Study: a third of new online content is written by artificial intelligence

An analysis of half a million web pages by the Pew Research Center found that at least 10% of all active internet pages show signs of AI creation. Among materials published after the launch of ChatGPT, the share of machine-generated content exceeds one-third. Commercial sites lead in the use of AI texts, while academic and government resources remain on the sidelines. In Europe, there are fears of a 'model collapse,' and in Israel, warnings about a decline in the quality of Hebrew.

A large-scale study by the American Pew Research Center, based on an analysis of half a million English-language web pages from the Common Crawl database, found that at least 10% of all active internet pages show clear signs of being created or edited using artificial intelligence. Among materials published after the launch of ChatGPT in late 2022, the share of machine-generated content exceeds one-third. AI texts are most actively used on commercial sites (about 10%), while non-profit organizations (about 5%) and especially academic and government resources (about 1%) still maintain low levels. Researchers attribute this to the direct economic incentive for commercial platforms. In Europe, amid the entry into force of the AI law, concerns are being expressed about the possibility of a 'model collapse'—a scenario in which future AI systems are trained on low-quality synthetic texts. In Israel, a decline in the quality of Hebrew texts is noted due to the introduction of AI tools for search engine optimization. The study also revealed a characteristic 'signature' of machines: a sharp increase in the use of the em dash, the serial comma, and repetitive words ('interweaving,' 'component,' 'accentuation,' 'deep dive'). Detection tools such as Open Pangram, GPTZero, and Turnitin acknowledge that they are in a continuous arms race with evolving language models.