LinkedIn longform is 41% AI-written, Pangram study finds

A new study of what people actually see when they scroll their feeds puts hard numbers on something everyone has been complaining about. According to research from Pangram, an AI detection company, reported by 404 Media, 41% of longform posts on LinkedIn now read as fully AI-generated, with X close behind at 25% fully AI-written and another 23% flagged as AI-assisted. The methodology matters here: Pangram used a Chrome extension to sample roughly one million posts over two months across LinkedIn, X, Reddit, Substack, and Medium, so the numbers describe content users are actually being served rather than a raw universe of what has been posted somewhere on the web.

The platform gap is the interesting part. Reddit and Substack both come in around 10% for longform, roughly a quarter of LinkedIn's rate. LinkedIn also built AI writing tools directly into its posting interface, which lowered the friction to zero, and the reporting notes LinkedIn has since adjusted its AI writing assistant placement. Reddit, meanwhile, launched a campaign emphasizing human users. X and Substack declined to comment.

Why this matters if you spend money or attention on those feeds: the longform LinkedIn post has for years been a reasonably reliable signal of a human thinking out loud about their industry, and by extension a plausible place to hunt for hires, clients, or expertise. If close to half of it is machine drafts, scanning that feed for real signal gets more expensive, and any tool that ranks or summarizes those posts is now optimizing on top of a lot of synthetic input. Pangram CEO Max Spero calling it 'a tax on readers' time' is the honest version of what marketers usually frame as engagement.

The caveat the reporting itself flags is that Spero calls the figures a 'lower bound', because the people who install an AI-detection extension are not a random sample of internet users. Take the specifics as reported, not settled. What the piece does not give you is a false-positive rate on Pangram's classifier, an industry-vertical breakdown, or any measure of whether platform algorithms are actively amplifying the AI content or just carrying it.

The forward-looking bet is that verified-human tiers, editorially voiced newsletters, and detection vendors all get more valuable from here.