Pew Research Center Findings on AI Content Volume
The Pew Research Center utilized specialized machine learning models to identify linguistic patterns, vocabulary choices, and grammatical structures typical of automated text generators. While an initial broad analysis placed AI involvement at roughly 9.6 percent across all historical web pages in the dataset, filtering out pre-existing human-written material revealed a much higher concentration in recent years, according to the center’s findings. By July 2026, researchers found that more than one-third—specifically 35 percent—of web pages published after the launch of ChatGPT displayed clear markers of AI generation or significant AI editing.
Commercial Websites Lead in AI Generation
The proliferation of machine-generated text varies significantly by domain type, showing a distinct concentration on commercial platforms. When conversational AI tools first emerged, linguistic patterns associated with automated writing appeared at similar rates across commercial (.com), public (.org), academic (.edu), and government (.gov) web addresses. By 2026, however, AI-influenced content on commercial .com domains rose to account for roughly 10 percent of sampled pages. In contrast, public sector sites hovered at 4.6 percent, while education and government portals dropped to just 1 percent.
Linguistic Markers in English and Korean Text
To detect automated writing, the Pew Research Center highlighted specific punctuation and formatting habits common in English-language AI outputs, such as a doubled frequency of em dashes (—) and a 63 percent increase in the use of Oxford commas. Similar stylistic predictability exists in other languages. According to the analysis, Korean-language AI outputs frequently rely on recurring transitional phrases and formulaic structures, including expressions that translate to “beyond simply,” “what is important is,” and “it is necessary to examine from various perspectives.”