New Study Reveals One in Three Recent Web Pages Show Signs of AI Generation

Recent research from the Pew Research Centre reveals that over one in three web pages published after the release of ChatGPT in November 2022 show significant signs of AI involvement, reaching 35 per cent. When including pages created before ChatGPT, the overall figure drops to 9.6 per cent, highlighting the rapid integration of AI into content creation.

Data shows that prior to ChatGPT, AI involvement on the web was negligible at around 1 per cent. Following its launch, and with the introduction of tools like Claude and Gemini, the share of AI-generated content steadily climbed. However, adoption varies significantly by domain: in 2026, 9.3 per cent of .com web pages showed signs of AI use, compared to 4.6 per cent for .org pages, and roughly 1 per cent for .edu and .gov domains, indicating that educational and government institutions are largely avoiding AI for official content. Additionally, a study by Graphite Growth indicates that AI is now responsible for producing as many online articles as humans for the first time.

Researchers at the Pew Research Centre also identified distinct linguistic patterns in AI-generated text since 2023. The usage of em dashes has roughly doubled, Oxford commas have increased by 63 per cent, and typical AI buzzwords—such as “align with”, “delve”, “interplay”, “showcase” and “testament”—have more than doubled. Stylistic structures like negative parallelism, which defines concepts by stating what they are not alongside what they are, have nearly tripled. While these individual markers do not definitively prove AI authorship, combined patterns allow researchers to determine AI involvement probabilistically.

Experts note important caveats, as AI detection tools remain imperfect and can misclassify both human and AI text. Furthermore, as AI-generated text proliferates online, questions remain regarding how it may begin to influence and alter human writing styles.