AI content generation is a process of inputs, transformation and outputs. Better inputs such as higher quality data and better processing “hidden function” will result into better outputs, but no process is 100% efficient, let alone 101%. Each iteration has a transformation cost; output divergence from input, causing AI hallucinations in many cases. How is this contributing to the internet demise.
74% of recent webpages were AI generated
A study conducted in 2025 by Ryan Law, Xibeijia Guan and Tim Soulo have concluded that 74% of recently published webpages had AI either completely generate the content or at least contributed in the making of said webpage. While, no detector is 100% true but it’s very much within reasonable usage, just like the study said, it’s very low cost usually free and embedded into content creation tools.
AI Crawling: Opt-Out by Default
AI model makers decided to make crawling opt-out instead of opt-in which makes the current progression fast; lots of human generated data input is good, but what’s coming next is statistically not human generated so AI is taking it’s own output as input for next generation making it….. Actually better! this technique already in use by AI model makers; using synthetic data to enhance performance and “accuracy”, usually introducing some kind of bias in the equation too, sooo… not better!
AI is already trained on synthetic, data crawling synthetic data, generating more synthetic data!
Actual results?
A study has been made about the use of synthetic data in making AI models in 2023 by Ranjeeta Bhattacharya “What Happens When We Train AI on AI-Generated Data?” concluding that the use of synthetic data is good to an extent; then it becomes progressively worse, unless fresh data human data introduced in the loop.
Plateau or Progress?
The gist is upcoming models will be either slightly better, same or eventually worse depending on where are we on the curve, due to the ever increasing usage of AI in content generation.
I believe we already reached a plateau on AI textual generation, and evidently new models don’t bring smarter capabilities but more integrations. All we seen in 2024 and 2025 were new integrations of AI, but the model themselves didn’t get much smarter or begin to step into new intelligence territory. I believe we reached the top of this technique’s exponential growth. Either a new fundamentally different technology (i.e. GPT) needs to be made to take this into higher levels, or its will be like this for most of its existence.
