When was the last time you read something on the internet and wondered if a human actually wrote it? That question is no longer philosophical. A landmark study from Stanford, Imperial College London, and the Internet Archive has an answer that should concern everyone who publishes, searches, or trusts anything online: roughly 35% of websites created since 2022 are AI-generated or AI-assisted. By mid-2025, the web had crossed a threshold that took decades of human effort to build in the first place.
The research, titled "The Impact of AI-Generated Text on the Internet," analyzed 33 months of web content from August 2022 through May 2025. Using Pangram v3, an AI-detection tool the researchers validated as the most accurate among available options, they classified websites from archived snapshots pulled via the Wayback Machine. Before ChatGPT launched in late 2022, the figure was effectively zero. By 2025, more than a third of new web content came from machines.
Jonáš Doležal, an AI researcher at Stanford and co-author of the paper, called the pace "staggering." "After decades of humans shaping it, a significant portion of the internet has become defined by AI in just three years," he told 404 Media. "We're witnessing, in my opinion, a major transformation of the digital landscape in a fraction of the time it took to build in the first place."
The implications extend far beyond the headline number. The study tested six common fears about AI content: whether it shrinks viewpoint diversity, increases disinformation through hallucinations, produces sanitized and overly cheerful prose, fails to cite sources, generates low-density semantic strings, and creates a homogenized writing monoculture. On style, the data confirmed what many readers have felt intuitively: AI-generated text trends toward the upbeat and the economical. The web is becoming more cheery and less verbose. Whether that constitutes progress depends entirely on what you think the internet was supposed to be.
For SEO practitioners and content marketers, the numbers represent both opportunity and existential threat. When a third of competing pages are generated in seconds by language models, the old rules about keyword density and backlink strategy become relics. Fresh, human-generated content that demonstrates genuine expertise—something AI still struggles to replicate authentically—may become a premium asset precisely because it's scarce. Search engines face their own crisis: training on AI content to find AI content risks creating feedback loops that degrade relevance over time.
For researchers, journalists, and anyone who uses the web as a primary information source, the trust calculus has shifted. The study's authors note that AI-generated content is "feared to contribute to a degradation in semantic and stylistic diversity, factual accuracy, and other negative developments." Those fears are no longer theoretical. They are measured, quantified, and growing.
The Dead Internet Theory—the idea that much of the web now consists of bots talking to bots—started as conspiracy speculation. Three years of aggressive AI deployment have made it partially true. The question is no longer whether machines can produce convincing text. They clearly can. The question is whether humans will notice, care, or develop new literacies to navigate a web where the ratio of synthetic to authentic content keeps climbing.