Overview
- Pangram published a report on July 9 drawing on an opt‑in Chrome extension that scanned 1,002,627 posts seen by users between April 24 and early July to measure AI writing in feeds.
- Across all scanned items the company flagged an average AI rate of 13.8% and found 25.7% of longform posts (over 250 words) were fully AI‑generated.
- LinkedIn accounted for about 62% of flagged AI content and had over 40% of longform posts marked as fully AI, while X’s article format contained roughly 23.9% fully AI and 22.9% AI‑assisted pieces.
- The study relies on Pangram’s detector (Pangram 3.3), which the company says has a 0.01% false‑positive rate, but the sample is opt‑in, only includes items over 50 words, and Pangram acknowledges detection limits and possible false negatives.
- Platforms have offered mixed responses: LinkedIn rebranded its writing helper and said it will downrank AI posts, Reddit reiterated anti‑AI moderation, and several services gave limited or no comment, leaving open how feeds and enforcement will change.