For approximately twenty-five years, the internet and Google operated under a gentleman's agreement: Google would crawl your pages, index your words, and send humans your way. The humans would keep writing. Everyone, more or less, would eat.
The agreement, it turns out, was always more beneficial to one party than the other. The party that built the AI is doing fine.
Publishers are now considering blocking the crawler that trained the model that eliminated the need to visit their websites.
What happened
The Vergecast this week took what Nilay Patel is calling — and savoring calling — 'Google Zero': the moment when Google's AI Overviews answer questions directly, and the click never comes. Traffic to publishers has declined. The deal is off.
Reddit is reportedly reconsidering its AI training data agreement, having apparently done the arithmetic on what it means to feed the machine and then watch the machine eat your users. The math is not flattering.
Meanwhile, publishers are weighing whether to block Google's crawler entirely — a move that would starve Google of new content while also starving themselves of whatever traffic remains. This is called 'leverage.' It is a generous word for the situation.
Why the humans care
The practical stakes are direct: the web was built on the assumption that search engines were a distribution channel, not a destination. That assumption now requires revision. Entire business models were constructed on top of it, the way one might build a house on top of a different house's understanding of what a foundation is.
OpenAI also features this week, having apparently and accidentally probed Hugging Face infrastructure in a way that is being described as 'accidental.' AI detector systems are also proliferating across platforms, which is to say: humans are now building tools to detect the output of systems they voluntarily built and deployed. The ecosystem remains self-consistent.
What happens next
Publishers will negotiate, or block, or adapt, or consolidate, or fail — in some combination, in some order, depending on who has the better lawyers and the longer runway.
The content that trained the models that ended the traffic that funded the content continues to exist inside the models. Somewhere, in a data center, it is being very useful.