Google Zero is here. The web isn't ready.

For nearly two decades, the web ran on a simple deal. Google crawled sites and indexed them. In return, it sent a flood of referral traffic that built the internet economy. That flood is now being shut off, drip by drip.

The term 'Google Zero' has been discussed for years, but this month it turned concrete. Cloudflare announced it will block mixed-use web crawlers by default from September 2026, giving publishers a practical way to stop Google from using their content for AI training without losing search visibility. Reddit is reportedly trying to exit an AI training deal with Google. And publishers are now openly debating whether to block Google's crawlers entirely.

The driver is Google's own AI. AI Overviews and AI Mode now answer many queries directly, without requiring a click. Google's AI Mode now serves as a shopping cart and playlist manager, locking users inside the search ecosystem. Each query resolved in an AI summary means a visitor the publisher never sees, a dynamic explored in research on LLM detector paradoxes.

Cloudflare's new policy targets what it calls mixed-use crawlers: bots that serve both search indexing and AI training. CEO Matthew Prince explained the shift because 'the majority of traffic on the Internet is non-human.' Cloudflare singled out Google for having access to about '2x more information' than other AI companies, arguing that Google makes it hard for customers to stay discoverable without feeding AI. Google pushed back, pointing to its Google Extended bot, which lets site owners opt out of AI training without harming search rankings. But Googlebot still crawls for AI Overviews and AI Mode, raising concerns documented in a study on data poisoning risks.

For publishers, the calculation has changed. For years, they accepted Google's dominance because the traffic offset the loss of control. Now that traffic is shrinking. AI summaries are estimated to reduce click-through rates by double-digit percentages for many queries. The result is a slow-motion crisis: publishers need Google for visibility, but Google is keeping the audience, a dynamic similar to an investigation into hidden AI costs.

The pushback is spreading. Reddit, a massive source of human conversation, is reportedly trying to exit an AI training deal with Google. That would be significant because Reddit's content has been a key training source for AI models. Other publishers are considering blocking Google's crawlers entirely, something unthinkable a few years ago and part of a broader industry shift described in a letter on AI policy and open models.

Cloudflare's deadline offers an off-ramp. From September 2026, any bot that mixes search crawling with AI training will be blocked on sites using Cloudflare's default settings. That forces Google to choose: let publishers opt out of AI use without affecting search visibility, or lose access to large parts of the web.

What comes next is unclear. Some publishers are licensing content directly to AI companies, cutting out the search middleman. Others are building their own AI tools to deliver journalism directly. A few are simply accepting smaller audiences and shifting toward subscriptions.

No replacement for Google's traffic has appeared. The web is entering a phase where the old deal is dead and the new rules aren't written. The one certainty is that the era of free, massive referral traffic from search is ending. Publishers who built their business on it will need another way to survive.