Over a Third of Post-ChatGPT Webpages Show Signs of AI Authorship, Pew Study Finds

August 20, 2026

More than one-third of webpages published following the debut of ChatGPT show signs of being written or substantially edited by artificial intelligence, according to findings from Pew Research. The analysis offers a detailed look at how machine-generated content is spreading across the internet and how AI presence differs across domain types.

What Happened

Pew Research analyzed nearly half a million English-language webpages from the Common Crawl web archive, covering a period of about five years beginning prior to ChatGPT’s November 2022 release. To identify machine-generated text, the researchers utilized detection technology provided by Open Pangram.

When assessing a random sample of 10,000 webpages gathered in July 2026, researchers found that approximately 10% showed significant signs of AI authorship. Because that broad sample included older pages published before modern AI writing tools existed, Pew isolated pages published specifically after ChatGPT was released. Among those newer pages, 35%—more than one in three—displayed evidence of AI authorship or heavy AI editing.

Key Highlights

Over 35% of webpages published after the release of ChatGPT show indications of AI authorship or substantial AI editing.
Commercial domains (.com) showed signs of AI writing at roughly 10 times the rate of government (.gov) and educational (.edu) domains, both of which registered around 1%.
Websites using the .org domain showed an AI authorship rate of 4.6%.
Writing patterns often linked with AI text, such as the use of em dashes, Oxford commas, and phrases like ‘it is not X, it is Y’, have increased over time.
The report follows data from Cloudflare showing that automated bot traffic has outpaced human web traffic earlier than previously projected.

Why This Matters

While previous reports from network infrastructure providers focused on automated traffic browsing the web, Pew Research examined the material being browsed, pointing toward an environment where bots frequently consume text produced by other bots. The study also noted that while detection tools like Open Pangram can occasionally misclassify human writing, the large-scale findings remain directionally informative regarding the growth of AI-generated content online.

What to Watch Next

The continued tracking of automated content relies on ongoing assessments of detection technologies and their ability to differentiate between human and synthetic text across diverse domain categories.

Frequently Asked Questions

What percentage of post-ChatGPT webpages show signs of AI writing?

According to Pew Research, 35% of webpages published after ChatGPT’s launch show significant signs of being written or heavily edited by artificial intelligence.

Which domain extensions have the highest and lowest rates of AI content?

Commercial .com domains showed AI authorship signs at approximately 10 times the rate of .edu and .gov domains, which both stood at around 1%. The .org domains showed an AI authorship rate of 4.6%.

How was the Pew Research study conducted?

Researchers examined nearly 500,000 English-language pages from the Common Crawl archive and applied Open Pangram detection software to evaluate the likelihood of AI authorship.

Source: Pew Research, reported via TechCrunch.