In 2020, 99% of online content was produced by humans. By 2025, more than half is generated by AI. This shift is not just a statistic. It is a point of no return that is transforming the very nature of the Internet — and threatening the foundations on which artificial intelligence itself depends.
We have entered a feedback loop that nobody knows how to break: AI produces content, the Internet fills up with that content, and the next generation of AI models trains on it. The inbreeding of content has begun. And its effects are already measurable.
Model collapse: when AI eats its own waste
Researchers have a name for this phenomenon: model collapse. The principle is simple and devastating. A language model is trained on Internet data. It generates content. That content gets published online. The next model is trained on this same Internet, now contaminated with synthetic material. With each generation, quality drops. Nuance disappears. Diversity collapses.
Research teams from Oxford and Cambridge have demonstrated that models trained on predominantly synthetic content progressively lose their ability to produce rich, varied responses. Vocabulary shrinks. Structures repeat. Hallucinations propagate. It is a digital ecosystem feeding on its own waste — and degenerating.
In my consulting work, I see this playing out concretely. Companies that rely on AI to produce marketing content at scale are starting to observe a race to the bottom. All articles look alike. All emails sound the same. Brand voice vanishes into a fog of generic output.
The Internet is no longer the Internet
The problem extends far beyond AI itself. The entire Internet is mutating. Google search results are now populated with sites entirely generated by AI, designed for SEO and devoid of substance. Amazon is flooded with books written by ChatGPT. Technical support forums are overrun with generated answers that are sometimes wrong but always confident. Social media is drowning in synthetic posts that imitate human tone without carrying human substance.
This is not "just" tech. It is a societal shift. The web as we knew it — a space where humans shared knowledge, experiences, opinions — is disappearing beneath a layer of synthetic content. The "Dead Internet Theory," once dismissed as a fringe conspiracy theory, is becoming a factual observation.
On the ground, the consequences are already here. The data science teams I work with spend increasing amounts of time filtering their training data. Distinguishing human content from synthetic content is becoming a profession in its own right. And this is only the beginning.
The paradox: human content has never been more valuable
The irony is striking. We thought AI would make human content obsolete. The exact opposite is happening. The more AI produces, the rarer authentically human content becomes — and therefore the more precious.
The tech giants understood this before everyone else. Massive partnerships are multiplying: OpenAI with Reddit, Google with news publishers, Apple with publishing houses. What they are buying is not content. It is certified human data. The new oil is not data in general — it is verifiable human data.
For businesses, the signal is clear. Generic content mass-produced by AI will lose all value. What will matter is proven expertise, field experience, the nuance that no model can invent. Your competitive advantage in content will no longer be volume — it will be authenticity.
The countermeasures emerging
Facing this threat, the ecosystem is organizing — but solutions remain partial.
Synthetic content detection: tools like GPTZero or Originality.ai attempt to distinguish generated content from human writing. But it is an arms race: every improvement in detection pushes generators to improve. False positive rates remain problematic.
Watermarking: Google, OpenAI and others are working on invisible watermarks embedded in generated content. Promising in theory, but trivial to bypass through paraphrasing or translation.
Human curation: platforms that guarantee human-verified content are gaining value. Wikipedia is strengthening its verification processes. Stack Overflow has banned AI-generated content. Substack thrives thanks to identified, authentic voices.
Private and proprietary data: the most advanced companies are building internal datasets — cleaned, human, contextualized. This is a strategic asset whose value is exploding.
What this changes for organizations
For the executives I advise, this phenomenon has concrete and immediate implications.
First, content strategy must evolve. Publishing 50 AI-generated articles per week is worthless if your competitors are doing the same thing. What makes the difference is field experience, proprietary data, insights that only a human can provide. AI should amplify a human voice, not replace it.
Second, internal data becomes a major strategic asset. Your knowledge bases, customer feedback, expert reports — everything that is authentically human and contextual takes on considerable value for training domain-specific models.
Third, critical thinking becomes a key competency. In a world where any text can be generated, the ability to evaluate, verify, and contextualize information becomes more valuable than the ability to produce it.
AI will not collapse tomorrow. But the Internet it relies on is undergoing a profound mutation. And organizations that fail to grasp the scale of this change risk building their digital strategy on foundations that are eroding beneath their feet.
I help companies navigate this transition, balancing AI adoption with preserving what makes them unique. If you want to assess your exposure to this risk, let's talk.