Pew Research: 35% of new web pages post-ChatGPT show signs of AI

A new study on the origins of text published on the internet has revealed that digital publishing is changing rapidly due to artificial intelligence tools. According to an analysis by the Pew Research Center, traces indicating AI generation were detected in 35% of English-language web pages published in the period following the launch of ChatGPT.

In the study, nearly half a million English-language pages from the last five years were examined using the Common Crawl archive. The texts were scanned with the Open Pangram system to determine the probability of AI generation. Across the entire sample set dated July 2026, the proportion of pages with clear AI traces was measured at 10%.

However, when the period before the widespread adoption of generative AI tools was excluded and only the post-ChatGPT process was evaluated, the rate was much higher. In this calculation, it was stated that more than one-third of new web pages contained an AI signature.

The study also points out that AI usage varies significantly depending on the type of site. While AI traces are more frequently encountered on commercial sites with ".com" extensions, the rate remained at approximately 1% on official or education-focused sites with ".gov" and ".edu" extensions. On non-governmental organization sites with ".org" extensions, this rate was recorded at 4.6%.

Researchers emphasize that AI detectors carry a margin of error, albeit a limited one. Nevertheless, repetitive phrasing patterns, punctuation preferences, and specific writing habits in the texts show that the impact of generative AI on web content is becoming increasingly visible.