Creeping AI: One in three new web pages weren't generated by a human

A growing number of web pages being published online appear to be influenced by artificial intelligence, according to new research from the Pew Research Centre. The findings suggest that one in three new web pages, approximately 35 per cent, shows signs of AI involvement, marking a sharp increase compared to earlier years. The study analyzed hundreds of thousands of web pages to assess the extent of AI’s role in content creation. Web pages published after the release of ChatGPT in November 2022 exhibited significantly higher levels of AI influence, with 35 per cent showing indicators of AI generation or editing. When pages created prior to ChatGPT’s launch were included in the analysis, the percentage dropped to 9.6 per cent, highlighting the rapid rise in AI-driven content over the past few years. Prior to the launch of ChatGPT, AI involvement in web content was minimal, with less than 1 per cent of pages displaying signs of AI contribution. However, following the introduction of large language models such as ChatGPT, Claude, and Gemini into mainstream content production, the use of AI in generating or editing web pages increased steadily. By 2026, the proportion had risen to 9.3 per cent for .com websites, roughly double the rate observed for .org sites, which stood at 4.6 per cent. Educational (.edu) and governmental (.gov) websites saw even lower rates, with AI involvement remaining near 1 per cent. These figures suggest that certain sectors, particularly education and government, are more cautious in their adoption of AI-generated content. An additional study conducted by Graphite Growth found that AI has now reached a critical threshold in online journalism. For the first time, AI-generated articles have matched the volume produced by human writers, indicating a major shift in how content is created and distributed across digital platforms. Pew researchers also examined linguistic patterns in AI-generated text to better understand its distinctive characteristics. Since 2023, the use of em dashes has more than doubled, while Oxford commas have increased by 63 per cent. Words and phrases commonly associated with AI writing, such as “align with,” “delve,” “interplay,” “showcase,” and “testament,” have more than doubled in usage. Writing structures like negative parallelism, which involves defining something by contrasting what it is not with what it is, have nearly tripled in frequency. While none of these patterns alone can conclusively identify AI-generated content, collectively they help estimate the likelihood of AI involvement. It is important to note that AI detection tools remain imperfect. Some human-written content has been misclassified as AI-generated, while others have been incorrectly labeled as human-created. This raises concerns about the reliability of current methods for distinguishing between AI and human-generated material. As AI continues to shape the landscape of online content, questions arise about its broader implications. Could the increasing prevalence of AI-generated text begin to alter the way humans approach writing? With AI becoming a standard tool in content creation, the boundaries between human and machine authorship may continue to blur.