Pew Research: Over One-Third of New Web Pages Created with AI

Research by the Pew Research Center indicates that around 10% of all web pages on the internet exhibit signs of being created or significantly edited by artificial intelligence. Notably, among content published after the release of ChatGPT, this figure rises to over one-third.

https://twitter.com/pewresearch/status/2090862804420931804

AI Gradually Fills the Internet

The researchers conducted an analysis of a random sample of 10,000 English-language web pages collected in July 2026 via Common Crawl. They employed a machine learning model to identify linguistic features typical of generative AI in the texts.

Approximately 10% of all pages in the sample showed significant signs of AI authorship. While this may seem relatively small at first glance, the sample included a considerable amount of older content, created prior to the emergence of current generative models.

Source: Pew Research Center.

However, when excluding older pages and focusing solely on content published after the launch of ChatGPT in November 2022, the scenario shifts dramatically: signs of AI authorship are found on more than one-third of these pages.

This indicates a growing prevalence of synthetic content alongside the rise of ChatGPT and its competitors, such as Claude, Gemini, and other models.

AI Content Predominantly on Commercial Websites

The distribution of AI-generated content across domain zones is uneven.

In 2026, signs of AI generation were found on about 10% of pages within the .com domain, which is roughly double the rate of 4.6% in the .org domain.

On .edu and .gov sites, the proportion was approximately ten times lower, around 1%. Researchers note that at the onset of ChatGPT's popularity, the differences among the main domain zones were much less pronounced.

This discrepancy may reflect the nature of the content itself. Commercial sites tend to publish larger volumes of textual material that can be relatively easily automated, including product descriptions, informational articles, SEO content, and news updates.

Study Does Not Claim Every Such Text is AI-Written

An important caveat regarding the methodology is that Pew did not verify the creation history of each page or ask authors whether they used ChatGPT or other tools.

Instead, researchers analyzed linguistic features statistically associated with AI generation. Therefore, the findings should be viewed as an estimation of probable authorship rather than an exact count of pages authored by neural networks.

Moreover, the study specifically addresses content that is "written or substantially edited" by AI. Minor text adjustments made using neural networks may not necessarily fall into this category.

It is worth noting that in August, Anthropic introduced global labeling for content generated by Claude models.