On 28 August, MediaPost documented the results from agencies that had already run campaigns on ChatGPT: there were clicks, but no sales. On Monday 31 August, three days later, OpenAI announced that its advertising business had crossed its first billion. A few hours later, the European Commission designated ChatGPT a "Very Large Online Search Engine", placing it in the same regulatory category as Google Search.
None of these three stories says on its own what they say together.
A billion that hasn't been collected
The figure OpenAI announced is an annualized revenue run rate: you take last month's revenue and multiply it by twelve. It isn't money in the bank. It's a snapshot of today, projected into the future. The announcement itself says so plainly, as does Digiday, which ran the numbers.
Less than 200 days after launch, ChatGPT Ads counts tens of thousands of advertisers across more than 40 markets. For an ad product, that is unusually fast.
The context is less flattering. The $1 billion amounts to roughly 2.5% of the company's total $40 billion run rate, as CNBC calculates. And the target OpenAI has given its investors is $2.5 billion in 2026 and $100 billion by the end of the decade. The gap between one billion and a hundred isn't closed by momentum. It requires a compound annual growth rate of more than 200%, every year, without pause.
What the advertisers who paid are saying
This is where it gets interesting. A few days before the announcement, MediaPost gathered accounts from agencies that had already run campaigns.
Canada's Choice OMG spent $415 across three campaigns. It got 8,940 impressions, more than expected, and 58 clicks. That's a 0.65% CTR, when the same team typically sees 2% on search ads. Roughly $7 per click. Leads: zero. The clicks came from real people, not bots. Not one of them was interested in buying. The company published the full breakdown.
SE Ranking ran a larger test: 97,000 impressions, 1,263 clicks, a 1.30% CTR and barely any sign-ups. The two biggest ads accounted for about half of all impressions.
Nicholas Verity of Cleverly pulled his campaign early: a decent CTR, zero conversions. The numbers didn't even agree with each other. ChatGPT reported 57 clicks; Google Analytics showed fewer than 20 visits. Daniel Johnson of We Scale Startups summed it up in the comments on his post: until the platform's reporting becomes reliable, any spend there is a research budget, not a channel you scale.
The audience that doesn't buy
The obvious explanation is "it's early, the targeting tools only just launched." That's partly true: geotargeting, custom audiences and conversion optimization only arrived in August.
But there's a second explanation, and it's much harder to fix. SE Ranking put it into words, and it is, I think, the most important sentence in the whole story: the more someone looks like your ICP, the more likely they are to be on a paid, ad-free plan.
Ads on ChatGPT are shown exclusively to Free and Go users. Anyone paying for Plus, Pro, Business, Enterprise or Education never sees them. In other words: the professional with a budget, the developer with a company account, the marketer testing tools, the team shopping for software, everyone who has proven they're willing to pay for something, has already bought their way out of the inventory.
OpenAI built an ad network and then, through its own pricing, removed its most valuable viewers from it.
And the inventory is already filling up: according to Similarweb data, roughly one in four ChatGPT responses now contains a sponsored ad. Saturation is arriving before proof of performance.
Europe is the expensive part
On 24 August, ads expanded to 31 European countries. On 31 August, the self-serve Ads Manager opened to European advertisers. The same day, the Commission designated ChatGPT a Very Large Online Search Engine under the DSA, with 159.1 million monthly users in the EU against a threshold of 45 million.
What this means in practice: a public ad repository where anyone can see what ran, who paid for it and how it was targeted. Annual independent audits. A ban on targeting minors and on using sensitive data. Fines of up to 6% of global turnover. Compliance deadline: four months.
The biggest market in the expansion is also the most regulated. And the precedent isn't only about OpenAI: every future consumer assistant that crosses 45 million European users automatically falls under the same regime.
Why does it matter?
Because behind the numbers, what's being decided is which monetization model wins. There are two.
The first charges for attention: you gather a billion weekly users and sell access to them. The second charges for work: you get paid by the people who use the model to produce something measurable.
The second quarter of 2026 delivered the first serious scorecard. OpenAI posted $6.7 billion in revenue, up 18%, with operating losses widening from $9.3 billion to $12.3 billion. Anthropic, which has publicly committed never to put ads in its products, posted $11.6 billion and a small operating profit, overtaking OpenAI in revenue for the first time.
Both are heading toward an IPO. One of them will have to explain to public investors why the channel that's supposed to close the gap contributes 2.5% of revenue and is served mostly to people who have chosen not to pay for anything.
So the question isn't whether ads on ChatGPT will work. It's whether the "free product, paid attention" model, the one that built Google and Meta, transfers at all to an interface where access to the people with money is sold separately, to those very people.
