← Content
AI · 5 min read · May 1, 2026

AI text now comprises 35% of new web content, but fears outpace evidence

A 2025 study finds AI-generated text widespread online yet shows mixed support for claims about diversity loss, accuracy decline, or stylistic homogenization.

Source: arxiv/cs.AI · Jonas Dolezal, Sawood Alam, Mark Graham, Maty Bohacek · open original ↗ ↗
Share: X LinkedIn

By mid-2025, 35% of newly published websites contain AI-generated or AI-assisted text; public fears about quality decline exceed statistical evidence.

  • AI-generated or AI-assisted text rose from 0% to 35% of new websites between late 2022 and mid-2025.
  • Increased AI text correlates with lower semantic diversity and higher positive sentiment in measured samples.
  • No statistically significant link found between AI text prevalence and factual accuracy or stylistic diversity loss.
  • US adults surveyed believe all four negative hypotheses despite mixed empirical support from the data.
  • AI users and favorable-view holders report fewer concerns than non-users or skeptics about internet quality.
  • Internet Archive and state-of-the-art detection methods enabled first large-scale measurement of AI text prevalence.
  • Public perception diverges sharply from measured outcomes, suggesting perception-reality gap on AI impact.

Frequently asked

  • According to Dolezal et al. (2025), approximately 35% of newly published websites between 2022 and mid-2025 were classified as AI-generated or AI-assisted. This represents a sharp rise from zero before ChatGPT's launch in late 2022. The measurement used Internet Archive data and state-of-the-art detection methods on a representative sample.

Related