less than two hundred days
ChatGPT’s ad business hit a billion-dollar run rate faster than any ad product I can find a comparison for. The revenue is not the interesting number. The mix is.
Sensor Tower’s measurement, disclosed by Digiday on 31 August: ChatGPT advertising is at roughly $83 million a month, which annualizes to a billion. The pilot started on 9 February. That is under two hundred days.
The intermediate marks are worse, or better, depending on what you own. A hundred million annualized by 26 March — six weeks in. Self-serve Ads Manager to all US businesses on 5 May. The UK, Japan and South Korea in July. Brazil and Mexico in mid-August. Thirty-one European markets before the month was out.
That is the number everyone quoted. Here is the one nobody did.
Between April and August, the category mix inverted.
Shopping and retail went from 37% of spend to 21%. Financial services went from 2% to 13%. Travel went 5% to 11%. Health and wellness, 1% to 5%. Software crept from 13% to 16%.
Retail halved. Financial services went up more than sixfold. In April, the only financial services advertiser anywhere in the top hundred was Intuit. By August, four of the top ten spenders on the entire platform were financial services companies.
An ad platform’s category mix is the most honest description of what the surface actually is, because it is a revealed preference bought with real money by people who measure. Advertisers do not have opinions about a medium. They have attribution data.
And what the data said, apparently, within about four months: this is not a shopping aisle.
The initial assumption was obvious and wrong. Product queries, purchase intent, a recommendation engine you can pay to be inside — build the catalog and take a cut. That is what 37% retail looks like. It is what everyone building on a search-shaped surface assumes, because search is where you go when you already know you want a thing.
What replaced it is the set of categories where the purchase is expensive, infrequent, and preceded by someone explaining something to you. Mortgages. Brokerage accounts. Insurance. Trip planning. Health decisions. Categories with enormous customer acquisition costs, long consideration windows, and an existing habit of paying humans to sit across a desk and answer questions.
The surface does not behave like a search box. It behaves like an advisor’s office, and the industries that already budget for advisors found it in a quarter.
The growth engine is the part I would watch, because it is the part that runs out.
Ads per user per hour rose 163% from April to August — averaging 26% month over month. Unique US advertisers reached about 1,200 in August, growing 43% a month; Adthena counted 7,378 distinct advertisers globally in a single week in July. A new carousel format launched in August and was already in roughly 23% of US desktop ads by the second half of the month.
Read those two rates against each other. Advertiser count is growing fast. Density per user is growing faster, and density is the term you can turn up unilaterally. Roughly half of this trajectory is not more people wanting to buy ads. It is more ads being shown to the same people.
Every ad-supported product in history has had that dial and every one of them has turned it. The dial has a hard stop, and the stop is not a business metric — it is the point at which the thing stops feeling like an answer and starts feeling like a result page. Nobody knows where that is on a conversational surface. It has never been measured because the surface did not exist.
They are going to find it empirically, in public, on a product that nine-figure numbers of people use as a thinking aid.
One more thing lands in January. The EU’s designation of ChatGPT as a very large online search engine carries a public advertising repository obligation — every ad, every advertiser, queryable.
The fastest ad business ever built is about to become the most legible one. Not because anyone chose that. Because a 2022 statute about search engines got there first.
Sources: PPC Land ↗ · OpenAI ↗ · Winbuzzer ↗ · TechAfrica News ↗ · StartupTalky ↗