A third of new web pages show signs of AI writing, Pew finds

The Pew Research Center looked at roughly 500,000 English-language web pages from the Common Crawl archive and estimates that about 35 percent of pages published after ChatGPT's November 2022 launch show signs of AI authorship. The detection came from Open Pangram's classifier, and the analysis compared pages from before and after that date. As TechCrunch reports, the effect is heavily concentrated by domain: roughly 10 percent of .com pages carried AI signals, against about 4.6 percent of .org pages and near 1 percent for .edu and .gov.

The tell-tale markers are worth noting, partly because they are the same habits editors now police in AI drafts: more em dashes, more Oxford commas, and constructions like "it's not X, it's Y." Pew is candid that detectors such as Pangram can misclassify individual pages, so the figure is best read as a direction rather than a precise count.

The finding lands alongside other data showing bot traffic has passed human traffic on the web, which points to a loop where AI systems increasingly read text produced by other AI systems.

Why it matters

If your work depends on web data, for training, search, or research, a growing share of what you scrape is machine-written and skewed toward commercial domains. Treat post-2022 .com content with more suspicion, and weight higher-trust domains accordingly.