Who Is Writing the Internet Now? What the Latest Data Shows

Less than four years after they first appeared on the scene, chatbots have left their imprint on a large portion of the web. Based on a new analysis from Pew Research Center, approximately 10% of English webpages that were sampled in July 2026 showed strong evidence of being substantially created or edited by artificial intelligence.

The finding comes from a data essay published August 20 by Pew’s Data Labs team and written by senior data scientist Samuel Bestvater. It offers one of the clearest long-range pictures yet of how quickly AI-generated text has spread online.

How the study worked

Pew used the Common Crawl web archive to compile almost half a million English-language pages (about 490,000) from snapshots taken between January 2021 and July 2026. The sample precedes the debut of ChatGPT in November 2022, giving the researchers a baseline for comparison. They then took the text of each page and ran it through an open-weight AI detection model called Open Pangram, which looks for statistically significant patterns in word-choice and sentence-structure use to distinguish human writing from computer-generated writing.

Why the Real Number Is Bigger Than It Looks

Ten percent might not seem like much, but the web is full of material that predates generative AI. When Pew limited its view to pages published after ChatGPT’s release, the picture changed sharply: more than one in three of those pages showed signs of AI authorship in the July 2026 snapshot, according to the study. Pew notes that this is in line with other research reaching similar conclusions about recently published content.

In other words, the older web dilutes the headline figure. The newer a page is, the more likely it is to carry AI’s imprint.

Commercial sites lead, institutions lag

AI-written text is not spread evenly. Pew found that in 2021 and 2022, signals of AI authorship appeared at similar low rates across .com, .org, .edu and .gov domains. By 2026 they had diverged.

.com sites: roughly one in ten pages, up from about 1% in early 2021
.org sites: 4.6%, about half the .com rate
.edu and .gov sites: around 1% each, roughly a tenth of the .com rate

Put simply, the commercial web has absorbed AI writing far faster than universities and government agencies have, at least on the measure Pew used.

The telltale habits of machine prose

Pew also looked at which stylistic habits have grown more common online. Because AI models learn from huge volumes of human writing, they tend to overuse certain features. Comparing the web of 2026 with a 2023 snapshot, the study reports:

Em dashes appear about twice as often.
Oxford commas are up 63%.
Words favored by AI models, such as “delve,” “interplay” and “testament,” have more than doubled in use.
The “it’s not just X, it’s Y” construction, which Pew calls negative parallelism, has nearly tripled, though it remains fairly rare.

Pew is careful to add that none of these habits proves a text was written by AI, since people use em dashes and Oxford commas too. The signal emerges only when looking across enormous numbers of documents, and detection models weigh many subtler patterns beyond these surface features.

What the numbers can't tell us

The study carries real caveats. Pew acknowledges that AI detectors are imperfect and can mislabel individual human-written pages as AI-generated, and the reverse. The analysis was also limited to English-language pages, and “written or substantially edited” by AI covers a range of possibilities from fully automated text to a human draft polished by a chatbot. The study looks at the proportion of AI-style text, not whether it is accurate, useful or trustworthy.

What This Means for Readers and Publishers

The trend lines up with changing habits. Pew’s June 2026 survey on Americans and AI found that around half of U.S. adults use AI chatbots, and 24% use them daily. As that use grows, the text flowing onto the web grows with it.

For publishers, educators, search engines and everyday readers, the takeaway is that the web is a mixed environment, and the mix is shifting toward machine-assisted writing. Institutional sites have held steady for now, while commercial pages are changing fast.

This article is based on “How Much of the Internet Is Written With AI?” by Pew Research Center (Aug. 20, 2026), by Samuel Bestvater, Aaron Smith, Carson TerBush, Chris Baronavski and Janakee Chavda. Read the full study and methodology at pewresearch.org.