AI writing now measurable across one-third of new webpages
More than one-third of webpages published after ChatGPT became publicly available in November 2022 show signs of AI involvement, according to an analysis of nearly 490,000 English-language pages collected through the Common Crawl web archive. The finding, from research by Open Pangram, tracks how generative AI tools have changed the composition of the public web.
The shift matters for writers and technology teams alike. As AI-assisted material becomes a larger share of online content, the information feeding search engines, enterprise AI systems, and automated agents increasingly mixes human and machine-produced text.
Adoption has climbed steadily since ChatGPT's release
In the July 2026 sample, 10% of all pages showed significant signs of AI writing or editing. The increase began around ChatGPT's November 2022 launch and continued in the years that followed. Older material still makes up a large portion of the internet, which explains the gap between the 10% figure across the full sample and the higher rate among newer pages.
When researchers limited the sample to pages published after ChatGPT's launch, more than one-third showed significant signs of AI authorship or editing. The pattern holds across different types of websites, though not evenly.
Commercial sites lead; education and government lag
Signs of AI involvement appeared at similar rates across .com, .org, .edu, and .gov domains around the time ChatGPT became available. By 2026, roughly 10% of .com pages showed signs of AI authorship, compared with 4.6% of .org pages and about 1% of .edu and .gov pages.
The lower rates on institutional domains may reflect slower adoption, stricter publishing workflows, or both. The research does not establish causation.
Writing patterns have shifted alongside the tools
Several language features associated with AI models have become more frequent across webpages since 2023. Em dash usage has roughly doubled. Oxford comma usage increased 63%. AI-favored vocabulary more than doubled in use. Negative parallelism, including constructions similar to "it's not just X, it's Y," nearly tripled, though it remains relatively uncommon.
Human writers use many of the same words and habits associated with AI-generated text, so none of these features can establish authorship on their own. Detection models can misclassify individual documents. The findings are most useful for tracking how AI-associated writing has changed across large collections of online content over time.
For writers watching these trends, the data confirms what many already sense: AI-assisted drafting has become a mainstream practice. Resources like AI for Writers and the AI for Technical Writers learning path address how professionals can adapt to this shift without losing editorial judgment.
Why this matters for writers
The web is now a mixed corpus of human and machine-produced text, and your work competes for attention inside it. If AI-associated patterns are becoming more common, writing that reads distinctively human may carry more value. The practical takeaway: know what AI-assisted prose looks like, use the tools deliberately, and treat editing as the step where you remove the fingerprints of machine-generated phrasing.
Your membership also unlocks:
