A third of web pages published since ChatGPT’s launch show signs of AI authorship, study finds

The landscape of the internet is undergoing a seismic shift. According to a comprehensive new report from Pew Research, more than one-third of all web pages published following the debut of ChatGPT exhibit clear indicators of artificial intelligence involvement. This data reinforces growing concerns that the modern web is increasingly becoming a digital ecosystem populated by content generated or heavily refined by machine learning models.

A Web Written by Bots, for Bots

This study arrives on the heels of a significant milestone reported by Cloudflare, which recently revealed that automated bot traffic has officially surpassed human web traffic—a shift that occurred much faster than industry analysts had anticipated. While Cloudflare’s findings focused on the consumers of web content, the Pew Research report shifts the lens to the creators, highlighting a reality where bots are increasingly writing for an audience of other bots.

To conduct this analysis, researchers utilized the Common Crawl web archive, aggregating nearly 500,000 English-language pages spanning the last five years. By employing Open Pangram’s detection technology, the team assessed the likelihood of AI authorship across this massive dataset.

Key Findings from the Study

• The Post-ChatGPT Surge: When looking at a broad sample of the web, roughly 10% of pages showed significant signs of AI influence. However, when researchers isolated content published specifically after November 2022, that figure jumped to 35%.
• Domain Disparities: The prevalence of AI-generated content varies wildly by domain type. Commercial .com sites are roughly 10 times more likely to feature AI-authored text compared to .edu or .gov domains, which hovered around a 1% rate. Meanwhile, .org domains showed a 4.6% rate of AI usage.
• Linguistic Markers: Beyond algorithmic detection, the study noted a rise in stylistic "tells" associated with AI writing, including a higher frequency of em dashes, specific phrasings like "it’s not X, it’s Y," and the consistent use of Oxford commas.

"In a random sample of 10,000 web pages collected in July 2026, around 10% showed significant signs of AI authorship. However, Pew pointed out that a random sample like this would inevitably include older web pages—ones published before AI writing tools even existed."

Navigating the Limitations of Detection

While the findings provide a compelling look at the state of the internet, the researchers acknowledge that AI-detection tools are not infallible. There is always a margin for error, with some human-written content potentially being misclassified as machine-generated. Nevertheless, at this scale, the data provides a reliable directional map of how the web is evolving.

As the barrier to entry for content creation continues to vanish, the internet is becoming a place where human-authored text is increasingly diluted by automated output. Whether this trend enhances efficiency or degrades the quality of information remains a central question for the future of the digital age.