A new Pew Research Center report found that about 10 percent of English-language webpages sampled in July 2026 showed significant signs of having been written or substantially edited by artificial intelligence. When researchers looked only at webpages published since ChatGPT became publicly available in November 2022, the proportion climbed to roughly 35 percent, or more than one in three.

Pew reported separately in June that 49 percent of American adults now use AI chatbots such as ChatGPT, Gemini or Copilot, compared with 33 percent in 2024, while 24 percent say they use chatbots daily. The new research suggests that this growth is no longer merely changing how people obtain information. AI is increasingly involved in creating the information they encounter.

Generative AI has made producing articles, marketing pages, product descriptions, and other online material extraordinarily fast and inexpensive. That could make useful information easier to create and translate, but it could also flood search engines with repetitive, inaccurate, or low-quality material. As AI systems themselves increasingly learn from information found online, future models could be trained partly on content generated by earlier models, making verification of original human sources increasingly difficult.

Pew conducted the study using Common Crawl, a nonprofit archive that periodically collects enormous snapshots of the public web. Researchers examined nearly 490,000 English-language webpages, randomly sampling about 10,000 pages from each of 49 web crawls conducted between January 2021 and July 2026. This timeline gave Pew an important baseline because it included nearly two years of webpages from before ChatGPT triggered the generative-AI boom. Researchers then analyzed the text using Open Pangram, an AI-detection model that searches for statistical patterns in vocabulary, phrasing, and sentence construction associated with machine-generated writing.

The results showed a dramatic change beginning around ChatGPT’s release. In the complete July 2026 snapshot about 10 percent showed substantial indications of AI authorship. But among pages created after November 2022, the figure reached approximately 35 percent. AI writing was also distributed unevenly. Roughly one in ten pages on commercial .com domains showed AI signals, compared with about 4.6 percent of .org pages and only around 1 percent of .edu and .gov pages.

Pew did not prove that 10 percent of everything on the internet is AI-generated. The research examined English-language webpages collected by Common Crawl, not every social-media post, video, image, private website or app. AI detectors are also imperfect: They can mistakenly identify human writing as AI-generated and can fail to recognize AI text that humans have extensively rewritten. Pew therefore describes the results as pages showing ‘significant signs’ of AI authorship rather than definitively declaring them machine-written.

Pew’s study suggests an ongoing evolution of the internet. The internet was once overwhelmingly a collection of information written by humans for humans. It is becoming an ecosystem in which humans and machines jointly create what other humans and other machines read. The central challenge may soon be less about whether AI is present online than whether users can still determine who created the information they encounter, where it came from and whether they should trust it.