Social feeds were built to reward fluent performance. That is why Pangram‘s new dataset lands less like a surprise than a measurement of something many users already feel: AI writing is becoming part of the texture of public conversation, especially where professional identity is on display.
The company analyzed 1,002,627 posts collected from opted-in users of its Chrome extension since April 24, 2026, across LinkedIn, Medium, Substack, X/Twitter and Reddit. Pangram says each item was counted once, only posts longer than 50 words were scanned, and the analysis used Pangram 3.3, its latest detection model. The result is not a full census of the social web, but it is a useful window into what people actually encounter while scrolling.
The first pattern is length. Across all scanned items, Pangram found an average AI rate of 13.8%, but longform content carried a much heavier signal: 25.72% of posts over 250 words were flagged as fully AI-generated. That makes operational sense. Generative tools are most useful when the task is to turn a rough idea into a polished paragraph, a motivational post, a thread or an explainer. Short replies still benefit from human timing, context and friction.

LinkedIn is where that incentive structure becomes most visible. Pangram found that LinkedIn represented roughly a third of scanned items but nearly 62% of all AI content it flagged. More than 40% of longform LinkedIn posts were identified as fully AI-generated, and the company’s own summary says one in three top LinkedIn posts carried an AI signal. This is the platform where polished professional language is already a currency, so AI does not have to disguise itself as spam. It can look like productivity.
The mechanism is bigger than individual laziness. LinkedIn has offered writing assistance directly inside the product, now framed as post enhancement, while its ranking system rewards consistent posting, confident tone and broad engagement. That creates a loop: AI lowers the cost of producing professional-sounding content, the feed rewards volume and polish, and users see more posts that feel optimized before they feel personal.

Pangram’s comparison with Reddit shows why moderation cannot treat all AI writing as the same problem. Reddit had the highest scan volume in the dataset, but only 4.4% combined AI share, partly because replies made up most Reddit items and were overwhelmingly human-authored. Top-level Reddit posts were much more likely to be AI-written, while on LinkedIn a top-level post was 1.35 times more likely to be AI-generated than a comment. Visibility, not just volume, is the pressure point.
The unresolved test is whether platforms can separate useful AI assistance from synthetic social performance without turning detection into another opaque ranking layer. Pangram’s data does not prove that LinkedIn is overrun by bots. It suggests something more subtle: a professional network can become less human even when the accounts are real, because the language that fills the feed is increasingly produced by systems optimized to sound employable.
Discover more from TekCrispy
Subscribe to get the latest posts sent to your email.



TekCrispy