Study Reveals AI Now Generates a Third of New Websites, But Defies 'Dead Internet' Fears
Key Takeaways
- ▸35% of newly published websites since late 2022 are AI-generated or AI-assisted, representing unprecedented speed of internet transformation in just three years
- ▸AI-generated content does reduce semantic diversity and produces more uniformly positive tone, but does not significantly increase factual inaccuracy or reduce source attribution
- ▸The research highlights a subtle but important risk: the homogenization of internet voice and perspective, rather than proliferation of outright disinformation
Summary
Researchers from Stanford, Imperial College London, and the Internet Archive have published findings showing that approximately 35% of newly created websites since late 2022 are AI-generated or AI-assisted, up from nearly zero before ChatGPT's launch. Using AI-detection software Pangram v3 and samples from the Wayback Machine, the team analyzed website content from August 2022 through May 2025 to test six common concerns about AI-generated text, including whether it increases disinformation, reduces semantic diversity, or diminishes source attribution.
The research delivered mixed results that challenge some apocalyptic predictions about AI's impact on the internet. While the data confirmed that AI-generated content does reduce semantic diversity and makes the web more uniformly positive in tone, it refuted predictions of rampant factual inaccuracy—human fact-checkers found no significant increase in verifiably false claims. Additionally, AI-generated websites did not reduce outbound link density or source attribution as feared. "The most surprising result was that our Truth Decay hypothesis wasn't confirmed," noted Stanford AI researcher Jonáš Doležal, though he cautioned that AI may be increasing unverifiable claims that current fact-checking tools cannot detect.
The study illuminates the stunning velocity of AI's integration into digital infrastructure: in just three years since ChatGPT's November 2022 launch, a significant portion of the internet has been fundamentally reshaped by generative AI. The research provides empirical grounding for debates about AI's societal impact, showing that while the technology has broadly homogenized web content, it has not produced the information apocalypse some predicted.
Editorial Opinion
This research offers a sobering but nuanced perspective on AI's reshaping of the internet. While it deflates some doomsday narratives about AI-generated misinformation, it reveals a potentially more insidious problem: the rapid consolidation of internet writing into a narrow, optimistic, semantically simplified register. The researchers' concern about unverifiable claims lurking beneath the surface of AI-generated content is particularly sharp—absence of detected lies doesn't guarantee intellectual diversity or truthfulness. Three years in, we're seeing not a 'dead internet' of bot-to-bot communication, but a domesticated one.
