AI Search Broke the Web's Implicit Contract
For 25 years, finding something online meant one thing: Google sent you somewhere. Not a summary — a destination. You clicked a blue link, landed on a page, and the person who wrote that page got your visit counted. That visit was money, or the promise of it.
That's over now.
Google's AI Overviews, which rolled out to all US search results in May 2024, now deliver synthesized answers at the top of the page for hundreds of millions of queries — health questions, recipe steps, product comparisons, code snippets. The answer appears. You don't click. You're already done.
The design is useful. The economics it replaced were not incidental.
The Deal That Built the Web
The open web was never a charity project. Publishers — from individual bloggers to major newsrooms — created content because search engines indexed it and sent readers their way. Google's PageRank model, introduced in 1998, made this trade explicit: create quality content, earn links, receive traffic. Traffic converted to advertising revenue, subscriptions, or leads. The web ran on this engine for a generation.
The deal wasn't written down. It didn't need to be. It was structural — baked into how search worked. Google needed content to be useful. Publishers needed Google to be visible. The exchange was stable because it was mutual.
AI Overviews changed one variable: Google no longer needs to send you anywhere to answer your question.
The content still comes from somewhere. The traffic no longer returns.
What AI Overviews Did to Click Rates
The first major impact data emerged from Sistrix's 2024 analysis of click-through rates in search results before and after AI Overviews appeared. Queries where AI Overviews were present saw organic CTR drop between 34 and 65 percent depending on category. Health and how-to queries — historically reliable traffic drivers — took the hardest hits. The summary answered the question. The publisher's page stayed empty.
Search analyst Rand Fishkin at SparkToro had been documenting zero-click search trends since 2019, but the AI Overviews rollout sharpened the pattern. By late 2024, an estimated 58 to 60 percent of all Google searches in the US ended without a click to an external site — not because users failed to find what they wanted, but because they found it without leaving.
The irony is in the sourcing. AI Overviews synthesize answers from publisher content. The New York Times, WebMD, Stack Overflow, niche recipe sites, small-business how-to pages — their text populates the answers. The traffic they historically generated by writing that content no longer flows back to them.
Cloudflare's 2024 transparency report documented the parallel problem: AI company crawlers collectively accounted for a significant and rising share of bot traffic across the web. These crawlers extract content for training and inference. They do not send users back. The extraction runs on content created under the assumption of a different deal.
The Extraction Model Has a Deadline
The economics matter because content is not self-generating. The web's store of useful, accurate, nuanced writing was produced by people who expected readers — or, more precisely, who expected the traffic that meant they could afford to keep writing.
That flywheel is seizing.
Publishers including Condé Nast, Dotdash Meredith, and numerous independent outlets have shed editorial staff since 2024, citing collapsing organic traffic and advertising revenue tied to it. Independent writers who built audiences on long-tail search traffic are reporting search-driven revenue declining 30 to 50 percent year-on-year, a figure consistently appearing in communities like Mediavine and Raptive where creator economics are tracked.
This isn't a technology problem. It's an incentive problem. If producing high-quality, accurate, well-sourced content no longer generates enough traffic to sustain the people who produce it, those people find other work. The content stops being made. The corpus AI search draws from stops updating.
What replaces it is already visible: SEO-optimized content written specifically to satisfy AI crawlers, not to inform readers. Thin, formulaic, structured for extraction. The feedback loop produces exactly the information environment that AI search claims to be an improvement on.
Who Gets to Set the New Terms
Google has not framed AI Overviews as a contract renegotiation. The product is described as a better experience for users. And for individual users, in the short term, it is. Getting a medical answer without clicking through three ad-heavy pages is better. Getting a recipe without scrolling through a thousand words of personal narrative first is better.
The problem is that user experience and ecosystem health operate on different time horizons. The experience improves now. The ecosystem degrades over years.
Some publishers are attempting to negotiate new terms: opting content out of AI training, requiring licensing agreements before content can be summarized. The New York Times sued OpenAI in late 2023. Several publishers have struck licensing deals with Google for Gemini training data. These are individual arrangements — not a restored contract, but a series of bilateral transactions.
What's structurally missing is a mechanism for publishers to participate in the value AI search creates from their content. A click was worth very little in isolation — a fraction of a cent in programmatic advertising for most traffic. But it was the thread connecting content creation to content consumption economically. Cut that thread while continuing to extract the content, and you're describing a model where the open web subsidizes Google's AI products indefinitely, without any mechanism of return.
The web was never a public resource. It was a network of private contributions, held together by shared incentive. The incentive just changed — unilaterally, at scale, without acknowledgment.
When the content stops being made because the economics disappeared, what will the AI summarize?
The structural shift in how search works is one dimension of a broader pattern: AI tools changing what gets built and who can afford to build it. The question of who bears the maintenance cost is the same question, in different form.
Photo by Click Jeth via Pexels.