Google is devouring the publishers whose content built its search monopoly.
The numbers are precise enough to foreclose euphemism. Between June 2025 and June 2026, organic search traffic from U.S. users to USA Today fell by nearly half. Business Insider lost more than 85 percent. Politico dropped 23 percent, CNN 25 percent. The Semrush measurement data tracks what happens when a platform that controls the chokepoint between content producers and their audience decides the content is more valuable as raw material than as a destination.
It is true that Google says its AI search features “send billions of clicks to the web every week.” The statement is accurate in the narrow sense in which those statements are usually accurate: the number is not zero. It was also higher last year, and higher still the year before. Google’s AI Overviews answer the query on the results page — a restructuring of the search box that rolled out this spring — eliminating the click-through entirely. The content that makes the answer possible is researched, written, and edited at real cost by real newsrooms. The traffic that once compensated that work is retained on Google’s page. The publisher’s content is consumed; the publisher is not compensated. In engineering terms, this is a successful data pipeline. In political economy, it is extraction.
Cory Doctorow has described the mechanism at book length. A platform is good to its users and good to its business customers until the switching costs are high enough that it can claw back value from both. Search was good to publishers for twenty years — sent them traffic, made them legible, built the dependency. A German court has already ruled Google liable for what AI Overviews do with that content. AI Overviews are the third act of the enshittification sequence: the dependency is harvested.
The constraints that historically limited this kind of extraction have been dismantled in sequence. There is no competing search-distribution channel of comparable scale; Microsoft’s Bing holds single-digit market share, and social-media referral traffic has declined across every major platform. The most meaningful regulation to date is a June ruling in the U.K. requiring Google to let publishers opt out of AI features without losing their ranking in traditional results — a principle Google has agreed to test in one country before deciding whether it applies anywhere else. A German court has already ruled Google liable for AI search summaries. Self-help means blocking Google’s crawler, which means forfeiting whatever audience remains.
Reddit is the chokepoint logic in miniature. The company struck a sixty-million-dollar-a-year deal with Google in 2024, granting AI training rights to its user-generated content, and is now — according to people familiar with the matter — reassessing whether the arrangement is worth continuing. The deal was made when Google was the indispensable distribution channel, when appearing in Google search was the price of being found at all. Reddit’s content — years of human conversation, searchable and indexed — is the raw material Google’s AI Overviews consume to answer queries without sending readers to Reddit. The sixty million dollars was not payment for that content. It was rent for continued access to the chokepoint. The chokepoint is now extracting more value than it returns.
The structural trap is precise and applies to every publisher in the dataset. You cannot leave Google without losing your audience. You cannot stay without feeding the system that is replacing you. The question for publishers now is what leverage they actually hold — and on that measure, the picture is nuanced. Politico employees have discussed adding a registration wall requiring humans to log in. A wall that excludes Google’s crawler from newly published content would, in principle, degrade AI Overview quality over time: the machine answers are only as good as the material Google can reach, and a publisher going dark to crawlers means the summaries drawn from that source age out while competitors’ fresher content is scraped in. That degradation of answer quality is the kind of asymmetry that could, if it materialized at scale, give publishers real negotiating capital. But sitting across the table from that possibility is a calculation Google continues to make: enough publishers will continue feeding the crawl because the alternative is oblivion. USA Today’s CEO, Mike Reed, says the company is willing to turn off Google’s crawler entirely if the trend continues. People Inc.’s CEO calls it “100% on the table.” These are acts of self-help by publishers who understand that unilateral action is insufficient — that without the competitive and regulatory constraints reinstalled at structural scale, blocking one chokepoint means falling through the floor.
The adaptations are telling. Cloudflare reports that bots now account for more than half of all web traffic. Time is creating text-based advertisements invisible to human readers but designed to be picked up by AI crawlers, hoping that Google’s Overviews will reference the material. “There are now two audiences,” Time’s chief operating officer told the Journal. The first audience is human readers, declining. The second audience is non-human, growing. Publishers are building for the machine that is consuming them, optimizing for an algorithm they do not control, under extraction terms that change without notice. The technical term for an arrangement in which one party’s labor is consumed by another party’s infrastructure and monetized exclusively by that second party is not partnership.
The regulatory cracks are real but thin. The German court has ruled Google liable for AI search summaries. The U.K. has required Google to let publishers opt out of AI features without sacrificing traditional search visibility. These are the first structural constraints reinstalled on a chokepoint that has operated without friction for two decades. Google’s response — test the U.K. ruling in Britain before rolling it out globally — reveals the posture: comply where the law requires, extract where it does not. A working U.K. opt-out mechanism, even one designed purely as a minimal-compliance demonstration, would create a functional model that other regulators and publishers could point to. But Google has given no indication it would adopt such a model voluntarily outside a jurisdiction that compels it, and no jurisdiction has yet followed the CMA’s lead. Interoperability mandates that would let publishers distribute content through competing channels, privacy frameworks that limit how scraped content feeds AI systems, and antitrust enforcement that treats the conditioning of search visibility on AI-content submission as the exclusionary conduct it is — these are the structural remedies that would make the publishers’ self-help viable rather than desperate. Without them, blocking Google is a publisher’s way of saying enough is enough to a platform that has already decided how much is enough.
Google will test the U.K. ruling in one country before deciding whether publishers anywhere else retain the right to control whether their content feeds a system designed to make them unnecessary. The question Google is answering is not whether the principle is correct. It is whether compliance is cheaper than litigation.