Artificial intelligence is no longer just changing how people search, work and communicate. It is increasingly changing the internet itself.
A new study from Pew Research Center has found that more than a third of web pages published since the launch of ChatGPT show significant signs of being written or substantially edited by artificial intelligence.
The research analysed almost half a million English-language web pages collected from the Common Crawl web archive between 2021 and 2026. Researchers then used an AI detection system to examine how much of that content showed signs commonly associated with AI authorship.
In a random sample of web pages collected in July 2026, around 10% showed significant signs of AI authorship.
However, when researchers removed older pages published before ChatGPT became publicly available in late 2022, the figure rose sharply. More than one-third of pages published during the AI era showed signs of being written or heavily edited by artificial intelligence.
The findings offer one of the clearest pictures yet of how quickly generative AI has become part of online publishing.
AI writing tools are now widely available to businesses, marketers, students, bloggers and media organisations. They can generate articles, product descriptions, marketing copy and other written content within seconds.
That convenience has created an enormous increase in the amount of content being published online.
But the growth also raises questions about what the future internet will look like.
If AI systems are increasingly trained on information found online, a growing volume of AI-generated content could eventually become part of the material used to train future AI models. Critics have warned that this could make it more difficult to separate original human reporting from recycled or machine-generated information.
The Pew study also found differences across the internet.
Pages on commercial .com domains showed much higher signs of AI authorship than content on .edu and .gov websites. Educational and government domains had significantly lower rates, although the study noted that AI detection tools cannot determine authorship with complete certainty.
That limitation remains important.
AI detectors can make mistakes, and certain writing styles may appear similar to patterns associated with AI. The study therefore does not prove that every page identified by the system was entirely written by a machine.
Still, the overall trend is difficult to ignore.
Less than four years after ChatGPT introduced generative AI to millions of people, AI-assisted writing has become a visible part of the modern web.
The question is no longer whether artificial intelligence will influence online content.
It already is.
The bigger question may be whether readers, publishers and search platforms will be able to maintain trust in an internet where humans and machines are increasingly writing side by side.
