This article sums up a multinational study tracking how AI is shaping the web by looking at archived pages from the Wayback Machine. Researchers from Imperial College London, Stanford University, and the Internet Archive dug into data from late 2022 through mid-2025, hoping to quantify AI’s role in new websites and test out six big ideas about how AI-generated content might affect information quality, tone, and diversity online.
What the study did
The team used old snapshots of the internet to see how often AI helped or fully created new websites. They focused on the explosion of generative AI after ChatGPT launched.
They also compared their findings to other industry signals about bot and AI activity, pulling in reports from Cloudflare and Imperva. Their goal wasn’t just to count AI-generated pages—they wanted to get a sense of how AI might shift the whole information ecosystem over time.
AI-generated share of new websites
Here’s the headline: 35.3% of newly published websites had some AI help, while 17.6% were entirely AI-generated. This whole trend really took off with ChatGPT’s arrival, which changed the pace and scale of AI adoption for web content almost overnight.
- 35.3% of new sites were made with AI tools—people leaned on AI to draft, design, or publish pages.
- 17.6% were fully AI-generated, meaning AI could spit out publish-ready content without much (or any) human touch.
- The ChatGPT era marks a sharp shift in how people build and fill websites.
Quality and factuality of AI content
People worry about misinformation, but the study didn’t find that AI-generated content was much more factually incorrect on average. In quite a few cases, AI-written articles even included links to credible sources, which hints at some built-in accountability when those sources are available.
The researchers didn’t see a total erasure of personal writing styles, either. AI content wasn’t always just long, empty blocks of bland text. That said, as AI’s role grew, a sanitized and overly cheerful prose style started to creep in, nudging the overall tone in a certain direction.
Impacts on ideas and tone
When AI stepped in, the range of ideas and viewpoints got noticeably narrower. That could shape public discourse by squeezing out the diversity of perspectives that usually make the web interesting and useful.
AI speeds up content production, but it also tends to concentrate voices. And those tone shifts—more upbeat, more polished—might subtly change how readers trust or relate to what they’re seeing, even if the facts don’t really change.
Potential harms and real-world risks
The researchers flagged some practical harms that come with AI-powered content production. Scammers and SEO farms can use AI to churn out fake sites and plagiarized news at scale.
There are even stories of political groups running campaigns with mostly AI-written articles to attack critics or manipulate opinions. All of this makes it clear: we need better detection, attribution, and quality checks as AI gets woven deeper into everyday publishing.
What comes next: monitoring and policy implications
The study team wants to build a continuous monitoring tool that tracks where and how AI-generated content shapes the web. They’ll look at changes by category and language.
This tool could give researchers, platforms, and policymakers a clearer view of shifts in AI-driven publishing. It might help spot high-risk domains and guide strategies to protect information integrity—without shutting down innovation, hopefully.
Here is the source article for this story: Dead Internet Theory Is 17% of the Way to Becoming Reality, Study Finds