The dead internet theory just got peer-reviewed data to back it up.
The Summary
- One-third of web pages published since ChatGPT's launch show signs of AI authorship, according to a new study tracking the web's transformation
- AI models are now authoring and editing much of the new web, marking a fundamental shift in how online content is created
- This isn't SEO spam in dark corners — this is the mainstream web being rewritten by machines in real time
The Signal
The web you're reading right now is one-third machine-written. That's the finding from a study tracking content published since ChatGPT's public debut in November 2022. Not one-third of spammy SEO blogs. One-third of all new web pages, period.
This is the flip side of the AI productivity story everyone's been selling. Yes, ChatGPT and its siblings let anyone generate content at scale. The question nobody asked: what happens when everyone does?
"ChatGPT and other AI models are now authoring and editing much of the new web."
The implications stack fast. Search engines already struggle to surface quality when volume explodes. Google's answer has been to add AI overviews that synthesize results, which means AI is now reading AI-written content to answer your questions. The training data for the next generation of models will be heavily contaminated with output from the current generation. We're watching the web eat itself.
This isn't about whether AI writing is "good enough." Some of it is excellent. Some of it is slop. The point is velocity and volume. Human writers produce content at human speed. AI produces it at machine speed. When one-third of new pages come from models, the economic math for human writers shifts hard.
Key shifts underway:
- Content creation speed has decoupled from human labor hours
- The web's signal-to-noise ratio is dropping as volume scales
- Model training on model output creates unknown feedback loops
The Implication
If you make a living writing commodity content, drafting basic web copy, or filling pages for SEO, your market just got flooded. The defense isn't to ignore AI tools. It's to do work machines can't replicate yet: original reporting, deep synthesis, cultural taste-making, or building trust with a specific audience.
For everyone else: get comfortable verifying what you read. The web's baseline assumption used to be "a human wrote this." That assumption just broke. One-third AI authorship today means half by next year, then two-thirds. The question isn't whether agents will write most of the web. It's what happens to the web when they do.