Research Summary: The uneven rise of AI-generated web content

An empirical analysis of more than five million crawlable webpages reveals that the proliferation of generative artificial intelligence across the public-facing web has expanded rapidly but unevenly over the past five years. Synthetic text concentrations differ significantly across languages and domain types, with commercial (.com) and English-language properties exhibiting markedly higher prevalence than educational (.edu) and governmental (.gov) environments.

Why it matters

The uneven distribution of AI-generated text alters the underlying integrity and informational quality of the public web, disproportionately affecting commercial and English-centric digital ecosystems. This divergence complicates automated intelligence gathering, data ingestion, and model training due to varying rates of synthetic data contamination. Organizations relying on open-source web data must urgently account for domain- and language-level variations in synthetic content concentration.

Key insights

• Generative artificial intelligence has drastically altered public-facing web content in under five years.
• The increase in AI-generated text across the web is uneven, showing pronounced differences by domain type and language.
• English-language pages contain a noticeably higher proportion of AI-generated text compared to pages published in other languages.
• Institutional domains (.gov and .edu) exhibit significantly lower shares of AI-generated text than commercial (.com) domains, a discrepancy that persists after controlling for compositional effects.
• The presence of synthetic text correlates with structural characteristics, specifically a page's position within the web connectivity structure.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2610.36631