Q TECH — We still read the internet with an assumption inherited from another era: that behind every page there is a person who wrote it. That assumption has just been refuted by a measurement.
The Pew Research Center, the leading research center for digital trends, analyzed 490,000 English-language web pages from the last five years, taken from the Common Crawl archive, and ran them through Open Pangram, a tool for detecting automated writing.
The result: in the sample from July 2026, 9.6% of all pages show strong signs of having been written or extensively edited by AI.
The figure seems modest until you read the fine print. The internet has accumulated decades of material that no machine could have written. When Pew filters pages prior to November 2022, the month OpenAI launched ChatGPT, the number shifts dramatically: 35% of pages published since then show AI authorship or editing.
More than a third of the new web no longer comes from human hands, or not entirely.
The question was never whether AI would write the internet, but when we would stop noticing it. Pew’s data says that moment has already passed.
Artificial text is not evenly distributed. In the “.com” domains, the commercial ones, 9.4% of pages show signs of AI. In “.org” domains, half that figure is 4.6%. In academic (.edu) and government (.gov) domains, it’s barely 1% and 0.8%, respectively. When ChatGPT appeared, the four domains were at almost identical levels, all below 1.1%.
This gap is the most revealing finding of the study. Where there is an economic incentive to produce cheap volume, the machine writes. Where institutional reputation is at stake, the human resists. The adoption of AI in writing did not follow the technology: it followed the business model.
The machine has already changed how the internet sounds
Pew also measured the linguistic footprint. Compared to 2023, the use of the em dash almost doubled: it went from 5.79 to 11.19 occurrences per 10,000 words.
The Oxford comma, the one that precedes the final conjunction in a list, increased by 63%. Models’ favorite words, like “delve” and “testament,” more than doubled their presence. And the “it’s not just X, it’s Y” structure nearly tripled.
The internet doesn’t just have more typed text: it sounds increasingly like a machine. Models’ style, honed on journalistic and academic prose, has become the dominant accent of the new web.
Artificial text abounds, but it doesn’t govern what we read
A previous study by the web positioning consultancy Graphite, confirmed by Axios, estimated that AI-generated articles would surpass 50% of new publications by the end of 2024. The same study provides a counterpoint: 86% of the articles appearing in Google search results are still human-authored, and chatbots cite human sources in 82% of cases.
Volume is not visibility. The tide of synthetic text exists, but the systems that organize the web still filter it out. Pew rightly warns that no detector is infallible on individual documents: the value of these figures lies in the aggregate, not in judging a single text.
The key takeaway is this: half of American adults already use chatbots, and 24% use them daily.
Each of these uses leaves a residue on the web. Human-generated text published online has begun to behave like a scarce asset: increasingly rare, increasingly valuable to search engines, to readers, and to the very machines that need fresh human prose to train themselves.
The web isn’t dying. It’s rewriting itself, and those who still write by hand have, for the first time in two decades, a measurable advantage.