Attention Merchants, Then and Now: The Attention Economy and AI Search Ads

The attention economy repeats itself

Tim Wu's The Attention Merchants is not a book about the internet. It is a history of a business model. Wu documents how, starting with the New York penny papers of the 1830s, a specific commercial idea took hold: gather a large audience with cheap or free content, then resell that assembled attention to advertisers. The reader was no longer only the customer. The reader became the product being sold.

What makes the book useful now is its pattern claim. Wu shows the same arc playing out over and over as new media arrive: a medium emerges, often idealistic and ad-free, wins a mass audience, and is then progressively industrialized to harvest and monetize that audience's attention. Radio in the 1920s, network television at midcentury, and the commercial web in the 2000s each followed the shape. The medium was new every time. The move to sell its attention was not.

Read that way, the arrival of a new dominant surface is also a forecast. Whatever medium a population's attention concentrates on next becomes, in Wu's account, the next terrain a market will form to buy and sell. That is the lens this article applies to AI search.

Why advertising always follows attention

The pattern is not a coincidence or a conspiracy. It rests on an economic fact stated clearly decades before the web existed. In 1971 the Nobel laureate Herbert Simon wrote that in a world rich with information, the binding constraint is not information but attention: a wealth of information creates a poverty of attention, and that scarce attention has to be allocated among an overabundance of things competing for it.

Once attention is understood as the scarce resource, the rest follows. Where a scarce, valuable resource concentrates, a market forms to price and trade it. Michael Goldhaber named this directly in 1997, arguing that as life moves online, attention rather than information becomes the actual currency of the network. Thomas Davenport and John Beck carried the same idea from net-culture and the academy into mainstream business strategy in 2001, framing attention as a currency every organization must both capture and ration.

So the sequence Wu documents has an engine underneath it. A new medium concentrates attention. Attention is scarce and therefore valuable. A market forms to resell it. Advertising is simply the mature form that market takes. When you see attention moving to a new surface, you are watching the early stage of a process whose later stages are well understood.

The pattern, medium by medium

Wu's history is worth seeing as a sequence rather than a single claim, because the repetition is the argument.

• The penny press (1830s): newspapers priced below the cost of production, made viable by selling the resulting readership to advertisers. The template for everything after.
• Commercial radio (1920s): a medium that began as a hobbyist and public-interest technology, rapidly reorganized around sponsored programming and the advertising break.
• Network television (1950s onward): the most efficient attention-capture machine yet built, with programming engineered around holding an audience between commercials.
• The open web (2000s): search and social platforms offered free utility at scale, then monetized the attention that utility gathered through targeted advertising, the dominant business model of the modern internet.

Attention is migrating into AI answers now

The reason Wu's pattern is worth revisiting in 2026 is that the precondition for it, a large migration of attention to a new surface, is now measurable rather than speculative. The Reuters Institute's Digital News Report 2026, the most rigorous cross-national media-diet dataset available for the news slice of the picture, records two shifts at once.

First, weekly use of AI chatbots for news rose from 7% in 2025 to 10% in 2026. That is still a minority behavior, but a fast-moving one, and it is the leading edge of buyers taking questions they used to type into a search box to a conversational engine instead. Second, and more broadly, social media and video platforms overtook all other sources to become the single most-used way to access news globally, at 54%, with over half of 18 to 24 year olds now naming social, video, or AI as their main news source.

A third figure matters for what comes next. Global trust in news fell to 37% in the same report, the lowest since the series began in 2015. A migrating, lower-trust attention environment is precisely the condition under which a synthesized, authoritative-sounding answer becomes the format people act on. That is the surface where attention is now concentrating, and by Wu's logic, where a market to monetize it will form.

What Wu's history predicts about AI search ads

Apply the pattern forward and the prediction is straightforward, and worth stating carefully as a prediction. AI answer engines such as ChatGPT, Perplexity, Gemini, and Google's AI answers are, right now, in the phase Wu would recognize as the useful, largely ad-free stage: they gather attention by being genuinely helpful. If the historical pattern holds, that stage does not last. The economic pull described above, scarce attention concentrating on a new surface, is the same pull that eventually turned free radio, free television, and free search into advertising businesses.

The concrete form this is likely to take is an ad load inside the answer itself: sponsored placements, paid inclusion, or promoted sources woven into the synthesized reply, the way paid results sit above organic ones in classic search. When that happens, the economics of being named in an answer change. Today an unpaid citation inside an AI answer is earned through relevance and authority. In an ad-supported answer surface, some share of that visibility becomes purchasable, and the earned slots get scarcer and more contested.

This is the part to label clearly. The historical pattern is established: Wu's evidence for it spans two centuries and multiple media. The application to AI surfaces is emerging: it is a reasoned extrapolation, not yet a documented outcome with published figures on AI-answer ad load. Exactly how fast or how heavily AI answers will commercialize cannot be stated with precision yet. What the pattern licenses is not a date. It is a direction, and a reason to act while the earned window is still open.

The caveats: what we do not yet know

A rigorous version of this argument has to hold its own uncertainties in view, because the history of predicting search and media has been a history of confident timelines that missed.

The window between migration and monetization

Put the established history and the caveats together and a strategic reading falls out. Every previous medium had a window between the moment attention arrived and the moment the surface was fully monetized. In that window, presence was cheap and earned. Early organic search rankings and early paid search inventory were both far cheaper to win than they became once the surface matured and competition and ad load rose. The businesses that established a position early carried an advantage into the expensive phase.

AI answer surfaces are plausibly in that window now. Being cited in an AI answer today is earned through entity clarity, corroborated authority, and content a machine can extract and trust, not bought. If Wu's pattern holds and paid formats arrive, that earned position becomes more valuable and harder to claim. The asymmetry is the whole point: the cost of establishing an AI-answer presence is lower before commercialization than after, and the businesses reading the pattern correctly are the ones acting inside the window rather than waiting for proof that closes it.

None of this is a guarantee about timing, and it should not be sold as one. It is a bet on a direction with two centuries of precedent behind it, sized against the acknowledged uncertainty about speed. The rational response to that combination is not to predict the date. It is to measure your standing now and improve it while improvement is cheap.

How to read your own position

Because there is no published per-vertical dataset, the pattern only becomes actionable through direct measurement of your own market. The relevant question is not "will AI search carry ads someday" but "when a buyer asks an AI engine the question that leads to my kind of business, am I named in the answer, and is that holding or slipping over time." That is a measurable quantity, sampled across real buyer questions and read repeatedly, and it is the starting point before any work is scoped.

Reading it now does two things. It tells you whether the migration has already reached your buyers, which no average can tell you, and it establishes a baseline you can defend later, when the surface is more contested and every position costs more to hold. In an environment the history says is about to be commercialized, knowing exactly where you stand is not a nice-to-have. It is the input that decides whether you act inside the window or after it.