People browsing LinkedIn and X are seeing a lot of writing that Pangram says was generated by AI, according to browsing data published by the AI-detection company. The finding is useful because it measures posts users actually encountered, rather than counting junk pages sitting unseen in some forgotten corner of the web.
Pangram said it analyzed roughly 1 million posts viewed by users of its Chrome extension over two months. Those users opted in to share detection results with the company. The extension runs in the background and labels text a user sees as likely human-written, AI-assisted, or AI-generated.
The numbers were highest for longer posts. Pangram split the material into shortform posts, between 50 and 250 words, and longform posts, above 250 words. It said about 40 percent of long LinkedIn posts in the sample were fully AI-written. On X, Pangram said a quarter of articles were fully AI-written and another 23 percent were AI-assisted. Longer posts on Reddit and Substack were lower, at roughly one in 10 identified as AI-generated.
Max Spero, Pangram’s chief executive, told 404 Media that the company was trying to measure how much AI text people run into during normal browsing. He described AI content as a cost imposed on readers’ attention.
What the data can and cannot prove
Pangram’s approach has one clear advantage over web-wide estimates: it tracks material that appeared in front of participating users. That distinction matters. A bot can fill the web with unread pages, but users experience platforms through feeds, recommendations, comments, and article formats that companies choose to surface.
The method also has limits. AI detectors are probabilistic systems, not truth machines. Pangram’s labels can include false positives, where human writing is flagged as AI, and false negatives, where AI writing passes as human. Spero told 404 Media that Pangram estimates its false positive rate at about one in 10,000 and works to reduce both kinds of errors.
Spero also said the company’s results may understate the amount of AI text seen by typical users. His reasoning: someone who installs an AI-detection browser extension may be more likely than an average user to block, mute, or avoid accounts posting obvious machine-generated material.
Platforms are already reacting, unevenly
The data puts pressure on platforms that have both encouraged AI-assisted posting and claimed to value human conversation. LinkedIn previously built AI writing tools into its posting flow. In May, the company said it was trying to reduce generic AI content and removed the writing assistant from the post button.
A LinkedIn spokesperson told 404 Media that professionals use the service to hear from real people and said the company works to reduce low-quality, automated, or generic content. The spokesperson said AI may help users get past a blank page, but LinkedIn’s focus is on surfacing professional conversations that help careers.
Reddit has taken a different public line. The company published a blog post Monday saying spam, bot activity, and inauthentic content are major concerns in the AI era. It has also run an ad campaign built around the message that its users are human. A Reddit spokesperson pointed 404 Media to that blog post when asked for comment.
Substack and X did not respond to requests for comment, according to 404 Media.
Pangram’s study does not show who wrote any individual post, and it should not be treated as a perfect census of platform content. It does show that, among the users in its sample, AI-labeled longform writing was not confined to spam sites or search-engine bait. It was sitting in mainstream feeds, especially where platforms reward long, polished, frictionless posting.
This story draws on original reporting from 404 Media.
