OpenAI's Ads Problem Isn't Speed—It's Everything Else Google Already Solved

2026-08-16 — business growth strategy India

Six weeks. That's how long it took OpenAI to cross $100 million in annualized ad revenue after launching its advertising pilot on February 9 for logged-in adult users in the United States. Most of us in this business read that number and saw validation. I saw a company moving so fast it's creating problems faster than it can solve them.

The speed part is real.

Most platform launches don't move like this. More than 600 advertisers were already running campaigns by the time the self-serve Ads Manager launched in April and May 2026. We're talking about Best Buy, Target, Expedia, Adobe, and Ford. Enterprise brands with serious budgets. The problem with watching this unfold is that OpenAI skipped the part where you actually get good at advertising.

Google took 15 years to build what it has. Google knows your search history. Google knows what time of day you search. Google knows whether you've clicked on a competitor's ad in the past week. That data lives in a system that's mature enough that nobody thinks about it anymore—it just works, most of the time, because it's been stressed and refined ten thousand ways.

OpenAI has 600 advertisers and a spreadsheet.

Actually, that's not quite fair. They have cost-per-click bidding now. But the friction is everywhere if you look. OpenAI is actively hiring for a head of measurement, which tells you something stark: they're running ads at scale without a mature way to prove they work. Compare that to Google, Meta, Amazon. These companies spent years building clean rooms with third parties, establishing industry standards for attribution, moving incrementally from CPM to CPC to ROAS-based models.

OpenAI is doing it in six weeks.

The click-through rates are instructive. Early CTR sits at 0.91%—which is seven times lower than Google. That's not a feature of conversational AI being worse at ads. That's a sign that nobody knows what a good ChatGPT ad actually looks like yet. The brands are guessing. OpenAI is watching the guesses land. They're learning in real time, which is fine for them—OpenAI gets paid either way. It's less fine for the advertisers who spent six figures figuring out that their targeting is loose.

Here's what keeps me awake about this:

OpenAI priced these ads at roughly $60 per thousand impressions, or three times the typical Meta CPM. They justified that with conversion data from Criteo, which claimed users referred from LLM platforms convert at 1.5x the rate of other channels. That's a credible number. But it's also a number from a company trying to sell something. Everyone in our world has learned not to trust early-stage platform conversion claims, and yet here we are, with brands paying NFL broadcast rates to run text ads inside a chatbot because the signal from conversational intent looks strong.

It probably is strong. But nobody's built the infrastructure yet to understand what strong actually means in this context.

The self-serve launch was supposed to solve this—lower the barrier, let smaller brands test, get real-world volume. They removed the $200,000 minimum spend, which is good. It's also not the hard part. The hard part is that OpenAI is trying to build measurement, targeting, audience segmentation, brand safety, fraud detection, attribution, and everything else Google spent two decades perfecting—all at once, under public scrutiny, with every advertiser who's spent money watching their performance numbers.

None of those are unsolvable problems.

But they're being solved on a schedule that prioritizes growth over maturity. The $100 million milestone was announced. The friction points—the stuff that makes this hard—are still being built in public. Google never had to do that. Google built in the dark and then opened the doors when it was already reliable.

OpenAI is opening the doors and building while people are walking through them.