Canonical: https://thonly.org/research/the-appreciation-economy · Licence: CC0 1.0
I had a sentence in my head for a while, and I liked it: the modern economy is largely driven by advertising. It felt obviously true. Then I went to check the number, the way I have been trying to make myself do, and the number said something else.
Global advertising revenue reached roughly $1.1 trillion in 2025, against a world economy of something like a hundred and ten trillion. Call it one percent. The United States runs higher — depending on whose series you use and what they count, somewhere between about one and a half and two percent of GDP. And the long-run series is the part that stopped me: across roughly ninety years of data, through radio, through television, through the entire internet, the advertising share of the American economy shows no trend increase. It moves in a band. It does not climb.
So my sentence was wrong. But the number that refuted it turned out to be more interesting than the sentence was, and it took me a few days to see why.
One percent is not the economy's mass. It is the economy's steering.
Advertising is how a seller gets to be considered at all. It is the discovery layer — the mechanism deciding which of the many possible sellers a person ever hears about. And ninety years of flat share means something specific and slightly unsettling: the attention economy did not grow the pie. It re-routed who gets steered toward whom, using roughly the fraction of the pie it always used.
I should flag a trap here, because I nearly fell into it and someone will eventually wave it at me. There are industry studies reporting that advertising "drives" something like a fifth of the American economy. Those are measuring a different quantity — economic activity supported by or attributable to advertising, which is a modelled figure with generous assumptions, commissioned by the industry it flatters. It is not spending on advertising. The one-percent figure and the twenty-percent figure are not in conflict, because they are not measuring the same thing, and only one of them is a number you can go and count.
The flat share is what makes the discovery layer worth arguing about, because a discovery layer is not neutral about who wins it.
Modern ad discovery is an auction. An auction gives visibility to whoever bids most per acquired customer, and the party who can bid most is reliably the one with the largest lifetime value per customer, the best measurement, the cheapest capital, and the widest geography over which to spread the cost of making the advertisement. Every one of those four favours scale. None of them is anybody's malice. A single-location business is not kept out of the discovery layer by a conspiracy; it is priced out by construction, because it cannot amortise a campaign over four hundred locations and its competitor can.
This is why the circulation we see runs between people and very large companies. It is not a claim about what people prefer. It is a claim about who can afford the steering wheel.
That was the next question I asked myself, and I want to answer it carefully, because it is the place where I am most likely to believe something because I would like it to be true.
There are three different claims hiding inside it, and they do not stand or fall together.
The first is about firms, and it is plausible. If discovery is allocated by ability to pay, and ability to pay tracks existing size, then advertising is a concentration mechanism — and firm concentration is a live subject in economics. The superstar-firms literature ties rising industry concentration to the falling share of income going to labour; there is related work on rising markups and on intangible capital as a driver of concentration. None of it was written about advertising specifically, but the mechanism I described above is the kind of thing it describes.
And here is the strongest objection to my own argument, which I would rather state than have found for me. Digital advertising is usually credited with lowering entry barriers, and the credit is deserved. A corner shop can buy fifty targeted impressions. In the broadcast era, the minimum viable ad buy was a national television spot that only a handful of companies on earth could afford. By that comparison the old system was far more exclusionary than the new one.
I think the honest repair is that the barrier moved rather than fell. It used to be a minimum ticket size: you either could or could not afford to play. Now it is a per-unit price you have to be able to out-bid, and that price is competed up to whatever the best-monetising bidder in your category can pay. Access got democratised. The cost got competed up to the ceiling. If you sell one lunch to one person, you are bidding against someone who will extract a decade of subscription revenue from the same click, and you are bidding for the same second of the same person's attention.
That yields a testable pattern — concentration should rise fastest in the sectors where discovery moved earliest and most completely to auction. I am not going to pre-register it, because I cannot clean it. The sectors that moved to auction discovery first are the same sectors that were platform-intermediated first, and I have no way to separate advertising from intermediation with anything I can get at. A prediction I cannot clean does not belong on a register whose only value is that its entries are clean.
The second claim is about households, and here I think I am wrong. If advertising were a major driver of household inequality, you would expect the two to move together. They do not. Ad share of the American economy has no trend across ninety years. American inequality over the same span is a U — high before 1929, compressed through the middle of the century, rising since about 1980. A constant cannot explain a variable. Whatever is driving household inequality, the size of the advertising sector is not it, and the serious candidates sit elsewhere: skill-biased technical change, trade, the decline of unions, tax policy, assortative mating, the returns to owning assets.
So the version of my claim that survives is narrower and better: it is not the quantity of advertising that matters, it is the allocation mechanism. The share stayed flat while discovery became auction-allocated and precisely measurable. The hypothesis has to be about the mechanism getting more efficient, not about the spending getting bigger.
The third claim is the one I actually believe, and I had not seen it until I stopped borrowing other people's frameworks.
Attention is the one input in the whole economy that is distributed equally by nature. Every person alive has the same twenty-four hours. There is no way to be born with more of it, no way to inherit it, no way to accumulate it across generations. It is the most equally held asset there is.
And it is captured unequally by construction. The ad economy harvests that equally-distributed asset through a small number of concentrated owners who keep the proceeds. That is not a claim about anyone's spending habits — I want to be careful never to describe people by what they lack — it is a claim about the shape of a mechanism. An input held equally by eight billion people, converted into revenue held by a very small number of them, is an inequality engine regardless of what happens to firm concentration or to the labour share.
That is the version I would defend. And it is the one that leads somewhere, because witness is also distributed equally, and it cannot be harvested at a distance.
Here is the property I keep coming back to.
Attention can be bought and it adds up. A dollar buys an impression anywhere on earth; impressions accumulate; and anything that accumulates concentrates, because whoever accumulates fastest wins. That is the whole mechanism, and it is indifferent to the content of the advertisement.
Witness does not work like that. For me to thank you, you have to have actually done something for me. It cannot be purchased, because it must be earned by delivery. It cannot be transferred, because it attaches to the person who did the thing. And it is generated overwhelmingly where human beings physically are — which is to say, locally. A gratitude economy would therefore have a bias toward proximate, small, human-scale providers for exactly the symmetric reason the ad economy has a bias toward scale.
I want to be careful about what kind of claim this is. It is not people should prefer local businesses. It is not a campaign. Nobody has to prefer anything. The discovery layer simply cannot be bought by distance, in the way a layer made of purchased attention obviously can. It is a property of the material, not a virtue anyone has to keep practising.
That distinction matters to me more than it might seem to. I am trying to build an institution meant to outlive me and be handed to a successor I will never meet. Anything in it that works only because someone is choosing, every day, to be good is a thing I cannot promise will survive. Anything that works because of what it is made of does not need my promise.
I used to treat post-AI as a mood word. I do not think it is one here. There is a mechanism, and it runs in two directions at once.
AI drives the cost of manufacturing attention to nearly zero — the copy, the images, the video, and, more importantly, the testimony. A generated review, a generated recommendation, a generated enthusiastic customer are all free now and unlimited. But testimony is the attention layer's actual currency: what makes an advertisement work is that some part of it is believed. When belief can be manufactured at zero marginal cost, the layer's signal value collapses under its own supply. The attention economy is being devalued by its own inputs getting cheap.
At the same time, and this is the half people talk about less, AI collapses the cost of everything that can be delivered down a wire. What stays scarce is the work needing a particular human body, in a particular place, at a particular time: care, food, repair, craft, teaching, the physical maintenance of a neighbourhood. Those things were always local. They are about to be a much larger share of what remains genuinely scarce.
So the supply side is moving toward exactly the shape a witness layer is good at describing, at the same moment the attention layer's own signal is being counterfeited into worthlessness. That is a hinge, not a hope. What breaks the old layer is that attention became cheap to manufacture. What holds the new one up is that witness stays expensive — it still costs a real person actually receiving something from another real person.
I would rather put these here than have someone else find them.
Advertising pays for things, and appreciation does not — though this objection has a clock on it. This is the one I find hardest. Ads are not only a steering mechanism; they are a subsidy. They pay for search, for video, for a great deal of journalism, and the person consuming those things pays nothing at the point of use. That is a real transfer of value to ordinary people, and it is large. An appreciation layer has no equivalent.
What I have since realised is that the objection is time-limited rather than permanent, and I had been treating it as permanent. As robots and automated production make things cheaper to make, the subsidy matters less, because the thing being subsidised costs less. That is genuinely a partial answer.
But it is a smaller answer than it first looked, for two reasons I had to be argued into. Falling cost is not falling price. Which of the two you get depends entirely on who owns the machines: under concentrated ownership, a collapse in production cost shows up as margin, not as cheaper goods. Who owns the bots decides whether abundance reaches people or accumulates. And automation does not abolish scarcity, it moves it — to compute, to energy, to land, to materials, to the attention of skilled people. Something always stays rivalrous, and whatever stays rivalrous is what gets charged for. So the subsidy question does not dissolve. It relocates, and it relocates into a question about ownership, which is the same question the first half of this essay was already asking about the discovery layer. I asked two questions and they turned out to be one.
This has been tried, and it was captured. Consumer review platforms are witness layers for small local businesses. They were captured twice over — by a market in fake reviews, and by monetisation that sits uncomfortably close to placement. I do not think that happened because the people running them were worse than us. I think it happened because of a shape: reviews are fungible and they are added up into a per-business total, and both the fraud and the upsell attach to that total. Which means the only interesting question is whether the layer can be built without ever computing such a total. That is a design question with a checkable answer, and it is why I wrote a specification rather than only an essay.
Witness is bad at strangers. Advertising pushes: a seller pays to reach people who have never heard of them. Witness pulls: it travels along real relationships, one person telling another. That makes it good at bringing people back and structurally bad at cold-start, which is precisely when a new business most needs help. So the honest version of my thesis is narrower than the one I started with: appreciation can displace the repeat half of the advertising layer, not the finding-strangers half — unless you build ranking, which this whole line of work refuses to do. I regard that as the interesting problem rather than a hole to paper over.
And I should not romanticise the corner shop. A small local business is not virtuous by virtue of being small; some are exploitative, and large firms have delivered enormous real gains in cost and reliability that I benefit from daily. My interest in local circulation is not that local people are better. It is that a local economy is where gratitude is actually generated, because that is where people are in rooms together — and I would like the discovery layer to be made of the thing the economy is actually producing rather than the thing that can most easily be bought.
I should say plainly that I am not a neutral party to the argument I just made.
I am building a manufacturing arm, and it has committed by charter to selling at cost — to letting price fall as automation improves rather than keeping the difference. So when I write that who owns the machines decides whether falling cost reaches people, I am describing a general problem to which I have a specific and interested answer. That does not make the general problem less real, and I think the argument stands on its own. But a person arguing that ownership determines outcomes, who happens to own something, should say so in the same breath. I have also not yet proven that a commitment of that kind survives contact with the pressure of an actual business, and I will not know for years.
The thing I keep turning over is what happens if the abundance argument works.
Suppose it does. Suppose production costs collapse and most goods become nearly free. The obvious conclusion is that the discovery layer stops mattering, because when something costs nothing, choosing badly costs nothing either.
I now think the opposite is true, and it took me a while to see it. Abundance does not remove scarcity; it moves the scarcity from money to attention. If everything is free, what you spend is no longer your money but your time, and your time is the input the attention economy was already built to harvest. Every choice still costs you a piece of your life.
Which means a world of nearly-free goods is not a world where steering matters less. It is the attention economy's ideal environment — nothing left to buy, nothing left to charge for, and the only remaining scarcity is exactly the one it already knows how to take. If we get to abundance without having fixed the discovery layer first, we will have built a civilisation in which the only thing anyone competes for is the thing every human being holds equally, and the competition will be won by whoever is best at taking it.
That is the strongest reason I know to work on this now rather than later.
Very little, so far, and I would rather say that than let the argument imply otherwise.
I have written down a discovery mechanism with no per-business total anywhere in it, no impression count, no purchasable placement, and no surface facing the businesses at all — where who appears is decided by a published rotation, and admission tests whether a person is still circulating rather than measuring how much they have received. It is unbuilt. It is specified in a companion paper, which registers four predictions before there is any data, including one whose failure would tell me to abandon the design rather than tune it.
The prediction that matters most for what I have argued here is simple enough to state in a sentence: among gratitude payments that reach a business, the share going to single-location operators should materially exceed those operators' share of local advertising spend, and the gap should widen as generated advertising gets cheaper. If it turns out that appreciation flows to the same businesses that advertising already flows to, then witness is not structurally local. It is just a cheaper advertisement, and I am wrong.
I have written that down where I cannot quietly withdraw it.
What changed my mind was not discovering that advertising is enormous. It was discovering that it is small, and decisive anyway.
One percent of the world's spending decides which of the world's sellers the rest of us ever hear about. That is a very small lever attached to a very large door, and small levers are the ones worth examining, because they are the ones that can actually be replaced. You are not going to reorganise a hundred and ten trillion dollars of economic activity. You might, conceivably, change what steers it.
I do not know whether appreciation can do that job. I know it has two properties the alternative does not — it cannot be bought at a distance, and it is held equally by everyone before anybody captures it — and I know how I would find out whether those properties are enough.
A note on the figures. The global total is WPP Media's and WARC's 2025 estimates, which land between $1.1 and $1.2 trillion; the ratio to world GDP is my own arithmetic. The long-run American series is Robert Coen's, curated at McCann-Erickson and extended by Douglas Galbi, and the no-trend finding is the one reported in the recent aggregate-advertising literature. Sources differ on the American ratio — roughly one and a half percent on contemporary measured-media figures, closer to two percent on the longer series, the gap reflecting what each counts as advertising. The claim this essay leans on is the absence of a trend, not any particular level, and that claim holds across the versions I could find. If it turns out not to, the household-inequality argument in the third section loses its falsifier and I would want to know.
Written with Miss Aquarius, the named AI collaborator on this corpus. The ideas and the editorial control are mine. Dedicated to the public domain under CC0 1.0 Universal.