An opt-in scan of a million posts people actually saw puts AI writing across mainstream platforms, and it hits longform hardest.

An opt-in scan of roughly a million posts that people organically scrolled past online found 41 percent of longform LinkedIn posts flagged as fully AI-generated. Across five mainstream platforms, about one in four posts longer than 250 words was judged entirely machine-written.

The data came from Pangram, which builds an AI-detection Chrome extension. Users who had it installed volunteered their browsing data, and the company analyzed around a million posts they came across naturally on LinkedIn, Medium, X, Reddit and Substack over a two-month stretch. Because the tool works passively while people browse, it recorded only what its own users actually saw rather than a raw sample of the internet, something CEO Max Spero told 404 Media had not really been measured before.

Longer posts were more likely to be AI-written on every platform. Pangram treated posts of 50 to 250 words as shortform and anything longer as longform. By that measure, Medium's longform ran 31 percent fully AI and X's 29 percent; on X, close to half of longform articles were either fully AI-generated or AI-assisted once both are counted together. Substack sat lowest, at around 10 percent.

LinkedIn supplied only about a third of the posts examined, yet it accounted for close to two-thirds of the AI-generated content the scan turned up. The platform had spent years building AI writing tools directly into its product, which made such posts easy to produce and let them spread widely.

Pangram read a counterintuitive pattern into the numbers: people appear readier to let AI speak for them under their real, identifiable professional persona than in casual, anonymous settings. In its own summary, the company said AI-generated material is a problem on every platform it studied but falls disproportionately on longform.

LinkedIn, for its part, told 404 Media that its members come seeking authentic voices and real perspectives from actual people, that it takes active steps to cut down on generic, automated and low-quality content, and that its focus is on professional discussions that help people advance their careers. In May it announced a push to curb AI content and stripped the built-in writing assistant out of its post button.

The figures come from a company with a commercial interest in AI content being seen as a problem, and Spero told 404 Media they are likely a lower bound, since people who install an AI-detection extension probably care more than most about noticing and blocking machine-written posts. Even so, the results, from the maker of the leading AI-text detector, were described as both striking and plausible.