Pew Research: More Than 30% of Web Pages Show Signs of AI-Generated Content Following ChatGPT’s Release.

A new study released by the Pew Research Center on August 20 suggests that generative AI is reshaping the web at a rapid pace. According to the report, more than one-third of webpages published after ChatGPT’s launch in November 2022 show signs of being written or heavily revised with AI assistance.

The findings point to a noticeable shift in how online content is being created. Rather than remaining a niche tool, generative AI now appears to be influencing a substantial share of newer web pages, signaling broader changes across the internet’s content ecosystem.

How the research was conducted

Pew based its analysis on the Common Crawl web archive, examining roughly 500,000 English-language webpages collected over nearly five years. To identify potential AI involvement, the researchers used Open Pangram technology to detect patterns associated with machine-generated writing.

In a random sample of 10,000 webpages taken in July 2026, about 10% showed clear signs of AI-generated content. But that overall figure included many older pages published before generative AI tools became widely available. After removing those earlier pages, Pew found a much sharper pattern: among webpages created after ChatGPT became public, around 35% carried indications of AI writing.

Different parts of the web show different levels of AI use

The study also found significant variation depending on the type of website.

• .com sites showed AI-writing signals at a far higher rate—roughly ten times the level seen on .edu and .gov domains.

• .edu and .gov sites were at the low end, with only about 1% showing such signs.

• .org sites fell in between, at approximately 4.6%.

These differences suggest that commercial websites may be adopting AI-assisted publishing much faster than educational or government platforms.

The linguistic fingerprints of AI are becoming more common

Beyond direct detection, the report noted that several writing traits often associated with AI-generated text are appearing more frequently online. Among them are heavier use of em dashes, more frequent Oxford commas, and formulaic constructions such as “not X, but Y.”

While such features do not prove a page was created by AI, their growing presence adds to the broader picture of changing writing patterns across the internet.

Researchers caution that detection is not perfect

Pew emphasized that AI-detection tools can still produce false positives and false negatives. That means the numbers should not be treated as exact measurements of how much web content is AI-made. Even so, the overall direction of the trend remains meaningful: AI-assisted writing is becoming harder to ignore in newer online content.

A web increasingly shaped by machines

The study arrives at a time when concerns about the internet’s evolving structure are growing. Cloudflare has previously reported that bot traffic has surpassed human traffic online. Put together with Pew’s findings, that raises the possibility of a web where AI is not only helping produce content, but also increasingly involved in consuming it.

As AI-generated material spreads and machine-driven traffic continues to rise, questions around authenticity, sourcing, and trust are becoming more urgent. For readers, publishers, and platforms alike, identifying where information comes from may become just as important as the information itself.