Was the 90% AI Content Prediction Right? Sort of, but It Measured the Wrong Thing.

In 2023, a widely circulated Europol report warned that as much as 90% of online content could be synthetically generated by 2026. That year is here. So how did the prediction hold up?

The short answer: AI has taken over new content creation at a scale that surprised even optimistic observers, but the framing of "90% of the internet" was always measuring the wrong thing.

What the research actually shows

The actual data for 2025-2026 is striking enough without any embellishment. An Ahrefs analysis of 900,000 newly created web pages found that 74% contained AI-generated content. A Stanford AI Index figure cited by multiple outlets puts AI-generated text at roughly 52% of all new content published online. A more conservative academic estimate, published in an April 2026 arXiv paper, suggests at least 30-40% of text on currently active web pages comes from AI sources.

These numbers are genuinely large. They're also not 90%, and they apply primarily to new content. The internet as a whole, the entire archive accumulated over three decades, still skews heavily human-generated. The balance is changing fast, but "90% of the internet" was never realistic.

Why the prediction was misleading

The Europol report was focused on a specific concern: synthetic media used for disinformation. It was talking about deepfakes, manipulated images, AI-generated news articles designed to deceive. The "90%" figure was pulled into a broader conversation about AI content in general, losing most of its context along the way.

The fear it was meant to communicate, that it becomes impossible to trust anything you read or see online, is legitimate. But that concern doesn't depend on the 90% figure being accurate. AI-generated disinformation is a real problem at much lower percentages. A lie doesn't need to be the majority of content to cause damage.

What's actually changed in 2026

What's genuinely new is that AI content is no longer easy to spot. The early AI writing was stilted and repetitive in ways that most readers could identify fairly quickly. Current models produce text that often reads as human without careful inspection, and AI detection tools have high error rates in both directions, per Ahrefs. The challenge in 2026 isn't the volume; it's the quality of the imitation.

The 90% prediction served its purpose as a wake-up call. But taken literally, it's wrong, and focusing on that specific number risks missing what's actually happening, which is more complicated and harder to quantify than a single alarming statistic. The internet isn't 90% AI. But the new internet is increasingly AI by default, and that's a real enough shift without the inflation.