Writing had been the preserve of human beings for thousands of years, but in a span of just three, the machines have taken over.
According to data from FT, AI-generated articles now constitute approximately 50% of all content published on the web, surpassing human-written articles for the first time in history. The crossover point came in late 2024, marking a profound shift in how information is created and distributed online.
The transformation has been remarkably swift. From 2020 through most of 2022, AI-generated content represented less than 10% of web articles, holding steady while human writing dominated at nearly 100%. Then came ChatGPT’s launch in late 2022, which catalyzed an explosive change. AI content production surged from single digits to nearly 20% within months, climbing steadily to around 40% by mid-2024 before crossing the halfway mark. Meanwhile, human-written content plummeted from its historical dominance to roughly 50% today, with the decline accelerating dramatically after 2023.
The implications of this reversal extend far beyond simple statistics. The internet as we know it was built on human expression, human expertise, and human creativity. Search engines, recommendation algorithms, and yes, the large language models themselves were all trained on this vast corpus of human-generated text. These AI systems learned to write by reading what humans wrote, absorbing our patterns of thought, our idioms, our ways of structuring arguments and telling stories.
Now those same systems are poised to replace the very foundation they were built upon. It’s a peculiar form of digital succession: the student becoming the teacher, the copy replacing the original. As AI-generated content continues to proliferate, future language models will increasingly train on text written by their predecessors rather than by humans. We’re entering an era of synthetic training data, where AI learns from AI in a recursive loop that could amplify both the strengths and weaknesses of these systems.
This shift raises uncomfortable questions for publishers, platforms, and internet users alike. If half of what we read online is now machine-generated, how do we distinguish authentic human perspective from algorithmic output? What happens to human writers as demand for their work contracts? And perhaps most critically, what gets lost when efficiency and scale take precedence over the messy, subjective, irreplaceable nature of human thought?
The machines haven’t just arrived at the table. They’ve taken the majority of seats.