Human attention is the only input in the global economy whose total supply is fixed by biology.
Which ought to make it the most expensive thing money can buy.
It does not feel that way from inside the industry. Ask anyone who buys attention for a living, in Mumbai or Lagos or Warsaw, and they will tell you reaching people is harder than it used to be, that the platforms keep pushing rates, that targeting is not what it was. Then they will show you a plan to purchase several billion units of the resource that is supposedly running out.
Nobody in that room is behaving like a person facing a hard physical limit. That composure is the thing worth explaining.
The arithmetic
Supply can grow two ways. More connected people, or more time from each.
The number of internet users rose by 59 million last year, an increase of one per cent. That is the slower channel, since the remaining unconnected population sits largely in South Asia and Africa, where connection is a decade-long infrastructure question rather than a marketing one.
The other channel has closed. Global time online climbed through the 2010s, peaked near seven hours a day during the pandemic, and settled back roughly to where it had been, which GWI described as a form of internet saturation.
Call total supply growth one per cent, and be generous.
Global advertising spend closed 2025 at $1.19 trillion, up almost nine per cent, on a path to $1.4 trillion by 2027.
Nine against one.
In any other market this resolves one way. Price rises until demand is rationed back to what supply can bear. It is the most reliable mechanism in the discipline. If a factory input behaved like this, procurement would be in crisis and every board member would know its name.
Prices did rise. Nowhere near enough, and nobody called it a shortage.
Something gave way, and it was not the price.
The unit gave way
Under the Media Rating Council standard, an impression counts as viewable if half its pixels are in view for one continuous second.
Read that again as a definition of human attention. Half an advertisement, for one second, with no requirement that a person was looking.
Define the unit that way and the ceiling disappears. A fixed number of human hours divides into a practically unlimited number of one-second fragments, and every fragment is saleable. Biology stops binding, because what is being sold has stopped being attention.
Note also where the threshold was set. A trade body in one country wrote it, and it became the reference unit for trading in markets that have nothing else in common with that country.
Currencies have done this for as long as there have been currencies. When a treasury owed more silver than it held, the remedy was rarely to admit the shortfall. It was to take metal out of the coin and leave the name on it.
Denominational drift is what happens when the name of a unit stays constant while the reality it measures quietly changes.
Markets rarely solve scarcity by creating more of what is scarce. More often, they redefine what counts.
Once you have the phrase, it stops looking like an advertising problem. European statisticians reclassified research spending from consumption to investment and added between half a point and four points to measured GDP. No additional economic activity occurred anywhere on the continent. The boundary moved, and every debt ratio, deficit target and growth comparison resting on that denominator moved with it.
Why this happens to every institution eventually
The question I keep returning to is not how drift occurs. It is why no institution appears to be immune to it.
The answer, I think, is uncomfortable and has nothing to do with dishonesty.
Reality is expensive to measure. Metrics are cheap to compare. Over time every institution drifts toward what it can compare, because comparison is what committees are actually built to do.
Consider what it costs to know whether an advertisement changed a mind. You would need controlled experiments, holdout groups, patience across quarters, and a tolerance for being told your last three campaigns achieved nothing. Now consider what it costs to know how many impressions you bought. One line, in an email, on a Monday.
The second number is not better. It is available.
And availability compounds. The cheap number gets reported, so it gets tracked, so it gets targeted, so it enters someone's variable pay, so it becomes the thing four departments negotiate over. By the time anyone asks whether it still measures anything, it has become the language the organisation uses to talk to itself. You cannot remove it without removing the ability to have the conversation.
This is why drift is not a failure of intelligence. Every individual in the chain made the sensible choice available to them at the moment they made it. The organisation ends up steering by a number nobody defends, assembled from a hundred decisions each of which was defensible.
Why nobody stops it
Somebody in the room always observes that the metric is imperfect. Everyone nods. Nothing changes.
I have watched some version of that exchange more times than I can count, and it took me years to understand what was happening in it, because from the outside it looks like collective negligence and it is nothing of the kind.
Watch what people actually do. They accept the number in public and discount it in private. Everyone applies a personal haircut. Nobody states theirs aloud, partly because it would be rude and mostly because the size of your private discount reveals what you really think of the last three years of your own reporting. So the meeting proceeds on a figure that not a single person present actually believes, and everybody leaves knowing that, and nobody is lying.
The reason is that a metric is only secondarily a measurement. Its first function is to be a medium of agreement. It is the thing that allows thirty people with conflicting interests to hold one conversation and reach a decision before lunch. Reject the unit and you do not become more accurate. You become unable to transact with anyone still using it.
There is also a quieter cost that keeps people seated. The first person to name the problem inherits the problem. Raise it seriously and you will be asked to lead the review, which means eighteen months of work, a report that concludes what everyone already privately knew, and a reputation for being difficult about numbers. So the observation gets made in a tone that signals it is not a proposal. That tone is the whole of the transaction. It lets the speaker be right without being responsible.
And no single firm can leave unilaterally. Report honest attention while competitors report impressions, and your figures fall while theirs climb. You will spend two board meetings explaining a decline you manufactured by telling the truth.
That is precisely why bad money outlives everyone's knowledge that it is bad. Not because the knowledge is missing. Because the knowledge is privately held, universally shared, and individually unactionable.
Who controls the mint
The redefinition is never performed by a neutral party. It is performed by whoever benefits from the count.
Alphabet, Amazon and Meta take 56.1 per cent of global advertising spend outside China, heading towards fifty-eight per cent. They issue the unit, define it, count it, and largely audit the count. No treasury in history has held all four functions and used them with restraint.
Occasionally somebody weighs the coin anyway. eBay ran controlled experiments on its own paid search and found that for searches already containing its name, 99.5 per cent of the traffic those advertisements appeared to deliver would have arrived regardless.
That is not a scandal. It is an assay result.
What holds its value
The instinct is to reach for duration. Time spent, honestly measured. But a person can sit in front of something for forty seconds and consider nothing at all, and duration is a proxy in exactly the way an impression was, debasable by the same route.
The property that resists drift is not duration. It is who controls the mint.
A unit degrades when the party who profits from the number also defines it. A unit holds when the count belongs to the person being counted. Someone typing your name into a search bar. A customer who returns without being reminded. A recommendation made to a colleague when you were not in the room.
None of these can be minted. The seller does not own the press.
The question worth asking
Every organisation is governed by numbers. Very few stop to ask whether the numbers still mean what they meant when somebody first decided to trust them.
Attention is only the clearest case because the metal was removed so quickly and so visibly. The same drift operates on any figure a business steers by, and it operates most powerfully where the number has been stable for years, because stability is what stops people looking.
So before changing the strategy, or the budget, or the people, there is a simpler question that usually comes first and almost never gets asked.
Did reality change?
Or did the denominator?
Sources: DataReportal and GWI on global internet users and time online. WARC Global Ad Forecast, December 2025. Media Rating Council, Viewable Ad Impression Measurement Guidelines. Bruegel on ESA 2010 and the European GDP revisions. Blake, Nosko and Tadelis, Econometrica (2015).
