Who Holds the Yardstick

2008 · A NUMBER THAT ISN'T THERE

A number that isn't there

A e6-measurement-006 ~18% YoY paid-click growth, Q4 2008 — no count, CPC, or query total disclosed. In its earnings release for the last quarter of 2008, Google told its shareholders one fact about its ads. People had clicked them about 18% more often than a year before. That was all. The company gave no total number of clicks. It gave no average price per click. It gave no count of searches.

Hold that gap in mind. It is the whole story of the American advertising market, read backward.

B e1-measurement-002 ABC founded 1914 in Chicago — the world's first circulation auditor. A e6-measurement-006 2008 self-count set against the 1914 audit — the one load-bearing juxtaposition (the hero hands off here). Set 2008 against 1914. In 1914, in Chicago, publishers and advertisers merged two rival groups into the Audit Bureau of Circulations. It was the first body in the world to audit circulation – how many copies a newspaper actually sold. Before it, a publisher stated its own circulation, and the advertiser buying the space took the publisher's word for it. After it, an outsider checked the figure. The market had just built the outside audit – the very thing 2008 would quietly take back.

1833–1914 · WHO KNEW THE REAL PRICE

Who knew the real price

Start at the price, because the price was once a public number.

64.9% advertising = 64.9% of US newspaper gross income, 1914 — $184.1M from advertisers against $99.5M from readers. (e1-sellers-003, A) B e1-events-001 Penny Sun, 1833: ~15,000 copies/day within ~2 years — the advertiser-funded model. The model is old. The New York Sun went on sale in 1833 at a penny a copy, and within about two years it was selling roughly 15,000 copies a day. You clearly couldn't run a newspaper on a penny a copy; something else had to be paying for it. By 1914, across all US newspapers, advertising supplied 64.9% of gross income – $184.1 million from advertisers against $99.5 million from readers. The advertising was subsidizing the paper.

To help with the buying and selling of advertisements, a new kind of business emerged: the advertising agency. It acted as a middleman between the advertiser and the newspaper. It bought space on the advertiser's behalf, and often wrote the ad as well.

B e2-creators-002 Standard 15% agency commission; on radio it covered production too. For that work it charged a commission of 15% of what the advertiser spent on space. That commission was the agency's price, and it was public. Everyone in the trade knew the rate. On radio, the same 15% did two jobs at once, because the sponsor's agency both bought the airtime and produced the program that ran on it. The 15% was only the visible part of the middleman's take. The rest, as the next section will show, stayed out of view.

A e4-pricing-001 The 15% commission shows up in later figures: census commissions-received held near 14% (14.13% 1977 → 13.18% 1992). s-AYER: no 1875 open-contract artifact in the corpus. This open pricing is usually traced to N. W. Ayer, a leading early ad agency, and its open contract, dated by tradition to 1875. Under an open contract the agency showed the advertiser exactly what it had paid the newspaper, so the advertiser could see the real cost of the space and the size of the agency's cut. Our own record does not hold that 1875 contract, so the founding document is missing here. The price itself, though, is not missing. The 15% commission shows up in later figures this piece relies on.

1914 · WHO COUNTED THE READERS

Who counted the readers

Now the count. This is where the audit begins, and with it the ninety-year exception.

B e1-measurement-002 ABC founded 1914 in Chicago, the world's first circulation auditor. The Audit Bureau of Circulations, founded in Chicago in 1914, was the world's first body to audit circulation. In 1914 the medium was print, so "circulation" meant the number of copies a newspaper or magazine actually sold; that number stood in for how many readers an advertiser was reaching. Until then, a publisher simply stated its own circulation, and the advertiser had to take the figure on faith.

The Bureau put a stop to that. It was built by both sides of the market at once: it merged a group led by the publishers who sold the space with a group led by the advertisers who bought it. That gave it the standing to check the publishers' claims as a neutral outsider. For the first time, the number that set an ad's price no longer belonged to the seller being paid for it.

1944–1949 · THE COUNT STARTS TO SLIP

The count starts to slip

The independent count did not hold for long. It began slipping almost as soon as it appeared. Where you first see this play out is on radio.

In print there had been one audited circulation number. In radio there was nothing like it. Instead, rival services sold competing audience counts, and they did not agree.

~20% Hooper's count ran ~20% above the co-op's, on the same telephone method after 1944. (e2-measurement-003, B; ci80 15–25) Two of them mattered. One was the Cooperative Analysis of Broadcasting, an industry service paid for jointly by the advertisers who used it. The other was C. E. Hooper, a commercial firm that sold ratings for profit. After 1944 both used the same telephone method: an interviewer called homes and asked what the household was listening to at that moment. Even so, Hooper's numbers ran about 20% higher than the Cooperative's. A McCann-Erickson study traced the gap to small technical choices in running the same calls: how rings were counted, how unrecognised programs were handled, how busy signals were treated.

B e2-measurement-008 CBS guaranteed $3,000/wk per Hooper point Benny lost, Jan 1949; the vendor rating written into a deal term. By the end of the decade, that number had a price tag on it. Radio measured its audience in ratings: a rating point was the count turned into a share, one point meaning that one percent of all radio homes were tuned to a program. In January 1949, the radio comedian Jack Benny moved his show from the NBC network to CBS. To land him, CBS made a guarantee to his sponsor, the American Tobacco Company: for every Hooper rating point Benny's show lost by switching networks, CBS would pay the sponsor $3,000 a week. The rating was no longer a measurement off to one side. It was a term in the deal, with a dollar figure on every point.

1925–1938 · WHO ELSE TOOK A CUT

Who else took a cut

B e2-creators-002 The one published cut: the 15% agency commission. The 15% was only the cut everyone could see. Behind it, several other parties each took a piece of the same ad dollar, and most of those cuts were never public.

73/27 the split the 1938 FCC audit extracted: NBC+CBS kept 73% ($32,046,218), paid 27% ($12,267,560) to 253 stations. A number wrested from the seller. (e2-sellers-002, A) One of those hidden splits came out only because a regulator pried it open. In 1938 the FCC audited the radio networks. NBC and CBS did not reach listeners directly; they sold national air time to advertisers and then fed their programs to local stations that broadcast them in each city. The audit showed how the networks divided that money: they kept 73% of their network time sales ($32,046,218) and passed the other 27% ($12,267,560) to the 253 local stations that carried the programs. That 73/27 split had never been published. It took a government audit to make it visible.

C e3-pricing-003 Station reps took 5–20% on the net after the agency's cut; a hidden, unpublished cut. B e3-pricing-002 A 2% cash discount on the net, usually left out of take-rate comparisons. There were still more cuts stacked on the agency's 15%. A station representative firm, a company that sold a local station's air time to national advertisers, took a further 5% to 20%, figured on whatever was left after the agency's share. And the seller usually gave a 2% discount for paying the bill fast, which the agency pocketed rather than passed on. Neither of these ever appeared in the headline 15%.

So the middleman's take was never a clean 15%. It was three or four separate cuts, charged by different parties on different bases, and they cannot be honestly added into one number. The 15% was simply the one that was public, and it was the smallest of them.

1956–2003 · DID THE 15% DIE?

Did the 15% die?

The neat story is easy to tell: DEADthe 15% agency commission died sometime in the 1980s, killed off either by a government ruling or, later, by the internet. Neither is what happened. The record contradicts both.

A e3-creators-001 AAAA consent decree, 1 Feb 1956: ended collective 15% enforcement. It changed the law, not the price. B e3-creators-002 71% of the largest US advertisers still on commission in 1982 (ANA tracking). A e4-pricing-001 Census commissions-received held near 14%: 14.13% (1977), 13.92% (1982), 14.16% (1987), 13.18% (1992). The peak, not the death. The first supposed killer was the law, and it did strike a blow. The agencies ran their trade through a shared industry body, the American Association of Advertising Agencies. That body enforced the 15% collectively. It held its members to the standard rate, and media owners agreed to pay the commission only to agencies the industry formally "recognized." In 1956 the US government challenged this as anticompetitive. The case ended in a consent decree, a court-approved settlement, signed on 1 February 1956. The decree broke up the collective enforcement and the recognition system. But it changed the rule, not the price. As late as 1982, 71% of the largest US national advertisers were still paying their agencies by media commission: a set percentage of what they spent on advertising space, rather than a flat fee. And the commission agencies actually collected, which the US Census tracked, stayed close to 14% for years: 14.13% in 1977, 13.92% in 1982, 14.16% in 1987, and 13.18% in 1992. The 15% did not die in this stretch. This was its peak.

B e6-creators-002 Commission basis fell to ~10% by 2003 from 61% in 1994; followed unbundling, not search. The second supposed killer was the internet, and it arrived too late to be guilty. For most of the century one agency did two jobs at once: the creative work, meaning the ideas and the ads themselves, and the media buying, meaning the purchase of the space those ads ran in. The single 15% commission paid for the whole bundle. In the early 1990s the two jobs split apart. Specialist firms took over media buying, clients began paying separate fees for creative work, and the old model of paying one commission on media spend stopped making sense. That is what collapsed. The share of advertisers paying their agency on a commission basis fell from 61% in 1994 to about 10% by 2003. All of this happened before search advertising, the paid results you see on a search engine, existed at any scale. Our research does not attribute the fall to search. It followed the unbundling.

1987 · WHEN THE RULER CHANGED

When the ruler changed

In September 1987 the count changed again, and this time the change was in the ruler, not the world.

≈−10% Sept 1987 people meters measured ~10% fewer at network prime time — CBS/ABC each −13%, NBC −4%. The honest count cut the sellers' own numbers. (e4-measurement-001, B) By the 1980s a single company, Nielsen, produced the audience numbers the whole television business ran on. That count was the currency: advertisers and networks set the price of a commercial by how many people Nielsen said were watching. In September 1987 Nielsen switched the national standard to the people meter. It put a small box in each sampled home with a labelled button for every household member, set to their age and sex. When a person started watching, stopped, or left the room, they were meant to press their button, so the count was built from who actually said they were there. The method it replaced was a paper diary each household filled in by hand. Measured this new way, network prime time came in about 10% lower. The three big networks took the drop unevenly: CBS and ABC each fell 13%, NBC 4%. Nothing had happened to the audience. The same people sat on the same couches. What changed was the way they were counted.

1914–2008 · THE BUYER'S OWN METER

The buyer's own meter

One kind of buyer never trusted the seller's audience count, and never had to: the direct-response buyer, who pays only for the replies it can count itself. Its own meter, that count of replies, is the oldest independent number in this market.

C e1-buyers-008 Direct-response pool ~$192M in 1914, ~16% of the benchmark — wide, low-confidence. B e3-buyers-001 Direct mail = 24% of US national advertising in 1956, more than broadcast. This is the oldest model in advertising. In the early 1900s, mail-order companies and patent-medicine sellers, the big direct-mail advertisers of the era, ran ads with a coupon to clip and mail back, then counted the replies each ad brought in. The pool of this spending was already large. In 1914 it was very roughly $192 million, about a sixth of the whole US ad market that year. That early figure is a wide, low-confidence estimate, third-hand, and should be read as a range, not a point. It kept growing. By 1956, direct mail alone was 24% of all US national advertising, a larger share than broadcast, radio and television together. So paying only for measured replies, what we now call performance advertising, was not something the digital platforms invented. It had been the single largest national medium for about half a century before any platform existed.

A e6-buyers-008 57% of US internet ad revenue bought pay-per-action in 2008 (up from 51%); 39% CPM. Online, this buyer's model won. By 2008, 57% of US internet ad revenue was bought pay-per-action, meaning the advertiser paid only when someone clicked or bought, up from 51% a year earlier. Only 39% was bought by the impression, the older way, a flat rate each time the ad was shown whether or not anyone acted on it. So the buyer's independent count had finally become the main way ads were sold. But it won in the one medium where the seller, the platform, also owns the record of the very actions being paid for. The independent number everyone wanted had landed back inside the seller's own books.

1914–1971 · WHO GETS IN THE DOOR

Who gets in the door

Two things get blurred together that should stay separate: who is allowed to buy in this market, and where the market's money actually ends up. Pulled apart, they point in opposite directions.

C e1-buyers-007 1914 newspaper classified carried no agency commission and no cost of preparation. Buying an ad yourself, with no agency in the middle, is old. In 1914 a small buyer could place a newspaper classified, one of the little text ads in the back pages, with no agency commission to pay and nothing charged for making the ad. That buyer reached the public directly, a hundred years ago. So self-serve buying, placing your own ad with no middleman, is not something the digital platforms invented.

61.4→66.3% national advertisers' share of US ad spending, 2002→2007; the money leaned toward national buyers: a long tail of buyers, not of dollars. (e6-buyers-005, B) The money, though, did not spread out the way the story promised. That story runs like this: search opened the market to small buyers and pushed spending out to a long tail, the many small advertisers who, added together, were meant to rival the big ones. The dollars went the other way. National advertisers, the big brands that advertise across the country, saw their share of US ad spending rise from 61.4% in 2002 to 66.3% in 2007. Far more small buyers did get in. But the money leaned harder toward the big national buyers, not away from them. What opened up was a long tail of buyers, not a long tail of dollars.

B e3-buyers-004 Cigarette makers paid broadcasters ~$225M/yr before the 2 Jan 1971 ban. B e3-medium-011 1971 was the era's only down year for US TV advertising, −1.7%. The government controls the door too, and it can shut out a whole class of buyer at once. Before the federal ban on cigarette advertising on television and radio, which took effect on 2 January 1971, cigarette makers were paying broadcasters about $225 million a year for airtime. In 1971, US television advertising fell 1.7%, the only year in that era it dropped at all. So the market does not decide alone who gets to buy. The government decides too, and it can turn a paying customer away for good.

1975–2025 · THE DIAL

The dial

Advertisers spent a century turning one dial ever finer: how precisely they could choose who sees an ad. Precision, it turns out, was never the point.

442M Overture paid clicks, Q4 2001 — matched to typed words, "with no cookie and no profile," across ~53,000 advertisers. (e5-targeting-004, A) Then search advertising changed what the targeting aimed at. In the last quarter of 2001, Overture, the company that invented paid search ads, delivered 442 million paid clicks across about 53,000 advertisers. Each ad was matched not to a person but to the words someone typed into a search box, with, in its own words, "no cookie and no profile." This was the most powerful new targeting of its age, and it knew what you wanted without knowing who you were. So the ad was aimed not at a person but at the thing that could be counted, a typed query that belonged to no one in particular.

C e7-buyers-003 Vendor survey: 71% ran Performance Max in 2025 (up from 60%) — soft endpoint, wide band. B e7-targeting-002 Apple ATT opt-in ~16% at launch (May 2021) → ~25% by mid-2022. Later, buyers handed the dial back to the machine, letting Google's software choose the targeting for them. By 2025, a vendor survey of more than 170 advertisers found 71% running Performance Max, Google's fully automated campaign type, up from 60% the year before. Treat that number lightly: it comes from a small survey panel, not a count of Google's whole advertiser base, and it carries a wide margin of error. It is a soft endpoint, not the point of the story. And identity can be pulled back off the ad by whoever controls the device. Apple's App Tracking Transparency makes apps ask permission before tracking you. It drew opt-in from about 16% of users when it launched in May 2021, rising to roughly 25% by mid-2022.

2002–2024 · THE GOVERNED PRICE

The governed price

The price of a search ad is settled by an auction dressed up as a neutral machine. But the seller builds that machine, owns the record it runs on, and can quietly tune it.

A mech-adwords-001 AdWords Select, 20 Feb 2002: advertisers "pay the lowest amount possible". Google's search-ad auction, AdWords Select, launched on 20 February 2002 with a promise: advertisers would "pay the lowest amount possible." Here is how it worked. Each advertiser set a bid, the most it would pay for a click. The auction ranked the ads by that bid and by how often each ad got clicked, then charged the winner not its own bid but just enough to stay ahead of the next advertiser. It read like a fair machine. But the machine ran on a record that Google, the seller, owned, kept, and billed from.

A e5-pricing-004 Overture was pure first-price, ~$0.23/click, listings needed editorial approval. The first search auction was not even this fair-looking. Overture ran the original one in late 2001, and it was pure first-price: the ads were ranked by bid, highest first, and each advertiser simply paid its own bid, averaging about $0.23 a click. The gentler, second-price design came later, with Google; it was not how search advertising began. Overture also guarded its door: a person had to approve each listing before it could run. AdWords let advertisers in with no such approval.

A mech-rgsp-001 rGSP (2019): +5.91% top-slot CPC on PC/tablet, +4.85% mobile; advertisers cannot opt out. A mech-knobs-001 Three pricing knobs — 'Butternut Squash', format pricing, rGSP — 'intentional pricing'; exponent never disclosed. A mech-mehta-004 Court: Google tested raising text-ad prices 5%+ — "the results have been largely consistent — it can". Then the seller began tuning the price upward. In 2019 Google reworked the search auction; the court would later label that redesign rGSP. Google's own pre-launch tests showed it raising the cost per click for the top ad slot by 5.91% on PCs and tablets, and 4.85% on phones. Advertisers could not opt out. In 2024 a federal court set out its findings of fact, the facts a judge formally rules to be true. It found that Google used three controls to set prices in the search text-ad auction: one codenamed inside the company "Butternut Squash," plus a lever it called "format pricing," and the rGSP redesign. The court called all three "intentional pricing." That Butternut Squash setting, it noted, has never been disclosed to the public. The same court found that Google had tested whether it could raise text-ad prices by 5% or more without losing many advertisers, and wrote: "the results have been largely consistent — it can."

A mech-tac-003 Syndicated take rate ~9% (2002) → 24.7% (2006) → 21.3% (2008) — the ratchet reversed. One fact complicates the simple story that Google only ever squeezes. Its cut does not move in only one direction. Consider Google's syndicated inventory: the ad space it sells on other companies' websites, sharing the revenue with them. Its take there, its own share of that revenue, rose from about 9% in 2002 to a peak of 24.7% in 2006. Then it fell to 21.3% by 2008, as rivals competed harder for those partner sites. A pure profit-maximiser would not let its own cut fall.

1943–2024 · THE STATE MOVES THE TOP OF THE STACK

The state moves the top of the stack

A few big players come to dominate this market, and they reach the top by spending money, not by the government's design. What the government mostly does is rearrange who sits at that top, and cap how large any one of them can grow. It very rarely breaks the concentration apart.

A e6-sellers-001 Overture led paid search: $667.7M revenue 2002 vs Google's $439.5M total. The money point is plain in the numbers. Entering the 2000s, Overture, not Google, led the paid-search business by revenue. In 2002 Overture took in $667.7 million, against Google's $439.5 million in total revenue across the whole company. The leader of search advertising was the one with the bigger business, and our research does not credit Google with inventing the pay-per-click auction. Overture built it and led it.

B e2-events-006 Edward J. Noble bought NBC's Blue Network for $8M, 12 Oct 1943 — a sale, not a build. A e3-sellers-002 From 1954 the FCC capped owners at 7 TV stations, max 5 VHF. B e4-sellers-004 RBOC Yellow Pages sold $9,517M of advertising in 1993 — more than radio or magazines. What the government mostly does is move existing players around and set ceilings on them. Take 1943. A third national radio network was created not by building one but by selling one off: Edward J. Noble bought NBC's Blue Network, one of NBC's two radio networks, for $8 million on 12 October 1943, and it became ABC. An existing giant's spare network changed hands; no newcomer built a rival from scratch. (What forced that sale is not in this record, so I state only the sale.) Then the government set a ceiling: from 1954 the FCC, the broadcast regulator, let no single owner hold more than seven television stations, and no more than five of those on the stronger VHF channels. It can create sellers as well as cap them. After the 1984 breakup of AT&T, the regional "Baby Bell" phone companies became the main publishers of the Yellow Pages, the phone-directory ad business. In 1993 they sold $9,517 million of advertising there, more than radio or magazines.

A e6-targeting-004 Targeting entered by purchase: DoubleClick $3.1bn (Apr 2007), aQuantive ~$6bn, Right Media $680M, 24/7 $649M — largest deals of the year. The tools for targeting audiences arrived the same way, by purchase rather than by building. The big companies bought the ad-technology firms that had them. Google agreed to buy DoubleClick for $3.1 billion in April 2007. Microsoft bought aQuantive for about $6 billion, Yahoo bought Right Media for $680 million, and the ad group WPP bought 24/7 Real Media for $649 million. These were the largest such deals of that year, not the typical ones.

B e2-sellers-006 Postwar boom tripled the AM dial: ~900 stations (1945) → ~2,500 (early 1950s). One real spreading-out did happen, and the government allowed it without planning it. After the war, the number of AM radio stations roughly tripled, from about 900 in 1945 to roughly 2,500 by the early 1950s, as the FCC handed out the licences.

$26.3B search default revenue share Google paid in 2021 — the court found it foreclosed competition, then left the machine intact. (e7-events-009, A) A mech-mehta-004 The court called Google's search text-ad prices "supracompetitive" — a seller-owned number surfaced by a court finding. And when a court finally called the modern structure unlawful, it stopped short of taking it apart. In 2024 a court found that in 2021 Google had paid $26.3 billion to be the built-in default search engine on other companies' devices and browsers. It ruled that this spending had shut competitors out. It also found that Google charged more for search text ads than a competitive market would bear, what the court called "supracompetitive" prices. The ruling named the conduct as illegal, but it left the machine standing.

1919–2025 · EVERY NEW UNIT IS BORN UNCOUNTED

Every new unit is born uncounted

A e6-medium-006 Coen series understates internet by ~half ($10,529M vs $21,206M, 2007) and stops at 2007. The clearest sign of this is the standard yardstick itself failing at the end. For decades, the go-to measure of US advertising spending was the Coen/McCann series, which ran from 1919. As the internet arrived, that series undercounted it badly. It put internet advertising at $10,529 million for 2007, while the industry's own count was $21,206 million, about twice as much. Then the series simply stops, at 2007. The standard measure went dark exactly as the new unit arrived.

B e2-medium-006 TV entered the series in 1949 at $58M, 1.1% of total. B e4-medium-002 Total TV first passed newspapers in 1992 — $31,079M vs $30,737M. A e7-medium-001 US internet ad revenue $23,448M (2008) → $294,593M (2025), 12.6×; only down year 2009. A e6-scale-009 Help-wanted classified fell 91% from its 2000 peak; real-estate classified fell with housing, not search. A related story fails on the same data: the idea that each new medium conquered and killed the one before it. Television first showed up in the figures in 1949, at $58 million, just 1.1% of all US ad spending. Television did not pass newspapers until 1992, and even then only by about 1%: $31,079 million against $30,737 million. The shares moved slowly, and no medium wiped out the last one in a single stroke. The internet did grow fast, once someone counted it: from $23,448 million in 2008 to $294,593 million in 2025, about twelve and a half times over, with a single down year in 2009. And even "the internet killed the classifieds" is too simple. Newspaper help-wanted ads, the job listings, did collapse, down 91% from their 2000 peak. But real-estate classifieds, the property listings, peaked in 2006 and fell with the housing market, not with search.

1922–2025 · WHICH TOTAL?

Which total?

Every "share of advertising" depends on which total you divide by, and that total depends on what you choose to count. So a share is never a plain fact; it is a choice of definition. Direct mail shows this most clearly.

B e2-medium-004 Direct mail: 14.5% of US ad spend in 1949, peak 16.8% in 1938. B e7-scale-001 2025 US total ~$405B on media-owner basis (MAGNA $398B) — excludes direct mail entirely. For most of the years the data breaks down by medium, direct mail was the second-largest one of all: 14.5% of spending in 1949, and a peak of 16.8% in 1938. Yet the headline 2025 total leaves it out. That total, about $405 billion (the research firm MAGNA puts it at $398 billion), is built on what the media owners take in, the revenue collected by the sellers of ad space. Direct mail has no media owner taking in that revenue, so this basis leaves it out entirely. The biggest old medium is simply written out of the new total.

C e7-scale-006 National-brand pool ~$145.7B, 36.6% of $398B; IRS $473B vs MAGNA $325B (2022) = $148B rail wedge. These different ways of counting, call them rails, disagree by more than the footnotes admit. Stay on the media-owner-revenue rail, and the "national brand" pool for 2025, spending by advertisers who sell nationally, is about $145.7 billion, or 36.6% of MAGNA's $398 billion. That makes it the largest or joint-largest pool, though on a wide, low-confidence estimate. Switch to a different rail and the whole picture moves. For 2022, the IRS recorded $473 billion of advertising written off on corporate tax returns, against MAGNA's $325 billion for the same year. That is a $148 billion gap between two counts of the same thing. So any chart of this has to name its rail right on the axis, not bury it in a footnote. There is no single US advertising total waiting to be found.

B e2-scale-004 Advertising peaked at 3.0% of US GDP in 1922. B e2-scale-005 Advertising bottomed at 1.2% of US GDP in 1944. C e7-scale-002 ~1.32% of US GDP in 2025 — different basis, not commensurable with the old series. The same warning applies to advertising's size relative to the whole economy. On the long Coen series, advertising peaked at 3.0% of US GDP, the total output of the economy, in 1922, and bottomed at 1.2% in 1944. A 2025 reading puts it near 1.32% of GDP. But that last figure rests on a different basis and cannot be lined up against the old series. The apparent fall is partly real, and partly just a change in what gets counted as advertising at all.

2025 · BACK TO THE HOLE

Back to the hole

The count is still a hole, and that hole is where we began.

>$26B the IAB's own estimate of misallocated US media spend a better count could unlock in a year or two — recoverable, not settled. (e7-measurement-005, B) The industry knows the count is broken. The IAB, the online-advertising trade body, estimates that better, AI-driven measurement could recover more than $26 billion of US ad money now being spent in the wrong places, within a year or two. That figure is the industry's own guess at how much the measurement failure is misdirecting. It is an estimate of what might be recovered, not a settled fact.

A e6-measurement-006 ~18% YoY paid-click growth, Q4 2008 — no count, CPC, or query total disclosed. B e7-events-008 Pew: result clicked on 8% of AI-summary visits vs 15% without — Google disputes the study. B e7-targeting-004 AI Overviews >2B monthly users by Jul 2025; CBO says it monetises at ~the same rate as search. Go back to where this began. Google still reports how fast its paid clicks are growing, not how many there are. And the old pattern is reappearing in the new answer box, the AI summary that now sits at the top of a search page. The Pew Research Center found that users clicked a search result on 8% of visits where one of those AI summaries appeared, against 15% of visits where none did. Google disputes that study, so both figures stand, unresolved. Google's AI Overviews reached more than two billion monthly users by July 2025, and Google's chief business officer said the format makes money at roughly the same rate as ordinary search. That, again, is Google's own account, not an audited fact.