Google Zero is here: if AI answers for the web, who will pay to write it?

THE THESIS. The open internet was built on a tacit barter: creators gave their content to search engines for free, and search engines sent back visitors, who turned into advertising and subscriptions. That pact is breaking, and it is no longer a hypothesis: it is third-party measured data. Our thesis is twofold. In the short term, "Google Zero" — the moment Google stops sending meaningful traffic to websites — is real, it is destroying jobs, and it will close newsrooms. In the long term, what is dying is not the web: it is one specific business model, the ad-funded cheap click. And the market emerging to replace it — content licensing and micropayments for bot access — could end up paying quality content better than the old one did. On one condition: that it isn't captured by the giants alone.

THE NUMBER THAT MATTERS. Google search traffic to news publishers fell 33% globally and 38% in the US between November 2024 and November 2025, according to Chartbeat data cited by the Reuters Institute. The cause has a name: AI Overviews, the AI-generated summaries Google places above its results. When one appears, users click through on just 8% of searches, versus 15% without one, according to Pew Research; and they click the source cited inside the summary barely 1% of the time. The aggregate result: 68% of Google searches now end without visiting any website (SparkToro's analysis of Similarweb data, January-April 2026), up from 60% in 2024. Google's new AI Mode — its conversational search — is even more closed: it refers traffic on only 1.6-2.5% of queries, versus 17-19% for classic search (Similarweb). The casualties have names: Business Insider lost 55% of its organic traffic between 2022 and 2025 (Similarweb) and laid off 21% of its staff; some tech sites have seen drops of up to 85% (Growtika). And news executives expect another 43% decline within three years (Reuters Institute survey). This is not a rough patch. It is a regime change.

THE SINGLE-CRAWLER BIND. Here is the technical detail that explains almost everything. A crawler is the program that roams the web copying pages. Googlebot, Google's crawler, feeds both classic search and AI Overviews: the same robot does both jobs. Google offers a separate switch, Google-Extended, but it only stops your content from training Gemini; it does not remove you from the AI summaries. The only real way out of AI Overviews is a tag ("nosnippet") that also degrades how you appear in normal search. Translated: either you accept that AI summarizes your work without sending visits back, or you also vanish from classic search, which is still 90% of the market. That is not a market with options; it is a toll road with a single gate. Even federal judge Amit Mehta, at a hearing in the Penske case against Google, described it in words rarely heard in a courtroom: it "seems really unfair."

THE PUBLISHER REVOLT (AND GOOGLE'S REPLY). The unthinkable is now on the table. According to the Wall Street Journal (via Nieman Lab, July 2026), People Inc. says blocking Google is "100% on the table"; Gannett (USA Today) warns it may cut off access if traffic keeps falling; Politico and Reuters are weighing registration walls; and Reddit is re-evaluating its $60-million-a-year deal with Google. In court, Penske Media (Rolling Stone, Variety) filed in September 2025 the first major antitrust lawsuit by a US publisher against AI Overviews, alleging click-through declines of up to 58%. Google replies that AI in Search drives "more queries and higher-quality clicks" and that total organic click volume is "relatively stable"; that is its account, and it is worth noting the company has not backed it with public figures. When one side publishes auditable data and the other publishes adjectives, our editorial line is clear: the data wins.

A MARKET IS BORN. If traffic no longer pays for content, what will? Three paths are emerging. One: direct licensing, like the News Corp-OpenAI deal, valued at more than $250 million over five years in exchange for the archives of the Wall Street Journal and other titles. Two: micropayments for bot access. Cloudflare — the company a huge share of web traffic flows through — launched "pay-per-crawl": when an AI bot requests a page, it gets HTTP code 402, "Payment Required," a status code reserved since the 1990s and never used at scale; pay up or stay out. Since September 15, 2026, Cloudflare also blocks mixed-use crawlers — those that index and train AI at once — by default on ad-supported pages, striking directly at the single-crawler bind. Stack Overflow adopted the model in February. And three, the most interesting: "pay-per-use" — charging not when the bot reads, but when your content appears in an answer. The uncomfortable question: million-dollar licenses only reach those with the muscle to negotiate with OpenAI or Google. Pay-per-crawl is more democratic — it is infrastructure, not negotiation — but its per-page rates are pennies. Today, the new money sustains News Corp; it does not yet sustain the local newspaper.

OUR TAKE, AND WHAT COMES NEXT. Let's separate hype from data. First: the web is not dying; attention is shifting, and answer engines desperately need fresh, reliable content. If AI Overviews strangle their sources, they degrade their own product: an answer machine without journalism has nothing to answer with. That dependency is publishers' best bargaining chip, and it explains why licensing money exists at all. Second: what is dying is the cheap-click economy — content written for the algorithm, not the reader. The death of clickbait is not entirely bad news. Third, and here we do not sugarcoat: the danger is the interregnum. Between the old model's death and the new market's maturity, whole newsrooms are being laid off and local news deserts are spreading — precisely when AI-generated disinformation makes lying cheaper. And fourth: if only the giants get paid, the diverse web becomes an oligopoly of content suppliers to four AI companies; that too is a concentration of power, even if it comes with a check. Our optimism, measured and long-term, is this: for the first time, both the technical plumbing (Cloudflare's 402) and the commercial precedent (licensing) exist to pay for content based on its value rather than its ability to generate clicks. If that market matures and opens up to smaller players — with price transparency and, probably, a regulatory push grounded in evidence rather than panic — journalism could come out better funded than it went in. The next 24 months will tell: the Penske ruling, real-world adoption of pay-per-crawl, and whether a major publisher finally dares to switch Google off. That day we will know whether the web's new pact is a market or a fiefdom.