Study Shows That One-Third of Web Pages Created Since ChatGPT’s Launch Display AI Authorship Indicators

A new study from Pew Research, published on Thursday, indicates that over one-third of web pages created after ChatGPT’s launch display traits of AI authorship. This finding corresponds with other research highlighting the substantial impact of AI on the generation and modification of contemporary online content.

The report follows remarks from Cloudflare, an internet infrastructure provider, which observed that bot-driven web traffic has exceeded human traffic sooner than anticipated.

Pew’s analysis centers on the content rather than the users navigating the web, revealing that a significant amount of this content appears to be bots consuming web pages written by other bots.

For this study, Pew analyzed data from the Common Crawl web archive, obtaining nearly 500,000 English-language web pages from the previous five years, starting a couple of years before ChatGPT’s introduction in November 2022. They then utilized Open Pangram’s technology to determine the proportion of pages likely authored or heavily revised by AI.

Out of a random sample of 10,000 web pages gathered in July 2026, about 10% showed “prominent signs of AI authorship,” according to Pew.

However, Pew acknowledged that such random samples inevitably encompass older web pages—those created before the emergence of AI writing technologies, and thus, not AI-generated.

To achieve a clearer understanding of AI’s effect on web content, Pew excluded older web pages and focused exclusively on those created after the ChatGPT launch.

In this refined examination, signs of AI authorship were found in more than one-third (35%) of the scrutinized pages.

Upon analyzing specific domains, Pew found that .com websites showed signs of AI authorship at nearly ten times the rate of .edu or .gov sites (both hovering around 1% AI-authored). Additionally, .org domains had a mere 4.6% rate of AI authorship.

It’s essential to remember that this analysis is not infallible. Pangram, like other AI detection tools, may misclassify some pages as AI-generated when they are not. Nevertheless, overall, the findings should offer a fairly accurate direction.

Furthermore, Pew observed that other markers of AI authorship, such as the use of em dashes, Oxford commas, and structures like “it’s not X, it’s Y,” have also increased over the years.