AI Is Eating Search, and the Web Has No Backup Plan

The old web had a bargain. Publishers, stores, forums, recipe sites, review blogs, newspapers, hobbyists, and software docs made pages. Search engines organized those pages. Users clicked through. Traffic became subscriptions, ads, affiliate revenue, brand awareness, or at least the satisfaction of being found. It was never a perfectly fair bargain, but it was legible: make something useful, get indexed, earn attention.

AI search is breaking that bargain by making the answer page feel like the destination. Google has been pushing further into generative search, describing AI as a way to handle longer and more complicated questions that do not have one obvious answer on the open web. In a CNN Business interview, Google Search product executive Robby Stein said people are asking “much longer and harder questions,” the kind that require synthesis rather than a familiar list of blue links. That is a real user need. It is also a direct challenge to the economic structure of the internet.

The business problem is simple: if an AI system reads the web, summarizes the web, and keeps the user inside its own interface, the value moves upward. The model provider gets the session. The search platform gets the data. The assistant gets the trust. The original publisher may get a citation, a tiny referral, or nothing at all.

This is not just a media problem, though media companies feel it first. Search traffic has long been the oxygen supply for huge parts of the digital economy. A small software company depends on documentation pages being discoverable. A travel site depends on itinerary searches. A medical publisher depends on symptom and treatment queries. A local business depends on people searching for services nearby. Even e-commerce depends on the messy middle of research: “best running shoes for flat feet,” “which laptop for video editing,” “how much solar do I need for a small house.” Those are exactly the kinds of questions AI systems are built to absorb.

For users, the change can feel wonderful. Instead of opening ten tabs, skimming search results, dodging pop-ups, rejecting cookie banners, and trying to guess which review site is secretly an affiliate funnel, an AI answer can produce a calm paragraph in seconds. It can compare options, explain trade-offs, and translate jargon. Google’s own explanation of generative AI in Search frames this as a way to “take more of the work out of searching.” That promise is powerful because searching has, in many corners of the web, become work.

The trouble is that the friction users hate is often the same friction that pays for the material being summarized. Ads are annoying, but they fund reporting. Affiliate links can be grubby, but they fund testing. Subscriptions are inconvenient, but they pay editors, photographers, engineers, and subject experts. If AI removes the visit while preserving the answer, it does not merely improve search. It changes who gets paid for knowledge.

The incentives are already shifting. Publishers have begun negotiating licensing deals with AI companies, blocking crawlers, suing over training data, or trying to make their brands valuable enough that users will seek them directly. Some will survive by becoming premium destinations. Others will become raw material. The open web’s weakest players are not necessarily the least useful; they are the ones without leverage. A niche expert with the best repair guide for an old appliance, a local journalist covering a city council, or an independent reviewer with years of testing notes may not have the lawyers, scale, or brand power to demand payment when their work becomes part of a machine-generated answer.

AI search also changes competition among the largest platforms. Google’s traditional search business was built around intent. A user typed a query, Google matched it with information and ads, and the user often left for another site. In an AI interface, the platform can keep more of the journey. It can answer the question, refine the question, recommend a product, compare prices, generate a chart, book a trip, or draft the email that follows. That makes AI search less like a directory and more like an operating system for decisions.

This is why the battle is not only Google versus OpenAI, or search versus chatbots. It is a fight over the default layer between people and the internet. ChatGPT, Perplexity, Gemini, Copilot, Claude, TikTok, Reddit, Amazon, and traditional Google Search all want to be where a question begins. The winner does not need to own all the information. It needs to own the moment of interpretation.

There is a risk that the web becomes more centralized at the exact moment it appears more conversational. A page of links at least showed that the internet contained many voices, even if ranking systems shaped what people saw. An AI answer can compress disagreement into a single confident surface. It may cite sources, but the user’s relationship is increasingly with the answer engine, not the underlying writers. Google has acknowledged quality problems before, including in its post about early AI Overviews mistakes, but factual accuracy is only one part of the issue. Even a good summary can weaken the habit of visiting, comparing, and judging sources directly.

The likely future is not that websites disappear. The web is too useful, too embedded, and too commercially important for that. Instead, it may split. Large platforms and authoritative brands will make deals, optimize for AI visibility, and become preferred suppliers of trusted information. Communities with strong identity, such as Reddit forums, Discord groups, newsletters, and specialized professional sites, will keep value because people go there for context and belonging, not just answers. Transactional businesses will adapt because buying, booking, and hiring still require action. The vulnerable middle is the vast layer of informational publishing that was built for search traffic and monetized by visits.

That middle layer includes much of what made the web useful. The internet was never only a library. It was a market of explanations, arguments, tutorials, tests, essays, databases, and personal obsessions. Search helped strangers find those things. AI can make them easier to consume, but if it drains the reward for producing them, the system starts eating its seed corn.

The most important business question is therefore not whether AI search is better for users in the short term. Often, it will be. The question is whether the companies building answer engines can create a durable economy around the information they consume. Citations are not a business model. Tiny trickles of referral traffic are not a replacement for a search ecosystem. Licensing deals help the powerful, but they do not automatically sustain the broader web.

A healthier version of AI search would treat sources as partners rather than scenery. It would send meaningful traffic when users need depth, make attribution unavoidable rather than decorative, share revenue where summaries substitute for visits, and give publishers clear controls over how their work is used. That is technically possible. Whether it is commercially attractive is another matter.

The web’s original promise was that publishing could be open, searchable, and economically viable enough for many kinds of people to contribute. AI search does not have to destroy that promise, but it does force a reckoning. If the answer engine becomes the main interface to knowledge, then the business model behind knowledge has to change too.

Otherwise, the internet may get faster, smoother, and emptier at the same time.