One-third of new web pages show AI authorship signals, Pew study reveals

A major finding from Pew Research published this week suggests that the internet's publishing landscape has shifted dramatically toward machine-generated content. The study examined nearly half a million English-language web pages and found that when focusing only on material published after ChatGPT's November 2022 launch, approximately 35% displayed measurable signs of AI authorship or substantial AI editing.

How the research was conducted#

Researchers built their analysis on data from the Common Crawl web archive, then applied Open Pangram's detection technology to identify pages likely shaped by AI tools. Their initial random sample of 10,000 pages yielded roughly 10% showing clear AI signals—but this figure included pre-ChatGPT content that couldn't have been machine-written. Once they filtered to only post-launch pages, the proportion jumped sharply.

Patterns across web properties#

The study revealed striking disparities based on domain type. Commercial websites (.com) displayed AI authorship signals at dramatically higher rates—roughly 10 times greater than educational (.edu) or government (.gov) sites, each hovering near 1%. Nonprofit domains (.org) fell somewhere between, at 4.6%.

What constitutes AI signals#

Pew's methodology has limitations—AI-detection tools can produce false positives—yet the scale of the finding suggests the directional trend holds. Researchers also noted that certain stylistic markers had proliferated alongside suspected AI adoption: frequent use of em dashes, Oxford commas, and formulaic phrases like "it's not X, it's Y."

The study arrives as internet traffic itself is shifting: Cloudflare recently reported that bot traffic now exceeds human browsing, raising questions about who is reading an increasingly machine-written internet.

Read the full Pew study: How much of the internet is written with AI?