Share of Answer: The Next Currency of Attention

The unit changes again

Every era of the attention market has needed its own priced unit, and each one arrived only after the previous unit stopped working. The verified reader, a circulation figure an outside auditor would sign off on, was the unit that let American advertisers buy print space with confidence starting in 1914. The rating point, a share of a broadcast audience a panel could measure, became the unit that let radio and television time be bought and sold at national scale. The click and the impression, priced in real time through programmatic exchanges, became the unit of the web for roughly thirty years. Each of those units eventually needed an institution or a market to make it trustworthy enough to trade.

The current transition is to a fourth unit, and it has a name. Generative engine optimization, GEO, is defined in the reference literature as the practice of structuring digital content and managing an online presence specifically to improve visibility inside AI-made responses, a discipline distinct from, though related to, answer engine optimization and artificial intelligence optimization. The term itself did not exist before generative AI systems were built into mainstream search and information-retrieval products. It was coined specifically as a response to that integration, which marks it as a genuinely new discipline rather than a rebrand of search engine optimization wearing a new label.

That newness is now visible in how the older discipline treats it. Mainstream search-engine-optimization practice, as documented in the canonical reference literature on the field, now explicitly names AEO and GEO as adjacent, model-targeted disciplines that a publisher has to practice alongside traditional SEO, not instead of it. The AI-answer channel is being treated as a parallel distribution surface with its own requirements and its own spend, sitting next to the search-results page rather than replacing it outright.

The pillar this article sits under, measurement and honesty, is not incidental to the story. Every priced unit in this history has needed two things before a market would treat it as real: a method precise enough to be checked, and an institution, or an industry, willing to check it in public. Circulation earned that starting in 1914. The rating point earned it over the decades that followed. Share of answer has earned neither yet, which is exactly why it belongs at the start of its own history rather than at the end of anyone else's.

The click that stopped happening

The mechanism behind all of this has a name too. A zero-click result is what happens when a search engine resolves a user's query directly on the results page, without requiring a click through to any source website. It is the structural hinge on which the whole shift turns, because the entire referral-traffic economy, the model that funded a generation of digital publishing through the CPM-priced banner and the real-time-bid exchange, assumed that satisfying a reader's question required sending that reader somewhere. An AI-made answer that resolves the question on the spot breaks that assumption. There is nothing left to click, so there is nothing left to serve an ad against, and a page that would once have earned a visit now earns, at best, a citation with no traffic attached to it.

The exchange layer that priced the click for those thirty years was itself a machine built on volume. A single page view could be auctioned in milliseconds to whichever advertiser would pay the most for that particular reader at that particular moment, and the entire architecture assumed an endless supply of pages carrying an endless supply of impressions to sell against. An AI-made answer that resolves a question without opening a page removes the inventory the auction needed in the first place, not merely the price the auction would once have set for it.

It is worth being precise about what is genuinely new here and what is not. Zero-click search predates generative AI by a considerable margin. Google's own results-page features, knowledge panels and featured snippets among them, were already resolving a large share of queries without a click long before any large-language-model-native answer engine existed. Read against that history, GEO and AEO are best understood as an acceleration of a trend that was already running, not as an entirely new rupture arriving out of nowhere. The referral-traffic economy has been under this particular pressure for well over a decade. What generative AI added was scale, fluency, and a conversational interface that makes the zero-click answer feel complete in a way a snippet box rarely did.

The commercial response to that pressure is now visible in how content gets budgeted. The reference literature on zero-click search describes publishers and marketers structuring content specifically for inclusion in generative AI outputs as a distinct practice area from traditional content marketing, separate in method and, increasingly, separate in spend. Read plainly, that is a description of a budget splitting in two: money aimed at a human reader who might click, and money aimed at a model that will summarize and may or may not name the source it summarized. The two audiences do not always want the same sentence, and a publisher now has to decide how much of its content budget serves each one.

Building the meter

Every prior priced unit eventually got an institution to measure it, and the instinct is repeating here, a century after the Audit Bureau of Circulations first did it for the printed reader. A distinct commercial GEO tooling market has already formed around the question of who a model names and how often. Otterly.ai, founded in 2024, monitors and helps improve a brand's presence inside large-language-model outputs, and Writesonic has built a comparable GEO and AI-visibility platform alongside it. Neither company existed, in this form, before the discipline they serve had a name. Their existence is the evidence that share of answer is being turned, in real time, into something an agency can sell and a client can be shown, the same commercial pressure that once built a circulation audit and, later, a national ratings panel.

The signal that this has stopped being an experiment is who has picked it up. Muck Rack, an established media-intelligence and public-relations platform with a long history of pitching journalists and tracking coverage, has added generative engine optimization as a core product category, describing it as a strategy for shaping how generative AI systems interpret and present a brand or a client. A category that a mainstream media-relations platform builds a standing product around is a category that has moved from a tactic a specialist tries to standard commercial infrastructure a client budgets for annually.

What the industry does not yet have is the audit. A circulation figure could be checked against a pressroom's own paper stock and a Nielsen rating against a fixed panel methodology, both processes an outside body could inspect and certify. A large language model's output is not fixed in the same way. It is non-deterministic, meaning the same question can return a different answer, naming a different set of sources, depending on the moment it is asked and the version of the model doing the answering. That makes share of answer structurally harder to audit with consistency than a circulation count or a rating-panel figure ever was, and it is an open, unresolved measurement problem for the entire nascent GEO industry, not a detail to be fixed by better tooling alone.

None of the earlier audits arrived quickly either. The Audit Bureau of Circulations took years to earn a publisher's full cooperation and decades to become the standard assumed outside the country that built it, and the ratings panel that followed it went through its own long argument over sampling method before advertisers trusted the number enough to buy against it. Measured against that timeline, a GEO tooling market that is only a few years old is not late. It is simply still at the stage where the method is being argued over rather than settled, the same stage circulation and ratings both passed through before either became dependable, unremarkable infrastructure.

Who owns the answer

The economic story has a geopolitical twin, and it follows the same shape the earlier eras of this history traced. A circulation audit needed an American institution with the standing to enforce it, and a rating panel needed an American company with the reach to run it nationally; both put a form of market power in American hands decades before the rest of the world built comparable bodies of its own. The same pattern is now repeating one layer up. A handful of American laboratories, OpenAI, Google, Microsoft, and Anthropic among them, and the foundation models they train, are the ones deciding which sources get retrieved, summarized, and named inside an AI-made answer. That decision happens inside a model, not on a public rate card.

This is the same concentration pattern that gave the American ad-tech exchanges control of the real-time-bid pricing layer for the better part of thirty years, arriving again at a new layer of the stack. A small number of firms once set the effective price of a click across most of the open web because most of the exchanges clearing that price sat in American hands. The same handful of countries, in practice the same handful of firms, now sit at the point that decides whether a business, a publisher, or a claim gets named at all when a reader asks a question. Control of the dominant medium of an age has repeatedly meant control of that age's market and, at scale, a real measure of its geopolitical weight. The identity of who builds the answering model is not a technical detail sitting outside that history. It is the latest chapter of it.

None of this requires bad intent to be worth naming as a problem. A laboratory tuning a model to give a useful, accurate answer is not conspiring against any particular publisher when its citation choices happen to favor large, well-structured sources over small ones. The concentration is a byproduct of scale, of who has the compute and the data to build a model people actually use, in much the way the concentration of the earlier ad exchanges was a byproduct of who had built the fastest auction rather than a plot against small publishers specifically. The effect on the market is the same either way, and the effect is what a market for attention has to answer to.

The opacity is, if anything, a step beyond the earlier eras. ABC published its audit method and could be inspected by any member that doubted it; a Nielsen panel's sampling methodology was, eventually, a matter of public documentation and industry negotiation. A model's citation behavior is neither published as a method nor open to a standing industry board the way either of those precedents was. There is no rate card for a citation, no appeals process for a source left out, and no independent body with the standing ABC or Nielsen eventually earned. The chokepoint exists; the institution built to hold it accountable does not yet.

What liberates, what concentrates

Every medium in this history has cut both ways, and this one is no exception. The liberating case is real: a small publisher or a specialist business that never had the budget to buy a top ad placement or the scale to build a large owned audience can, in principle, be named inside an AI answer purely because its content answered the question clearly and was structured in a way a model could parse and trust. That was never true of a paid search auction, where the highest bidder, not the clearest answer, generally won the slot. A citation inside a generated answer is not, at least in theory, for sale in the same way an ad impression was.

The concentrating case is just as real, and it sits on the other side of the same mechanism. Because the citation decision is made privately, inside a model whose training and ranking behavior is not published as an auditable method, the businesses and publishers being judged have no visibility into why they were named, why a competitor was named instead, or what would change the outcome. A paid auction was mercenary but legible: an advertiser could see the price and choose to pay it. A model's citation is neither. The same shift that removes the toll of an ad auction also removes the map of how to be included, and a market with no map concentrates influence in whoever can afford to guess well, or whoever is close enough to the model-maker to be told.

The pattern repeating

Read across the whole span of this history, the throughline is not the technology of any single era. It is that whoever controlled the dominant medium of an age controlled the price of attention inside that medium, and that pricing power translated, reliably, into a measure of economic and geopolitical weight that outlasted the medium itself. Print circulation built American publishing houses and, through them, an advertising economy other countries had to answer to. The rating point built the broadcast networks and, later, the cable and streaming firms that inherited their audience-measurement infrastructure. The click and the impression built the ad-tech exchanges that, for thirty years, cleared the price of nearly every banner and every search result on the open web. Each unit needed a market and, eventually, an institution to make it trustworthy.

What is established is the direction: the priced unit has moved from the click to the citation, the referral-traffic economy that funded digital publishing for a generation is under real and worsening pressure, and a commercial industry has already formed to measure and sell the new unit. What is still open is whether that industry gets its own version of the audit, an institution with the standing to inspect a model's citation behavior the way ABC once inspected a pressroom's paper stock, or whether the answer stays a private decision made inside a handful of laboratories with no outside eye on it at all. Both are plausible readings of where this goes, and neither should be stated as settled. What can be stated is that a business, a publisher, or an institution that is not readable to the systems now deciding who gets named is absent from the answer, whatever the eventual shape of the market that measures it.