AI-generated posts now make up roughly one in four long-form updates on social media, and LinkedIn is the biggest offender, according to a new analysis from detection company Pangram. About 41% of the platform's posts longer than 250 words were flagged as AI-written, the highest rate of any network studied.
The scale of LinkedIn's problem stands out. The professional network accounted for only about a third of all posts scanned, yet it produced close to two-thirds of every piece of AI content Pangram detected. In other words, machine-written text is concentrated where people go to talk shop. For more on how detection and generation are evolving, see our artificial intelligence coverage.
What the Pangram study found
Pangram pulled its data from a Chrome extension that scanned more than one million posts across five platforms between April and June 2026. The company sorted results into short-form (50 to 250 words) and long-form (over 250 words) buckets, and the pattern was consistent: the longer the post, the more likely a model wrote it.
Here's how the five platforms ranked on AI-written content, short-form and long-form:
- LinkedIn: 30% short-form, 41% long-form
- Medium: 28% short-form, 31% long-form
- X/Twitter: 9% short-form, 29% long-form
- Reddit: 3% short-form, 13% long-form
- Substack: 12% short-form, 10% long-form
On X, close to half of long-form articles were either fully AI-generated or AI-assisted. Substack posted the lowest long-form rate at around 10%. Reddit was a split story: replies were 98% human-written, but standalone posts contained AI text far more often.
The short-versus-long split is telling. On nearly every platform, longer posts carried a higher AI rate than short ones, which suggests people are most tempted to hand the writing to a model when a post demands real length. Substack was the lone exception, its long-form rate (10%) actually dipping just below its short-form figure (12%). The Reddit contrast points the same way: quick, conversational replies stay human, while the effort of an original post is where automation creeps in.
How reliable are the numbers?
Pangram says its Pangram 3 model carries a false-positive rate of 0.01%, meaning it rarely tags human writing as machine-made. The company also concedes the detector is likely better at confirming human text than at catching every AI passage, so the true share of AI content could be higher than the headline figures, not lower.
One important caveat: the study measures origin, not quality. A post being AI-assisted doesn't automatically make it spam, and a human-written one isn't guaranteed to be good. Still, the concentration on professional feeds helps explain why "AI slop" has become shorthand for low-effort, auto-generated updates, a trend we've tracked across recent tech news.
Why it matters
The findings land as platforms wrestle with how much automated writing their feeds can absorb before users tune out. LinkedIn itself appears to feel the pressure and has already started cracking down on AI-generated posts. Expect more networks to weigh similar limits as detection tools sharpen, a shift with real stakes for anyone building a professional brand online.
For readers, the takeaway is a healthy dose of skepticism. If a quarter of long posts in a feed are machine-made — and potentially more — the value of a genuinely human perspective only goes up.
Which platform has the most AI content?
LinkedIn, by a wide margin. It led both categories Pangram measured and supplied the majority of all AI text found across the five networks.
Can AI writing be detected reliably?
Detectors like Pangram 3 report very low false-positive rates, but no tool is perfect. They are generally more confident identifying human writing than flagging every AI-assisted sentence, so treat published percentages as a floor rather than a ceiling.
