Look, we all felt it. You scroll through LinkedIn and something feels profoundly off — the humblebrags are too structured, the ‘lessons learned’ too perfectly symmetrical. Now there are numbers to back up the uncanny valley. A new analysis from Pangram scanned over a million posts across five platforms and found that one in four social media posts longer than 250 words is AI-generated. That’s the baseline. On LinkedIn, the figure rockets to 41 percent.
What makes this particularly damning is the concentration. LinkedIn represented only about a third of all the posts Pangram’s Chrome extension scanned between April and June, yet the platform accounted for nearly two-thirds of all detected AI content. That means if you’re reading a lengthy professional hot-take, there’s a solid chance a bot wrote it. X wasn’t far behind, with close to half of its long-form articles flagged as AI-generated or AI-assisted — a stat that tracks if you’ve ever encountered those numbered, thread-style posts that read like a content marketing white paper had a fever dream.
Pangram claims its detection model, Pangram 3, has a false positive rate of just 0.01 percent, but the company is honest about the limitations. The tool is likely better at confirming human writing than definitively catching every AI-generated piece, which means the real numbers could be even higher. Reddit replies were a bright spot — 98 percent human — though standalone posts told a different, more synthetic story. Substack held the line too, with only about 10 percent of long-form posts flagged.
The study didn’t weigh in on quality, and that’s probably the right call. AI prose isn’t inherently worthless, but volume has a way of drowning out signal. LinkedIn seems to agree. The company has already started cracking down on AI-generated content, perhaps realizing that a feed full of synthetic thought leadership isn’t exactly the professional utopia anyone signed up for.
