Pew Study Finds AI Wrote 35% of Post-ChatGPT Web Pages

A newly released study by Pew Research indicates that artificial intelligence is rapidly rewriting the landscape of the internet. By analyzing nearly half a million English-language web pages archived over the last five years via Common Crawl, researchers found that 35% of pages published after the November 2022 launch of ChatGPT show strong indications of AI generation or heavy editing. To identify these patterns, Pew utilized detection technology developed by Open Pangram.

The study examined a random sample of 10,000 web pages gathered in July 2026. Within this broad sample, which included older pages created before generative AI tools existed, approximately 10% exhibited strong indicators of machine authorship. However, when the researchers isolated only the pages published after ChatGPT's debut, that figure jumped to more than one-third. The prevalence of machine-generated text varied wildly by domain type. Commercial sites with .com addresses displayed signs of AI writing at ten times the rate of educational (.edu) or government (.gov) sites, both of which hovered around 1%. Meanwhile, non-profit .org domains showed AI authorship on 4.6% of their pages.

Pew acknowledged that AI-detection software like Pangram is not flawless and can occasionally misclassify human writing. Nevertheless, the researchers noted distinct stylistic shifts that align with generative models, such as an increased use of Oxford commas, em dashes, and specific sentence structures like "it's not X, it's Y." This surge in AI-generated content coincides with recent data from Cloudflare showing that automated bot traffic has officially surpassed human traffic on the web, signaling a shift toward an internet increasingly populated and read by machines.