Pangram AI Detection Report Shows LinkedIn Dominates AI-Generated Content on Social Media

AI content now dominates longform posts on social media, and LinkedIn is the epicenter

Pangram’s first‑month data from its Chrome extension shows that more than 40 % of longform LinkedIn posts are fully AI‑generated, and overall AI‑generated content accounts for 13.8 % of scanned items across major platforms. This matters because professional feeds are being flooded with undisclosed machine‑written material, eroding trust in the very spaces where reputations and careers are built.

Key quantitative findings

One in four longform items (≥250 words) are fully AI‑generated (25.72 %).
Two‑thirds of all AI‑flagged posts come from LinkedIn, despite LinkedIn representing only a third of scanned items.
Nearly half of X/Twitter articles are AI‑generated or AI‑assisted (23.9 % fully AI, 22.9 % mixed).
Reddit shows the lowest AI share (4.4 %), but top‑level Reddit posts are 5.25× more likely to be AI‑generated than replies.
Substack is the only platform where longer posts are slightly less likely to be AI‑generated, with a combined AI rate of 21.9 %.

These numbers come from 1,002,627 posts scanned since the extension launched on April 24 2026, using Pangram 3.3, a model with a reported 0.01 % false‑positive rate.

Longform content is the low‑hanging fruit for AI writers

Across LinkedIn, X/Twitter, Medium, and other platforms, longer posts consistently show higher AI‑generation rates. The exception is Substack, where the AI rate stays flat or even dips for longer pieces. This pattern suggests that creators prefer to let LLMs handle the bulk of substantive writing, reserving shorter snippets for human‑only effort.

Why LinkedIn is the AI hotspot

Platform incentives – LinkedIn’s built‑in "Enhance post" (formerly "Write with AI") button encourages users to generate content with a single click.
Professional identity pressure – Users are willing to let AI speak for them in a setting tied to their real‑world reputation.
Volume effect – Although LinkedIn accounts for only 33 % of scanned items, it contributes 62 % of all AI‑flagged content.
Mixed‑format dynamics – Top‑level LinkedIn posts are 1.35× more likely to be AI‑generated than comments, yet when controlling for length, comments are slightly more AI‑heavy than posts.

Reddit’s composition paradox

Reddit contributed the highest scan volume (36.7 % of items) but has one of the lowest overall AI rates because 98.1 % of replies are human‑authored. Top‑level posts, however, match X/Twitter’s AI saturation (~11 %). This highlights a blind spot for anti‑bot measures that focus on reply spam rather than high‑impact top‑level submissions.

Community reactions and broader implications

Users on Hacker News note that LinkedIn’s AI slop feels inevitable, with comments like “AI has just accelerated what has always been out there” and “LinkedIn writing was always awful, now it’s AI‑enhanced.”
Some express frustration that AI‑generated posts are used for self‑promotion, especially in niche domains like anti‑piracy advocacy.
Others worry about a feedback loop: AI‑generated content fuels LLM training data, which in turn produces more AI‑generated content (see the “mycorrhizal fungi” anecdote).
Skepticism remains about detection accuracy; a comment asks for objective validation of Pangram’s claimed 99.98 % precision.

Methodology at a glance

Data source: Pangram Chrome extension scans any post longer than 50 words.
Timeframe: April 24 2026 – early July 2026.
Detection model: Pangram 3.3, advertised with a 0.01 % false‑positive rate.
Platforms covered: LinkedIn, Medium, Substack, X/Twitter, Reddit.
Opt‑in sharing: Users voluntarily contributed anonymized scan statistics to enable this research.

What this means for users and platforms

Transparency is essential – Without clear labeling, professionals cannot assess the authenticity of the advice or opinions they encounter.
Platform moderation may need to evolve – LinkedIn’s announced in‑house AI detection and down‑ranking is a start, but the fact that the announcement itself was AI‑generated underscores the difficulty of self‑regulation.
Users may need new tools – Extensions like Pangram’s provide real‑time alerts, helping readers make informed decisions about where to allocate attention.
Future detection arms race – As AI generation improves, detection models must keep pace; community expectations that stronger detectors will soon curb AI saturation are common but not guaranteed.

Conclusion

AI‑generated writing has become a pervasive problem across all major social media platforms, with LinkedIn bearing the heaviest burden—over 40 % of its longform posts are fully AI‑generated. While the trend mirrors broader internet‑wide estimates that roughly one‑third of new content is AI‑assisted, the professional stakes on LinkedIn make the issue especially urgent. Ongoing transparency tools and research, like Pangram’s extension, are critical for restoring user control over attention and preserving the credibility of online discourse.