A new study from the Pew Research Center has revealed a significant shift in the composition of the internet. According to the report released on Thursday, more than one-third of all English-language web pages published after the launch of ChatGPT show signs of being written or substantially edited by artificial intelligence.
The findings highlight the speed at which generative AI tools have been integrated into digital content creation. This development follows other recent industry data suggesting that the web is increasingly becoming an ecosystem where automated systems both produce and consume information. The study raises questions about the future of digital authenticity and the potential for a self-referential loop in which AI models are trained on data previously generated by other machines.
A Rapid Shift in Web Composition
To understand the prevalence of AI-generated content, Pew Research analyzed a massive dataset from the Common Crawl web archive. This archive is a well-known repository that provides a snapshot of the public web for researchers and developers. Pew collected nearly half a million English-language pages spanning the last five years, a period that includes several years of activity prior to the public release of ChatGPT in November 2022.
The research team used detection technology from Open Pangram to evaluate the likelihood of AI involvement in the writing or editing of these pages. In a random sample of 10,000 pages collected in July 2026, roughly 10 percent of the content showed what the researchers called significant signs of AI authorship.
However, the 10 percent figure includes many pages created before high-quality generative AI tools were available to the public. When the researchers filtered the data to focus specifically on pages published after the launch of ChatGPT, the results were much more pronounced. In that specific subset, 35 percent of the pages were identified as having been written or heavily modified by AI.
Variations Across Domain Types
The study found that the presence of AI-authored content is not uniform across the internet. There are stark differences depending on the type of domain hosting the information.
The highest concentration of AI-generated content was found on .com domains. These commercial URLs showed signs of AI authorship at a rate approximately 10 times higher than that of government (.gov) or educational (.edu) domains. Both .gov and .edu sites remained largely human-authored, with only about 1 percent of their pages showing AI influence.
Organizations using the .org domain occupied a middle ground. The study found that 4.6 percent of pages on .org sites showed signs of AI authorship. The high prevalence on .com domains suggests that commercial entities, such as marketing blogs, news aggregators, and product-focused websites, are the primary drivers of AI content adoption.
Identifying the Linguistic Markers of AI
While AI models are designed to mimic human speech, they often leave behind statistical "fingerprints" that detection tools can identify. Pew found that certain linguistic habits and punctuation patterns have become more frequent as AI usage has grown.
Key indicators identified in the study include:
- A higher frequency of Oxford commas.
- Increased use of specific punctuation patterns, such as long dashes for sentence structure.
- The recurrence of specific phrasing templates, such as "it is not X, it is Y."
- A formal or standardized tone that lacks the idiosyncratic variations typical of human writers.
The researchers acknowledged that AI detection is not an exact science. Tools like Open Pangram can occasionally misclassify human writing as machine-generated, or fail to catch sophisticated AI content. However, Pew maintains that at the scale of hundreds of thousands of pages, the data provides a clear directional look at how the web is changing.
The Problem of a Bot-to-Bot Internet
The increase in AI-generated content coincides with a shift in who is actually using the web. Recently, the internet infrastructure company Cloudflare reported that bot traffic has officially overtaken human traffic. This means that a majority of the activity on the internet, from data scraping to automated searches, is now performed by software rather than people.
The Pew data adds a new layer to this development. It suggests that not only is the web being navigated by bots, but the content they are finding is also increasingly being created by bots. This creates a potential "dead internet" scenario where the human element of the digital world is increasingly marginalized.
For the technology industry, this trend presents a serious technical challenge known as model collapse. If AI developers continue to scrape the web for training data, their future models will inevitably be trained on the output of current models. Researchers have warned that this feedback loop can lead to a degradation of quality, as errors and biases in one generation of AI are magnified in the next.
What the Results Show
The data from Pew Research corroborates other independent studies that have tracked the rise of automated content. The findings suggest that AI is no longer a niche tool used by a few early adopters; it has become a fundamental component of how the commercial web is built.
The results also point to a growing divide in information reliability. While commercial spaces are being flooded with automated content, institutional domains like those managed by universities and government agencies appear to be maintaining stricter standards for human authorship. This could lead to a future where users must be increasingly selective about the domains they trust for factual information.
What Happens Next
As AI authorship continues to grow, the industry is looking for ways to maintain transparency. Companies like Anthropic have begun experimenting with digital watermarks that would identify AI-generated text, but such measures are difficult to enforce across the entire open web.
Search engines are also in a difficult position. Companies like Google have had to adjust their algorithms to prioritize "helpful" content over mass-produced AI text that is often designed specifically to game search rankings.
What is clear from the Pew study is that the transition is happening faster than many expected. In the few years since ChatGPT became a household name, more than a third of the new digital landscape has been influenced by artificial intelligence. The challenge for the coming years will be determining how to preserve the value of human-generated information in an increasingly automated world.
Filed under: AI, TechNews, Software, DataScience, WebDevelopment, MachineLearning