users back to human drafting, the share of fully machine-written long-form posts falls from 41%. The company indicates it will also review posts flagged by its own tools for authenticity.

2

AI share of LinkedIn posts holds above 40%

Possible Resolves by Q2 2027

Discussed by: Thierry Poibeau and other researchers noting detection faces a moving target

AI writing tools keep improving while detectors chase a moving target, and creators keep generating posts faster than platforms can filter. If demotion is evadable and the retired assistant only slows new adoption, the fully AI-generated share of long-form posts holds at or above 41%. Detector accuracy claims also weaken as models train on AI-detector telltales.

3

False positive scandal hits AI detection

Possible Resolves by End of 2027

Discussed by: Fast Company coverage of the Orélien case and Stanford research on detector bias

Detectors mislabel distinctive human writing, and the AI disclosure penalty makes wrongful accusations costly. If a prominent author or public figure is publicly damaged by a false AI label, detection firms could face retractions, regulation, or legal action. That would reshape how confidently platforms apply demotions and private warnings.

Historical Context

2 moments from history that rhyme with this story — and how they unfolded.

February 2011

Content farms and Google Panda (2011)

In the late 2000s, companies like Demand Media mass-produced cheap articles matched to popular search queries, flooding Google results with thin, formulaic content. In February 2011 Google launched the Panda update, which re-ranked search results to demote mass-produced, low-value pages.

Then

Sites built on content farming lost large shares of search traffic almost overnight.

Now

Established the pattern of platforms using algorithm changes to counter automated content floods.

Why this matters now

LinkedIn's plan to demote AI-generated posts mirrors how Google answered automated content — a platform using its algorithm to police quality.

2017–2019

Deepfake video wave (2017–2019)

Face-swapping tools let anyone place real faces in fabricated videos, spreading false clips of politicians and celebrities. Detection research and platform policies raced to keep pace as public trust in video authenticity eroded.

Then

Companies and researchers built detection tools; platforms added takedown policies.

Now

Established a pattern where synthetic content generation outpaces verification, and suspicion falls on authentic content too.

Why this matters now

The same dynamic is repeating with text — generation capacity runs ahead of detection, and genuine human writing becomes suspect.

Sources

(11)