OpenAI opened its self-serve Ads Manager to U.S. Advertisers this week. No minimum spend, cost-per-click bidding next to the original cost-per-impression option, and the big holding company agencies already lined up as partners. Three months ago this was an invite-only pilot. Ads started showing to ChatGPT’s Free and Go users in February, and to get in you needed a six-figure commitment and a willingness to pay premium CPMs. Reuters reported the pilot hit a $100 million annualized revenue run rate in its first six weeksf, and OpenAI is telling people it wants $2.5 billion in ad revenue this year and $100 billion by 2030. Ambitious is one word for it.
I’ve seen a version of this movie before. Back in 2010, when Yahoo handed its paid search over to Microsoft, we wrote about getting your Bing campaigns ready for the switch and then about everything that went wrong in the first ten days. Reporting lagged, match types behaved differently than promised, and nobody could quite tell you why traffic looked the way it did. The Advertisers who did well were the ones who tested early, kept budgets small and paid attention, while everyone else waited for somebody to tell them it was safe.
There’s a nice bit of symmetry here, too. Two and a half years ago we wrote that Bing now ran on ChatGPT. Now ChatGPT is building its own ad business, and in a lot of ways it’s taking a run at the same search budgets Bing has been chasing since 2009. I’m not sure what to make of that yet, other than that it’s kind of funny.
Why I’m Interested Anyway
The case for ChatGPT Ads is pretty simple. Somebody typing “best water heater” into Google gives you three words. Somebody asking ChatGPT to compare two tankless water heaters for a family of five in a cold climate, with a budget, is telling you almost everything. If an ad can show up in that moment, and actually be relevant to it, that’s more intent than a keyword has ever carried. Google knows that too, which is a big part of why it’s racing to put ads into AI Mode.
The case against is also pretty simple. Measurement is brand new, the reporting is thin, and the early pricing was set for brands that wanted to be first more than for brands that needed a return. I’d also keep in mind who’s actually seeing these. Ads go to the Free and Go tiers. Paying subscribers don’t see them, and those are probably some of your heaviest, most engaged ChatGPT users. That doesn’t make the audience bad. It just isn’t everyone, and it’s worth remembering when someone shows you a click-through rate and calls it a benchmark. There’s a brand safety question nobody has a good answer to yet either. Your ad sits next to an answer the AI wrote, and you don’t control what that answer says about you or your competitors. OpenAI says ads won’t change the answers. I believe they’re trying. I’d still look.
How We’re Going To Test It
Small and deliberate. We’re picking one or two clients with clean conversion tracking and an offer that already performs on Google Search, so we have something real to compare against. We’ll buy on CPC rather than CPM, because paying for impressions on a brand new channel with immature reporting feels like paying for the privilege of not knowing. And we’ll judge it on cost per lead or cost per sale next to Google for the same offer, not on click-through rate or the platform’s own view of how things went.
If it works, great, we’ll know early and scale it before the auction gets crowded. If it doesn’t, we’ll have spent a little money learning that, which is a lot cheaper than finding out a year from now after everyone else has. That’s more or less the same thing we told people in 2010. It held up then.
Update, September 2026: OpenAI said at the end of August that ChatGPT Ads hit a $1 billion annualized revenue run rate less than 200 days after launch, with tens of thousands of Advertisers in more than 40 countries, and it’s opening self-serve buying across India, Europe, the Middle East and North Africa. Our own testing has yet to see much promise, as we serve primarily performance-based advertising. Those results echoes others that have yet see direct response success. Technologies evolve over time, Facebook Ad Manager v1.0 experienced the same growing pain at initial launch.


