10% of Web Pages Now Show AI Authorship, Pew Research Finds

Growing AI presence across the web

A new Pew Research Center study has quantified what many internet users have suspected: AI-generated text is becoming increasingly prevalent online. Researchers analyzed nearly 500,000 English-language webpages spanning five years and found that approximately 10% of pages sampled in July 2026 show significant signs of AI authorship.

The research team used Common Crawl, a comprehensive web archive, to examine content published both before and after ChatGPT's public release in November 2022. They employed Open Pangram, an AI detection tool, to identify machine-generated text, then manually reviewed 10,000 recent pages for linguistic patterns characteristic of AI writing.

Why it matters

This research provides the first large-scale empirical evidence of AI's growing role in creating web content. For business leaders and content strategists, the findings underscore a fundamental shift in the internet's information ecosystem—one that affects everything from search engine optimization to brand authenticity. When one in ten pages may be AI-generated, organizations must reconsider how they verify sources, maintain content quality standards, and differentiate their human-created materials in an increasingly synthetic landscape.

Commercial domains lead AI adoption

The distribution of AI-generated content varies significantly by domain type. Commercial .com websites show the highest concentration, with roughly one in ten pages displaying AI authorship. Nonprofit .org domains follow at 4.6%, while educational and government sites (.edu and .gov) show approximately 1% AI-generated content.

This pattern suggests that commercial entities have adopted generative AI tools more rapidly than institutional or governmental organizations, likely driven by cost efficiency and content production demands.

Linguistic fingerprints of machine writing

Researchers identified specific markers that distinguish AI-generated text from human writing. Machine-produced content uses em dashes twice as frequently as human authors and employs Oxford commas in lists 63% more often. These patterns emerge because AI models are trained on large datasets of human text and tend to over-apply certain stylistic conventions.

Certain vocabulary choices also signal AI authorship. Words like "delve" and "interplay" appear with unusual frequency in machine-generated text. Additionally, AI writing often exhibits repetitive explanations, overuses key terms from the original prompt, and maintains an unnaturally consistent tone that lacks the natural variation of human prose.

Accelerating trend since 2022

The researchers emphasized that their 10% figure represents a conservative estimate across all web content, including older pages created before generative AI tools existed. When filtering the sample to include only pages published after ChatGPT's November 2022 launch, the proportion of AI-generated content increases substantially.

The trend accelerated as additional AI systems entered the market, including Anthropic's Claude and Google's Gemini. This timing aligns with the broader adoption curve of generative AI tools across industries.

Detection limitations acknowledged

The Pew team noted that AI detection models remain imperfect, sometimes flagging human-written text as machine-generated and vice versa. This uncertainty complicates efforts to definitively measure AI's presence online and raises questions about content verification strategies going forward.

These findings were first reported by Pew Research Center in a study published in August 2026.