Pew study confirms sharp rise of AI-written text on the web since ChatGPT's launch
Pew Research Center analyzed nearly 500,000 English-language web pages and found over a third of pages published after ChatGPT's launch show signs of AI-generated text Commercial .com domains contain AI-generated text roughly ten times more often than .edu or .gov sites (~10% vs ~1%) Characteristic AI language patterns have spiked: words like "delve," "tapestry," and "pivotal" more than doubled in frequency; em dash usage doubled and Oxford comma usage jumped 63% since 2023 Current AI detection
Analysis
TL;DR
- Pew Research Center analyzed nearly 500,000 English-language web pages and found over a third of pages published after ChatGPT's launch show signs of AI-generated text
- Commercial .com domains contain AI-generated text roughly ten times more often than .edu or .gov sites (~10% vs ~1%)
- Characteristic AI language patterns have spiked: words like "delve," "tapestry," and "pivotal" more than doubled in frequency; em dash usage doubled and Oxford comma usage jumped 63% since 2023
- Current AI detection tools like Open Pangram cannot reliably distinguish between fully automated text, AI-assisted writing, or partial AI involvement, limiting the precision of such studies
- A separate April 2026 study by Imperial College London, Internet Archive, and Stanford found ~35% of newly published websites were fully or partly AI-generated, with 33% higher semantic similarity and more positive tone overall
Why It Matters
This research provides large-scale empirical evidence that AI-generated content has become a dominant force on the open web, fundamentally altering the landscape of online information. For AI practitioners and researchers, it highlights the urgent need for better detection methodologies and clearer definitions of what constitutes "AI text," as the current binary framework fails to capture the nuanced reality of human-AI collaboration in writing.
Technical Details
- Dataset: Nearly 500,000 English-language web pages from the Common Crawl web archive, analyzed in a July 2026 sample
- Detection tool: Open Pangram AI detection model used to identify signs of machine authorship across web pages
- Domain breakdown: .com sites at ~10% AI-generated, .org at 4.6%, and .edu/.gov at ~1% each, revealing a tenfold commercial vs. institutional disparity
- Linguistic markers tracked: Frequency analysis of AI-favored vocabulary ("delve," "interplay," "testament," "pivotal," "landscape," "tapestry," "bolstered," "crucial," "meticulous," "vibrant"), em dash usage (2x increase), Oxford comma usage (+63%), and "it's not just X, it's Y" negative parallelism patterns (nearly 3x increase)
- Corroborating study: Imperial College London/Internet Archive/Stanford April 2026 research found 35% of new websites AI-generated, with 33% higher semantic similarity and more positive tone in AI texts
Industry Insight
- Content platforms and search engines must develop more sophisticated detection and labeling systems that account for the spectrum of AI assistance rather than a binary human/machine classification, as partial AI use is likely the norm going forward
- Commercial website operators face increasing reputational and trust risks as AI-generated content saturates .com domains; brands should establish clear disclosure policies to maintain audience credibility
- The polarization around AI authorship stigma—documented in workplace studies and debates over watermarks like Anthropic's planned Claude watermark—suggests the industry needs nuanced frameworks that distinguish between unethical spam generation and legitimate AI-assisted productivity tools
Disclaimer: The above content is generated by AI and is for reference only.