When ChatGPT ads launched on February 9, the model was simple and expensive: US$60 CPMs, a US$200,000 minimum, sold as brand exposure. On May 6, I showed you prices falling and cost-per-click switching on. In the five weeks since, OpenAI has released weekly updates and rebuilt the thing into something closer to a Meta-style performance and commerce platform.

OpenAI’s self-serve ads manager is live for U.S. advertisers of all sizes, making access and pricing much more accessible. On May 28, OpenAI turned on cost-per-action ads, so advertisers pay only when a user clicks, signs up, or buys — the model small performance advertisers actually want. 

The Information reported OpenAI is now courting small local businesses (car washes, dry cleaners) and has dropped spend commitments for some buyers, competing directly against Meta for spend. The entry barrier keeps dropping, too: Criteo cut its ChatGPT minimum from US$50,000 to US$10,000 a month and added a dollar-for-dollar media match.

The clearest signal is the move toward commerce. OpenAI struck a deal with Skai to bring in retail and search marketers who buy Amazon, Walmart, and Google but never touch a programmatic platform. 

It gave ads a visual upgrade with shopping formats that pull in price and reviews, stack into a carousel, and carry dynamic “shop now” and “book now” buttons. It opened the pilot to financial services brands like Robinhood, and on June 8 it expanded to the U.K., Japan, South Korea, Brazil, and Mexico while testing multi-advertiser placements sold through a second-price auction. 

Quietly, an April 30 privacy update now lets OpenAI share limited identifiers with marketing partners and receive purchase data back, softening its old “never sell” stance.

So who actually loses here? 

My original assumption was mostly Google, but almost everything OpenAI shipped this quarter is more Meta’s playbook: cost-per-action, a conversion pixel, ex-Meta leadership under Dave Dugan, and an open courting of small and local advertisers.

So the first budgets at risk are the lower-funnel performance and small-business dollars Meta has owned for a decade. The second is retail media: The Skai and Criteo product-feed deals plug ChatGPT into the commerce buying workflows that today run on Amazon, Walmart, and Google. 

The third, and the one that should worry Google most, isn’t brand-name search but the high-intent commercial query, someone asking ChatGPT to compare two SUVs rather than typing a keyword. Some of this is also net-new experimental budget while the pilot is young, but the direction of travel is performance and commerce, which points more to Meta and Google and less to the open Web’s display dollars.

So I’m not sure this should worry publishers too much in the short term. But in the longer term, the product-research moment is the highest-intent slice of the funnel, exactly the territory Wirecutter, Hearst, Condé Nast, and any publisher with affiliate or commerce revenue depends on. 

OpenAI is reportedly heading toward an IPO that could value it near US$1 trillion and is telling investors ad revenue will go from about US$2.4 billion this year to US$102 billion by 2030, making up more than one-third of its revenue. Yet Digiday’s review found early execution defined by underdelivery and thin inventory, and ChatGPT’s daily time per user actually fell 18.3% between early March and mid-May. 

That raises real questions and should prompt real skepticism and a reminder there are a lot of chapters of this book still to be written. 

Step back, and the pattern that matters most to publishers is this: OpenAI is becoming another large, self-serve, automated, closed-loop ad platform, taking its place alongside Meta, Google, Amazon, and increasingly Netflix and Pinterest.

They all compete on the same three things: self-serve buying, automation, and closed-loop measurement at a scale no single publisher can match, which makes the long-tail, low-CPM open display market harder for publishers to compete in, not easier. 

It’s why publisher growth keeps showing up elsewhere: in events, branded content, and partnerships — the formats these platforms can’t self-serve. 

OpenAI’s commerce play is the bigger threat because that budget sits at the product-research moment publisher commerce and affiliate content was built to own. I lean toward the idea that the defense is to go upstream:

1. Protect and deepen the original review and recommendation journalism AI engines actually cite, and make it structured enough to be cited (clean product schema, machine-readable pages) so you’re the named source rather than the disintermediated one.

2. Compete on performance only where you’re genuinely differentiated. Compete instead on outcomes, first-party intent data, and premium context — not by chasing the platforms on commodity and scale (though I remain very open to seeing successful inventory bundling and alliance strategies take shape).

In other words, to use an Americanism, when everyone else zigs, it’s probably time to zag.

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

Banner photo: Adobe stock image by Thaspol.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .