Turns out, the bots haven’t taken over the web — not yet, anyway. A new analysis from Axios and Graphite reveals that AI-generated articles, which briefly outnumbered human-written content online in late 2024, have now plateaued.
Humans and machines are sitting at a near 50-50 split — but the tide might be turning back toward us.
Graphite analyzed 65,000 web pages and found that 86% of top-ranking Google results were written by humans.
Even chatbots like ChatGPT and Perplexity cite human-made articles more than 80% of the time.
So despite the flood of synthetic text since ChatGPT’s 2023 debut, readers — and algorithms — still seem to prefer the messy authenticity of human prose.
Some researchers say this shift is a survival instinct. If AI starts training mostly on AI-generated data, it risks collapsing under its own weight — an idea long dubbed the “model collapse” problem.
Ironically, that’s something Europol warned about years ago, predicting a web flooded by synthetic noise.
That apocalypse hasn’t quite arrived — maybe because we got bored of machine-perfect sentences.
But it’s not just sentimentality driving the trend.
Platforms like Google have been quietly demoting low-quality AI spam in search results, and many content farms are learning that churning out auto-written pages doesn’t pay like it used to.
In short, the bots can write, but they can’t rank.
Meanwhile, debates around authorship and ownership are heating up.
The U.S. Supreme Court is now weighing whether AI-generated works deserve copyright protection, and a growing movement of publishers is calling for “organic content” certification — labels that mark writing as human-made.
The big question isn’t whether AI will replace writers. It’s whether audiences can still tell the difference — or care.
For now, the numbers hint at a quiet rebellion: people clicking on stories written by people.
And maybe that’s the strangest plot twist of all — in a world where machines can mimic anything, it’s our imperfections that make us readable.
