Why does LinkedIn have the most AI-generated posts?

Pangram Labs reports that LinkedIn was "the most AI-saturated platform" among the five it scanned, with "more than 40% of longform posts flagged as fully AI-generated." The scan covers 1,002,627 posts from users who opted into sharing data through Pangram's Chrome extension, collected since April 24, 2026, and counts only items longer than 50 words. LinkedIn posts were a third of the scanned items but, in the excerpt's words, "accounted for nearly two-thirds (62%) of all AI content we flagged." Across all platforms the average AI rate was 13.8%, and 25.72% of items over 250 words were fully AI-generated. The excerpt's own reading is that people are "overwhelmingly willing to use AI to speak on their behalf in professional settings that are associated with their real identity," and less so on "casual and anonymous platforms."

Pangram offers that reason as an explanation, not as something it tested. The excerpt also lists LinkedIn's own part: a built-in "Write with AI" button, since rebranded "Enhance post," and an executive's announcement that the platform would detect and downrank AI-generated posts with an in-house algorithm. The excerpt notes that the announcement was itself AI-generated, and it is agnostic on whether the company is modulating AI in its feed. It does not report whether downranking changed any rate, so platform design and user preference both sit in the data without being separated.

Against the nearest notes, this is a measure of prevalence, not reach. Why do AI posts get likes without inviting conversation? argues that AI posts collect likes because they look comprehensive and authoritative. Pangram's figures say nothing about engagement, so they neither confirm nor test that mechanism. Does AI writing assistance change how readers perceive the writer? measures what assistance does to how a writer is perceived; the LinkedIn figure shows where identity-bound speech is most often machine-written, which is the exposure side of the same worry. The excerpt also cites the 35% figure for newly published websites as "researchers estimated," and How much of the internet is AI-generated now? is where that sample and method are recorded. The excerpt gives neither.

What the excerpt does not establish is substantial. It does not say whether the 13.8% average counts mixed human-and-AI items; it reports a "fully AI-generated" figure for LinkedIn, and for X/Twitter it adds a mixed category. That matters for the ranking, because with mixed content included X/Twitter was "the worst off," with almost half of its articles either fully AI-generated (23.9%) or AI-assisted (22.9%). The sample is self-selected users of one extension, not a random draw from each platform, and the excerpt does not say how items were chosen. The 0.01% false positive rate is Pangram's own figure for its Pangram 3.3 model. So the claim holds as one vendor's measurement in one opt-in sample: LinkedIn had the highest fully-flagged share of the five platforms. It is not an estimate of how much LinkedIn content is AI-written.