GoKawiilTwo months ago, we launched our Chrome extension to help combat the rising slop problem on social media. It lets users scan posts on social media as they scroll, flagging AI-generated content so they can make informed decisions about how they spend their attention.
Pangram is a research-first company, not just for our industry-leading AI detection algorithms, but for tracking the risk and prevalence of AI-generated content. Social media is one of the hardest domains to study here — much harder than, say, news articles, research papers, or Amazon reviews. But it's also one of the most crucial, because it's potentially the highest-volume source of AI-generated content we face.
We believe it's important to understand this problem so that we can better combat it. That's why we included an opt-in setting in our Chrome extension, to allow users to aid our research by anonymously sharing their scan statistics with us. We've compiled the first few months of this data into the report below.
Key findings:
One in four longform posts on social media flagged as fully AI-generated
Of all posts we flagged as AI-generated, two-thirds came from LinkedIn
Nearly half of X/Twitter articles contained AI writing
AI-generated content is hitting longform the hardest
AI-generated content appeared across all social media platforms in our data set. The average AI rate across all scanned items was 13.8%, but specific rates varied by platform and item length. On four out of five platforms, longer content was more likely to be AI-generated than shortform content. Across all platforms, one in four longform items (25.72% of items over 250 words) were fully AI-generated.
Substack was an exception; there, the rate of fully AI-generated content remained fairly flat, and longer, more substantial posts were actually slightly less likely to be AI-generated compared to shorter ones.
... continue reading