The Scale of AI-Generated Content

Recent data from Pew Research indicates a significant shift in web content creation. By analyzing nearly half a million English-language webpages using Open Pangram’s detection technology, researchers found that approximately 35% of content published after the launch of ChatGPT (November 2022) exhibits significant signs of AI authorship or heavy AI editing. When looking at the broader web (including pre-2022 content), the prevalence drops to roughly 10%, highlighting that the surge in AI-generated text is a recent and rapid phenomenon.

Domain-Specific Adoption Patterns

The study identifies a clear divide in how different types of domains utilize AI for content production. Commercial (.com) domains are the primary drivers of this trend, showing AI authorship rates approximately 10 times higher than those of educational (.edu) or government (.gov) domains, which hover around 1%. Non-profit (.org) domains also show relatively low adoption, with a 4.6% rate of AI authorship. This suggests that AI-driven content is heavily concentrated in sectors focused on traffic generation, SEO, and commercial engagement rather than institutional or academic publishing.

Detection and Stylistic Markers

While AI detection tools like Pangram are not infallible and may produce misclassifications, the study provides a directional view of the changing web landscape. Beyond direct detection, researchers noted an increase in specific stylistic markers often associated with LLM output, such as the frequent use of em dashes, Oxford commas, and repetitive rhetorical structures like "it’s not X, it’s Y." These findings align with broader infrastructure reports, such as those from Cloudflare, confirming that the internet is increasingly becoming a space where automated systems generate content primarily for other automated systems to consume.