For three decades, the web ran on a relatively simple exchange: someone produced information, search engines helped people find it, and a share of readers made their way to the source. Artificial intelligence is rewriting that arrangement. Not necessarily for the worse, but in a way that forces us to reconsider who creates value, who captures it, and who pays to sustain it.

The Business Behind an Answer

Consider an ordinary query. Someone wants to know why prices are rising, how a vaccine works, or what distinguishes two kinds of computer storage.

Until recently, the usual course was to turn to a search engine, scan a few results, and visit one or more pages. A university, a newspaper, an independent researcher, or a company had produced the content. The search engine brokered the encounter between the person asking and the party able to answer.

The arrangement had obvious flaws. Pages competed for the top spot, content written to please algorithms proliferated, and many sites turned reading into an obstacle course of advertising. Still, there was a recognizable economic relationship: producing information could generate visits, reputation, subscriptions, advertising revenue, or sales.

With generative artificial intelligence, a growing share of those queries ends differently. The user asks and receives a finished answer. There is no longer any particular need to visit the pages from which the information came.

The experience may well be better. The answer can bring together scattered explanations, adapt to the reader’s level, compare sources, and invite follow-up questions. The technological improvement is real. So is the economic transformation it sets in motion.

Information still costs something to produce, even as consuming it feels increasingly free and instantaneous to the user.

That is where the problem begins.

The Evidence: Answers Are Replacing Some of the Visits

There is no need to speculate about some distant future. There are measurements that let us observe the change.

In July 2025, the Pew Research Center published a study of how Google users behave when presented with AI-generated summaries. The researchers analyzed searches conducted by a panel of American adults.

When an AI summary appeared, users clicked on a traditional result in 8 percent of visits. When none appeared, they did so in 15 percent. The difference is considerable: the observed rate fell by nearly half.

The same study found that links embedded within the AI summaries themselves were clicked in just 1 percent of the visits that displayed them.

This does not mean that every AI-assisted search produces an equivalent loss of traffic. The percentages describe particular observed sessions, not a universal law of user behavior. Nor do they allow us to attribute every change solely to the presence of AI: the queries that trigger these summaries may differ from the rest.

But the finding is significant enough to show that the traditional economic mechanism is under strain (Pew Research Center, July 2025).

The evidence does not end there. The Digital News Report 2026, produced by the Reuters Institute at the University of Oxford, cites Chartbeat data covering more than 2,500 news sites. Between November 2024 and November 2025, traffic from Google’s organic search fell by 33 percent globally and by 38 percent in the United States.

It is a substantial decline, though it would be wrong to attribute it entirely to AI. Changes to search algorithms, shifting consumption habits, competition from other platforms, and the broader evolution of the media all play a part.

The figure demonstrates a drop in traffic to the sites studied; it does not, on its own, show how much of that drop artificial intelligence caused.

The publishers surveyed by the Reuters Institute also expect search-referred traffic to fall by 43 percent over the next three years. That last figure is an industry expectation, not a verified forecast (Reuters Institute, Digital News Report 2026).

The distinction between what has been observed and what is expected matters. Technological diagnoses tend to overreach when they turn an early trend into an inevitable destiny. There is no need to do so here. The measured changes are significant enough.

Google’s case is especially revealing because the company occupies two positions at once: it is one of the chief beneficiaries of the traditional model and one of the actors transforming it most rapidly.

For many years, its advertising business exploited an economic feature of search: the user expressed an intention, and Google could display ads related to it.

A search for hotels, loans, computers, or insurance revealed valuable information about a possible purchasing decision.

Advertisers paid to appear at that moment. The search engine earned revenue, and the sites that received visits could convert some of them into economic activity. Artificial intelligence changes the interface of that relationship.

When a query becomes a conversation, the answer can stretch across several exchanges. The user compares options, asks for explanations, requests recommendations, and may eventually reach a decision without ever leaving the system.

The platform no longer merely ranks possible answers. It takes part in constructing the answer and, increasingly, in the decision that follows. Commercially, this opens enormous opportunities. It also introduces difficulties.

In its third-quarter 2026 results, Alphabet reported roughly $63.3 billion in revenue from Google Search and related advertising, a 12 percent increase year over year. The business was still growing, and growing a great deal.

Yet growth had slowed from 19 percent in the first quarter and 17 percent in the second.

The company acknowledged that lower monetization of queries made through AI Mode contributed to the slowdown. It also cited other factors, including currency effects and changes in certain advertising categories. It did not attribute the entire slowdown to AI.

The contrast is telling: according to Google, query volume was growing at its fastest annual pace in five years, yet some of the new experiences generated less revenue per query than traditional search (Alphabet, third-quarter 2026 results and Search Engine Land, October 2026).

We should resist a hasty conclusion. Google has not stopped making money, and there is insufficient evidence to claim that its advertising business is doomed.

In fact, the company is expanding ads within its AI search experiences and maintains that conversational queries allow it to identify new commercial intentions. It can also profit from computing infrastructure, enterprise services, subscriptions, and other AI-related products. What emerges, then, is not necessarily a crisis for Google.

It is a transition between two ways of capturing economic value.

And a company able to participate in both can weather it far better than those that depend on only one.

Reddit and the Economic Value of Human Conversation

Reddit offers another view of the same phenomenon.

Much of its appeal derives from something AI models do not produce on their own: original conversations among people.

Consumer experiences, technical discussions, recommendations, testimonials, disagreements, and answers that do not necessarily appear in formal documents.

That material is valuable to users, but also to systems that seek to understand how people talk, compare, and reason.

For years, Reddit benefited from the visitors Google sent its way. Search engines surfaced its conversations and directed new audiences toward them. As AI-generated answers have grown, the balance has shifted.

Reddit’s content can help build an answer without the user ever needing to visit Reddit.

The company has responded with a combination of data-access agreements, restrictions on crawlers, and greater insistence that users sign in.

But there is an important nuance: Reddit is not simply undergoing an economic collapse. In the third quarter of 2026, it recorded roughly 130.3 million daily active users worldwide, up 12 percent from a year earlier. In the United States, by contrast, the figure fell 2 percent year over year, to 50.7 million.

The divergence between global growth and American decline rules out any talk of a uniform downturn across the platform (Reddit, third-quarter 2026 results).

The more interesting question is not whether Reddit makes or loses money in a given quarter.

It is that its content can now be monetized in two different ways: through users’ direct attention, or through agreements with platforms that need access to its data.

The traditional model consisted of drawing people toward the conversations. The emerging model also makes it possible to charge for supplying those conversations to systems that answer elsewhere.

Content ceases to be merely a destination and becomes a commercial input.

The difference may seem small. Its consequences are not.

From Traffic to Licensing: Another Market Takes Shape

The transformation can also be seen in the agreements between AI companies and content producers.

In May 2024, OpenAI and News Corp announced a multiyear partnership allowing the use of current and archival material from publications such as The Wall Street Journal, Barron’s, MarketWatch, and The Times, among others.

The agreement covers the use of that content to improve AI products and to present information in response to users’ queries.

The financial terms were never officially disclosed. Several outlets reported that the deal could exceed $250 million over five years, a press figure that should not be treated as a confirmed contractual amount (OpenAI and News Corp, official announcement).

Arrangements of this kind herald a new economic possibility: charging for access to content without depending solely on readers reaching the site that produced it.

This is not the first time data, documents, or distribution rights have been sold. News agencies, financial databases, and academic publishers have been doing it for decades.

The novelty lies in the potential scale of the market and in the use to which the content is put.

A platform can acquire access to thousands of publications to feed a system capable of answering millions of queries.

Here an important asymmetry appears. Large publishing groups can negotiate licenses. They have extensive archives, recognized brands, and legal resources.

A small regional outlet, an independent researcher, a teacher who publishes educational materials, or a specialized site is unlikely to have the same bargaining power.

A broader licensing market may develop, with intermediaries and compensation mechanisms accessible to small producers. But we do not yet know whether that market will distribute revenue in a reasonably balanced way.

Nor do we know whether licensing will replace the losses in advertising, subscriptions, or sales that follow from declining visits.

The existence of a new market does not guarantee that its benefits will reach those who produce the content.

The New Scarcity: Reliable Information and Access to It

There is an economic paradox in all of this.

Artificial intelligence drastically reduces the cost of generating texts, images, summaries, and explanations. Producing a page that resembles an article has never been easier. But that abundance is not an abundance of reliable knowledge.

Investigative journalism requires sources, verification, and accountability. A scientific experiment requires methodology, data, and the possibility of replication. A good technical text demands an understanding of the problem, not merely fluency in presenting it.

AI can help with all of these tasks. It may even lower their costs considerably. But it does not automatically erase the difference between generating a claim and establishing that it is true.

The economic consequence may be unexpected: as generic content grows cheaper, original, verifiable, and up-to-date information may acquire greater relative value. Not necessarily because every reader is willing to pay for it.

It may become valuable to systems that need it as an input in order to offer better answers than their competitors.

The Reuters Institute provides a relevant data point on the reach of the phenomenon. Its Generative AI and News Report 2026, which examines several markets, found that weekly use of generative AI systems rose from 34 percent to 51 percent between 2025 and 2026 in the six countries with comparable data.

It also found that 57 percent of respondents across the countries studied had encountered AI summaries built into search engines in the previous week (Reuters Institute, Generative AI and News Report 2026).

The technology has ceased to be a marginal tool and has become a new everyday interface for accessing information.

That does not mean users will stop consulting original sources. It does mean that more and more decisions about which information to display, synthesize, or highlight are being made inside intermediary systems.

Scarcity is beginning to shift: there is no shortage of text; what is scarce are guarantees of quality, provenance, timeliness, and accountability.

The Concentration Problem: When Closing the Web Benefits the Giants

Economic theory offers an explanation of why a less open Internet could benefit the dominant platforms.

Suppose that legally accessing quality information requires negotiating licenses, paying for databases, maintaining computing infrastructure, and sustaining legal teams.

For a company with hundreds of billions of dollars in revenue, those costs may represent a considerable but manageable investment.

For a small company trying to build a search engine or a specialized assistant, they could become barriers to entry.

The more expensive it becomes to access the data needed to compete, the greater the advantage of those who already have capital, infrastructure, and commercial relationships.

It is not merely a matter of economies of scale. Economies of scope also come into play: a single company can use the same infrastructure, models, and data to offer search, advertising, cloud services, productivity tools, e-commerce, and personal assistants. The cost of an investment can be spread across many products.

Moreover, an integrated platform can observe more interactions and gather useful information about what its users need, always within the applicable technical, contractual, and regulatory limits. This could reinforce dominant positions.

But we must be precise: the existence of scale advantages does not prove that concentration is inevitable.

Open models, lower-cost technologies, public knowledge bases, interoperable protocols, and competition regulation can lower some of these barriers. The contest is not settled.

What does seem clear is that access to content and to processing capacity is becoming a strategic dimension of competition.

The openness of the Internet is no longer a purely cultural question; it now has direct consequences for the structure of markets.

The Tension Between Individual Benefit and Collective Sustainability

Suppose someone needs to understand a legal concept.

They can read several articles, compare interpretations, and construct an answer. Or they can consult an AI assistant that offers an integrated explanation, complete with references and examples.

If the second option provides a correct, comprehensible, and sufficiently complete answer, it is reasonable to prefer it.

There is no moral obligation to choose the less efficient procedure in order to sustain a particular business model.

The problem arises when millions of perfectly reasonable individual decisions alter the incentives to produce information. The user saves time. The producer loses a visit. The platform keeps the interaction.

If visits were financing the production of content, the economic incentives that sustain that content could begin to erode.

In economics, situations of this kind call for the study of externalities: costs or benefits that are not fully reflected in the individual decisions of those taking part in a transaction.

Even so, it remains to be shown that reduced compensation through traffic will not be offset by new sources of revenue or by lower production costs.

Some content may disappear because it is no longer economically viable. Other content may appear precisely because AI tools make it possible to produce with resources that were previously insufficient. Both processes can occur at once. The challenge is to understand the net balance and, above all, how it is distributed.

A greater quantity of available information does not necessarily mean a greater diversity of independent sources.

Nor does a smaller number of visits automatically imply a lower quality of knowledge produced. These relationships must be investigated, not presumed.

The Attention Economy Does Not Disappear: It Changes Hands

Herbert Simon observed in 1971 that a wealth of information consumes what information requires of its recipients: attention.

For much of the Internet’s history, the competition for that attention organized its business models.

Sites needed to attract users, keep them reading, and bring them back. Search engines competed to order information. Social networks developed mechanisms to prolong engagement and stimulate interaction.

Artificial intelligence does not eliminate the scarcity of attention. It may reduce the time needed to find an answer, but it concentrates a larger share of the interaction in the intermediary that provides it. A long conversation with an assistant can replace numerous visits to different pages.

The user gains an integrated experience. The platform gains continuity in the relationship.

The difference is that the new intermediary does not merely select content: it synthesizes, reorganizes, and presents it in a voice of its own. This can be useful, but it also alters the distribution of visibility and reputation.

On the traditional web, users could easily recognize that an explanation came from a university, a scientific journal, or a personal blog.

In a synthesized answer, sources may appear as secondary references, while the platform occupies the principal role in the exchange.

The attribution of knowledge ceases to be a natural byproduct of browsing and becomes, in part, a design decision. And design decisions have economic and cultural consequences.

Not All the Evidence Points the Same Way

So far, the diagnosis seems to lead toward a pessimistic conclusion. But it would be a mistake to ignore the contrary evidence.

Google has stated publicly that the aggregate volume of organic clicks sent from its search engine to other sites has remained relatively stable, and that clicks coming from AI experiences tend to be of higher quality.

By higher quality it means, among other indicators, that users do not immediately return to the results page, which may suggest deeper interest.

The company also maintains that the new features generate queries that were not previously being made (Google, August 2025).

These claims should not be accepted without scrutiny, but neither should they be dismissed simply because they come from an interested party.

Aggregate click volume and the click-through rates observed for particular queries measure different phenomena. The two measurements may well be compatible: clicks per query may fall in certain categories while the total number of queries rises or traffic is redistributed toward other content.

To resolve the discrepancy, we would need comparable data broken down by period, category, country, query type, and kind of site.

It would also help to distinguish between clicks that generate revenue, clicks that build lasting relationships with readers, and clicks that merely produce brief visits.

A site may receive fewer visitors and nonetheless gain more subscribers or customers if those who arrive are more deeply interested.

The economically relevant question is not only how many clicks disappear, but how much value is lost, how much is created, and who is able to capture it.

Moreover, AI can benefit small producers. An independent researcher can translate their work, make their writing more accessible, and build tools that once required teams of programmers.

A small business can offer specialized service without maintaining a large commercial operation.

A teacher can create interactive, personalized materials at comparatively low cost.

That is why it is unconvincing to describe AI solely as a mechanism for extracting value. It is also a technology capable of lowering barriers to production and distribution.

The decisive question is not whether it creates value. It does. The question is how that value is distributed and which activities it stops paying for.

From Finding Information to Delegating Decisions

Until now we have focused on answers. But the deeper change may come when these systems move from answering questions to carrying out actions.

Booking a hotel today requires comparing options, checking terms, and making a reservation. An assistant capable of managing the process can turn that sequence into a single instruction.

Something similar could happen with purchases, administrative procedures, professional services, software procurement, or task management.

The platform would no longer be merely the place where a search begins. It could become the intermediary that proposes options and executes transactions.

Google has already identified agentic experiences, capable of filling out forms, booking services, and making purchases, as a line of commercial development (Alphabet, third-quarter 2026 results).

In that scenario, the competition to attract traffic could partly turn into a competition to be selected by an agent. The difference would be considerable.

Instead of designing a page to persuade a person directly, companies might need their products, prices, and terms to be intelligible and accessible to automated systems. New intermediaries, new forms of advertising, and new conflicts of interest would emerge.

If an assistant recommends a product, does it do so because the product best meets the user’s needs, because it has more complete information, or because a commercial arrangement exists? Who is responsible if it makes the wrong purchase? Can a platform systematically favor its own services?

These are technical, economic, and regulatory questions. They will not be resolved simply by making the model more intelligent.

And we should not confuse capability with adoption. The Reuters Institute study published in 2026 found that, among respondents in nine countries, only 14 percent had used an AI agent to carry out a real task in the previous week. The potential is considerable, but it is not yet a universal practice.

Who Will Pay for the Knowledge Everyone Wants to Use?

Here we arrive at the problem I consider most important. A society needs to produce information that does not always have an immediate buyer.

It needs scientific research, historical archives, investigative journalism, technical documentation, public statistics, educational materials, and spaces for debate.

Some of these goods can be financed through subscriptions or advertising. Others depend on universities, governments, foundations, donations, voluntary contributions, or companies that publish them as part of their activities.

Artificial intelligence changes the way this content can be used, but it does not eliminate its cost.

Scientific research does not happen because a system can draft a convincing summary. A corruption case is not uncovered because a model can answer questions about it. Someone has to gather the evidence, test it, and take responsibility for publishing it.

Technology can help enormously with these tasks. But we must distinguish between lowering their cost and rendering them unnecessary.

If platforms capture a growing share of the revenue associated with the use of information while its original producers lose their means of financing, a sustainability problem could arise. Not necessarily across the entire web, nor all at once.

Perhaps prestigious outlets will consolidate their subscriptions, specialists will sell personalized services, and producers of commercial information will find new ways to distribute their products.

Generic content could lose economic value, while original research and reliable data gain it. Public or cooperative funding for certain information goods might also increase. The reorganization, in short, does not run in a single direction.

But the question persists: how is the production of knowledge to be financed when consumption no longer necessarily leads back to the person who produced it?

The problem predates artificial intelligence. This technology simply makes it more visible and, perhaps, more urgent.

What Should Change

I do not believe the solution lies in trying to artificially preserve every previous business model.

The Internet was not born with an obligation to sustain particular newspapers, blogs, search engines, or platforms. Technologies change, and economic activities must adapt.

Nor does it seem reasonable to ask users to give up more efficient tools in order to protect other people’s revenue.

But there is a difference between allowing economic models to evolve and accepting any distribution of their benefits as though it were inevitable. A reasonable transition should attend to at least four conditions.

The first is competition. If access to content, models, and processing capacity requires investments that only a handful of companies can afford, the market could become more concentrated than is desirable. Open systems, interoperability, and oversight of anticompetitive practices take on new importance.

The second is attribution. Not every use of information requires compensation, but acknowledging sources and making them easy to consult helps preserve reputation, verifiability, and the possibility of economic sustainability. Attribution does not solve every problem of rights or revenue, but it is an important piece.

The third is a diversity of funding mechanisms. Advertising and clicks were never the only possible models. Subscriptions, memberships, licenses, specialized services, institutional support, and public funding can coexist. The task is to determine which of them work for each kind of knowledge, and which introduce problematic dependencies.

The fourth is transparency in intermediation. As assistants select information and take part in commercial decisions, it becomes necessary to distinguish recommendations grounded in the user’s criteria from those shaped by advertising incentives or distribution agreements.

None of this requires opposing artificial intelligence. It requires recognizing that its development is also a problem of economic and institutional organization.

The False Nostalgia for the Old Web

There is a temptation worth resisting: idealizing the Internet of links.

That system enabled an extraordinary diffusion of knowledge, but it also produced concentrated advertising markets, commercial surveillance, dependence on algorithms, and vast quantities of content designed to capture visits rather than to inform.

Misleading headlines, ad-saturated pages, and the industrial production of shallow texts were not born with generative models.

They were, in part, rational responses to the incentives of the attention economy.

Artificial intelligence can correct some of these defects. A good answer that spares us a visit to five redundant pages is a genuine improvement.

We should not mourn the disappearance of visits whose sole economic purpose was to display ads before offering some trivial piece of information.

The problem begins when the same mechanism harms redundant content and knowledge that demands original effort alike. If a technology eliminates unnecessary intermediation, it can increase welfare.

If, along with that intermediation, it also eliminates the incentives needed to produce valuable information, the result is more ambiguous. The difficulty lies in separating the two effects.

And that separation cannot be made solely from the perspective of the companies affected, because their commercial interests are not identical to the public interest.

Nor can it be made solely from the perspective of the immediate consumer, because a present benefit can coexist with a long-term collective cost.

The Real Change Is Not Technological

The phrase may sound like an exaggeration. Artificial intelligence plainly constitutes an extraordinary technological change.

But its deepest consequences may not depend solely on the quality of the models.

They will depend on who controls the infrastructure, who holds original information, who sets the terms of access, and who can sustain their activity economically in the new environment.

Economic history is full of technologies that raised productivity while reorganizing markets, professions, and institutions.

The printing press altered the material conditions for reproducing knowledge. Radio and television transformed advertising and cultural distribution. The Internet lowered the cost of publishing and shifted the power of traditional intermediaries.

Artificial intelligence introduces another shift: it allows a platform not merely to transport or locate information, but to build an answer out of it. And it is beginning to extend that capacity to recommendations and actions. Its economic significance lies in that displacement.

Before, much of the business consisted of facilitating the encounter between a question and a source. Now, a growing share consists of supplying the answer directly. The difference affects producers, distributors, advertisers, users, and competitors alike.

Conclusion: The Model Is Already Changing

We do not need a spectacular stock-market crash to recognize what is happening.

Studies show changes in user behavior. Media outlets report significant declines in certain sources of traffic. Platforms are experimenting with new forms of advertising, licensing, and subscriptions. Content producers are revising their strategies. Technology companies are investing extraordinary sums in the infrastructure needed to sustain the new system.

These are distinct processes, with multiple causes and still-uncertain consequences. But they converge on a recognizable transformation.

Artificial intelligence is changing the business model of the Internet.

And it is already being felt.

We do not yet know whether the result will be a more open, productive, and accessible web, or an ecosystem of concentrated platforms controlling access to a growing share of knowledge.

There will probably be no single outcome. Some markets will open, others will concentrate; certain activities will lose profitability even as opportunities emerge that we can scarcely imagine today.

What we can say is that the technology is shifting the boundary between producing information and capturing the value of its use.

For decades, publishing meant, among other things, offering something that others could find and visit.

In the new environment, publishing can also mean feeding systems that will answer without ever leading the reader back to the source. This is not necessarily an injustice. Nor is it necessarily economic progress for everyone. It is a change in the rules that govern how value is distributed.

The decisive question is not whether machines will be able to answer better. It is how we will ensure that it remains possible, and worthwhile, to produce the questions, research, data, arguments, and discoveries on which those answers depend.

For a society can benefit enormously from a technology that makes knowledge easier to access. But that benefit does not, by itself, guarantee the conditions for continuing to produce it.

And an Internet that answers ever better while gradually losing the capacity to renew its sources may discover, too late, that it had mistaken access to knowledge for its production.

Sources and References

  1. Pew Research Center (2025). Google users are less likely to click on links when an AI summary appears in the results. A study of user behavior when AI summaries appear in search results.
  2. Reuters Institute, University of Oxford (2026). Digital News Report 2026. Trends in news consumption, search traffic, and publishers’ expectations.
  3. Reuters Institute (2026). Generative AI and News Report 2026: From standalone tools to information infrastructure. Data on AI adoption, use for information, and the emergence of agents.
  4. Alphabet Inc. (2026). Third Quarter 2026 Earnings Call. Financial information and the monetization strategy for search and AI.
  5. Search Engine Land (2026). Google: AI Mode’s lower monetization slowed Search ad growth. Analysis of the company’s statements on monetization.
  6. Reddit Inc. (2026). Third Quarter 2026 Results. User trends and the platform’s results.
  7. OpenAI and News Corp (2024). A landmark multi-year global partnership with News Corp. An example of commercial agreements for the use of journalistic content.
  8. Google (2025). AI in Search: Driving more queries and higher quality clicks. Google’s perspective on traffic from AI-powered search.
  9. Simon, Herbert A. (1971). Designing Organizations for an Information-Rich World. A conceptual reference on the relationship between information abundance and the scarcity of attention.

A note on method. The data come from corporate reports, behavioral studies, and third-party analyses published through November 2026. The projections regarding concentration, the financing of knowledge, and market evolution are reasoned interpretations, not findings established by those studies. An observed association between AI adoption and declining traffic should not be mistaken for causal proof that the former fully explains the latter.