NYT: Google's AI Search Is Choking Open Web Referrals

The clearest read of what is happening to the open web this year comes not from a manifesto but from Cloudflare's server logs, and The New York Times has now made that the frame for how Google's AI is reshaping who visits websites at all. The reporting lands on a phrase publishers have started using in their own strategy decks, "Google Zero", the planning assumption that search referrals go to something close to nothing.

The numbers the piece leans on are worth taking seriously because they are infrastructure numbers, not survey numbers. In Cloudflare's own data, 52% of crawler requests as of June 2026 are for AI training, up from 22% in spring 2025, and mixed-use crawlers that blend search and AI now make up more than a third of activity. Google still accounts for roughly 88% of referral traffic, which sounds like dominance until you notice the other half of the story, that heavily crawled categories including software, IT, retail and finance have seen human traffic fall as much as 40% in under twelve months. The old bargain, crawl in exchange for clicks, is coming apart in public.

Why this matters if you are not a publisher: the web's answer to "who pays for original writing" has been ad-supported search referrals for two decades, and that plumbing is what is quietly being replumbed. Google's AI Mode and AI Overviews answer more questions in-place, so the same query that once sent a reader to a recipe site, a review site or a local paper now resolves inside the search page. The Times itself is not immune, its own search share of traffic reportedly slipped from 44% in 2022 to 37% in 2025, and it has more brand equity to fall back on than almost anyone else.

The honest caveat is that the article is a mosaic of Cloudflare telemetry, publisher anecdotes and one company's referral position, not a controlled study, so take the 40% and the 88% as the framing that a lot of publisher strategy is now built on, not as a settled measurement of Google AI's causal impact. What the reporting does not give you is a clean split between what Overviews are doing versus broader shifts in reader behavior, or how much of the lost search revenue is being clawed back through licensing deals with model makers.

The forward-looking part is that the response is starting to organize. Cloudflare is defaulting to block mixed-use crawlers on ad-hosting pages from September 15, 2026, which for the first time gives the supply side of the web something like collective leverage. Whether that produces a real payments market for training data, or just accelerates a consolidation into a smaller set of licensed, direct-audience publishers, is the actual story to watch over the next twelve months.