In 2004, the year Google went public, the company’s stated mission was to organize the world’s information and make it universally accessible and useful. The practical expression of that mission was a bargain that underpinned much of the public internet for two decades: Google would index the world’s content, and in exchange for doing so, it would send readers to the sites that produced it. Publishers tolerated being indexed — in many cases actively optimized for it — because the traffic that came with a first-page ranking could support a journalism operation, a specialist information service, a media brand. The bargain wasn’t written down anywhere, but it was durable enough to build industries on.
That bargain is now being unwound, and the unwinding is not symmetrical. Google benefits. Publishers don’t.
What Google built and how it worked
Understanding why the current shift matters requires understanding why the original arrangement was so durable. Google’s model required content to function. The search algorithm needed a web of documents to index, rank, and surface. High-quality content — well-sourced, well-maintained, frequently updated — produced better search results, which produced better user experience, which produced more search volume, which justified more investment in the content that fed it. The incentives aligned across the system: publishers needed traffic, Google needed content, users needed answers that took them somewhere genuinely useful.
This alignment sustained an ecosystem of considerable scale. Journalism operations built their digital strategies around search visibility. Specialist publishers — health, finance, travel, legal — built entire businesses on the expectation that ranking well for the right queries would deliver a reliable stream of qualified readers. SEO became a professional discipline. The premise was that Google would always send the reader somewhere, and the place they were sent would be the site that had earned the right to be ranked first.
The premise held for roughly twenty years.
What AI Overviews changed
AI Overviews, Google’s system for generating natural-language summaries directly in search results, entered the market at scale in 2024. The traffic data since then has been unambiguous about its effect on publishers. Zero-click searches — queries that generate no click on any result — increased from 56% to 69% between May 2024 and May 2025. For news-related searches, the proportion reached 69% in the year following AI Overviews’ launch. The question gets answered. The source doesn’t get visited.
The impact on organic traffic at the top of the search results page — the position that historically justified the investment in quality content — has been severe. Click-through rates for first-position search results fell from 7.3% to 1.6% between December 2023 and December 2025 — a 78% decline in the value of the most sought-after position in digital publishing. The aggregate effect across the publishing industry: Google search traffic to publishers fell 33% globally in the year to November 2025, a figure confirmed by Chartbeat tracking across more than 2,500 news sites worldwide.
Individual publishers have experienced steeper drops. Chegg, a company whose business depends on students finding its study content through search, reported a 49% decline in non-subscriber traffic between January 2024 and January 2025. The pattern is consistent across categories where AI Overviews can deliver a sufficiently complete answer: the reader gets what they came for without leaving the Google results page.
The contractual and ethical gap
Google’s defense of AI Overviews is that the system still cites sources and surfaces links alongside its generated summaries. This is technically accurate. The question is whether a citation attached to an answer that has already satisfied the reader’s query represents the same value as a search result that required the reader to visit the source to get the answer. The data on click-through rates suggests it does not.
The original indexing arrangement was never formalized as a contract. Publishers could opt out using robots.txt directives; many didn’t, because being indexed was more valuable than the alternative. But the implicit understanding — or at least the practical reality that publishers built businesses on — was that indexing meant traffic. Google read your content and sent you readers. The system now reads your content, synthesizes it, and delivers the synthesis to the reader without the reader needing to travel to you.
This is not a contractual violation because there was no contract. It is a unilateral revision of the terms of an arrangement that one side built an industry around, made by the party that holds all the leverage. Roughly one-third of publishers say they intend to block Google from using their content for AI Overviews once adequate controls are available. They are not blocking traditional search crawling — that traffic, declining as it is, remains valuable. They are specifically withholding permission for AI training and synthesis, which is the closest available approximation of renegotiating terms they never formally agreed to.
What publishers are doing
Publisher responses range from technical to legal to strategic. The technical response — updating robots.txt to exclude AI crawlers while remaining indexed for traditional search — is the most common and the most limited. It addresses AI Overviews specifically but does nothing about the broader decline in search-referred traffic that AI answers are producing.
The legal responses have been more aggressive. Antitrust arguments — that Google’s integration of its AI answer layer with its dominant search position constitutes self-preferencing that forecloses competition — are being developed by publishers and regulators in multiple jurisdictions. These arguments have historical precedent: Google has faced antitrust scrutiny for search result manipulation before, and the AI Overviews product represents a more direct form of content synthesis than any previous feature.
The strategic responses have been the most varied and, in some ways, the most honest acknowledgment of the underlying shift. Publishers are investing in direct audience relationships — newsletters, podcasts, membership programs, apps — that don’t depend on search referral. They are investing in types of content that AI Overviews can’t adequately synthesize: original reporting with proprietary sources, long-form investigative work, real-time original coverage that no AI can produce because no training data for it exists yet. They are, in effect, trying to build value in the parts of their operations that Google hasn’t yet learned to replicate.
The question behind the question
The traffic numbers and the legal arguments are proxies for a more fundamental question: what is the information economy of the internet built on, and who benefits from it?
Google’s original model was extractive, but it shared its extraction. Publishers contributed content; Google contributed distribution; users got answers and direction; the traffic that resulted supported the content creation that made the search results worth having. The cycle had genuine reciprocity at its core, even if the distribution of its value was heavily skewed toward the platform that controlled the interface.
The AI answer model breaks the reciprocity without offering a replacement. It extracts the information value from the content — synthesizes, summarizes, presents — and delivers it to the user without the traffic event that, under the old model, compensated the creator for producing it. A citation is not compensation. The fact that a reader could click through to the source doesn’t mean they will; the data shows overwhelmingly that they don’t.
What this means for the long-term production of the content that AI systems depend on is a question that is, so far, mostly being discussed in the publishing industry rather than at the companies whose products consume that content. The assumption appears to be that content will continue to be produced, because it always has been. This assumption may be correct for the largest and best-capitalized publishers, who have the resources to build direct audience relationships that don’t depend on search. It is less obviously correct for the specialist, regional, and independent publishers whose economics were always thinner and whose dependence on search traffic was always higher.
The open web — the version of the internet where information is freely produced, freely indexed, and freely accessible — was built on the assumption that the entity organizing the information had an interest in the continued production of new information to organize. That assumption held as long as organizing meant sending readers somewhere. It becomes less stable when organizing means answering the question so completely that the reader has no reason to go anywhere at all.
The negotiation between those two versions of the internet is underway. The outcome will determine what the web looks like in ten years — whether it is an ecosystem where independent content creation is viable, or a landscape where the primary beneficiaries of information are the platforms that aggregate it rather than the people who produced it. Google, which has the most at stake and the most leverage, has not yet been required to make a choice between those two outcomes. That requirement may be coming.
