OpenAI launched advertising in ChatGPT in early February, charging roughly $60 per thousand impressions and requiring a minimum commitment of $200,000. Six weeks in, the early advertisers have a problem: none of them can demonstrate that the campaigns drove measurable business outcomes, according to a report from The Information.
The $60 CPM puts ChatGPT ads in the same pricing tier as live NFL broadcasts and premium streaming inventory. Meta's average CPM sits around $10 to $20 depending on who's counting. Google Search ads run about $38. So OpenAI is asking for a premium that would make sense if the measurement infrastructure matched the price tag. It does not.
What $200,000 buys you
Advertisers in the pilot get impressions and clicks. That's it. No conversion tracking, no pixel installation, no attribution modeling. The reporting arrives as weekly CSV files, and buying the ads still involves phone calls and emailed spreadsheets, per The Information's reporting. For context, Google has spent two decades building systems that can trace a user from ad impression to checkout. OpenAI is selling at three times the price with roughly none of that plumbing.
Three of the world's largest agency groups are in the pilot: WPP, Omnicom, and Dentsu. CNBC reported that some brands committed between $200,000 and $250,000 to the test, which is about double a typical experimental ad buy. But the budgets are sitting largely unspent. The Information found that halfway through the pilot, advertisers had deployed only 15 to 20 percent of their allocated spend, which points to a straightforward inventory problem: there just aren't enough ad slots to fill.
The inventory squeeze
OpenAI is showing ads to a tiny fraction of its user base. Sensor Tower estimated that by mid-March, ads had rolled out to roughly 5 percent of ChatGPT's mobile users, up from 1 percent at the start of the month. Volume did jump about 600 percent during that stretch, but 600 percent of a small number is still a small number.
"We're in the early testing phase of ads in ChatGPT, and the goal right now is to learn and refine the experience for consumers before expanding it more broadly," OpenAI told CNBC. The caution is deliberate, and some agency sources acknowledged that the slow approach is probably the right call for building a sustainable ad business. But it creates an awkward dynamic: advertisers locked up six-figure budgets for a test that isn't generating enough data to evaluate.
Dentsu's EVP of Paid Search Meredith Spitz put it diplomatically, saying the firm is "eager to partner with OpenAI to further test, learn and evolve the offering." She noted that ad delivery is "quickly building momentum, with volume increasing week-over-week." That reads like patience on a timer.
So what does Criteo actually change?
On March 2, Criteo announced itself as the first ad tech partner to integrate with the ChatGPT pilot. The press release hit all the right notes about discovery and trust. Criteo CEO Michael Komasinski called it "an exciting step forward," which is exactly what CEOs say about partnerships that haven't proven anything yet.
The practical pitch is more interesting. Criteo claims that users referred from LLM platforms convert at about 1.5 times the rate of other referral channels, based on a sample of 500 U.S. retailers observed in February. That's a promising signal, though it measures what happens after someone clicks through to a retailer site, not whether the ad itself was worth buying. And 500 retailers in one month is a thin dataset for a company activating $4 billion in annual media spend.
What Criteo doesn't bring: programmatic auctions, real-time bidding, or the kind of granular measurement that performance marketers need to justify spend. eMarketer noted that ChatGPT's ad model remains impression-based with high CPMs tied to views rather than clicks, making ROI assessment difficult. OpenAI has said it won't share user conversation data with ad partners and has no plans for programmatic buying. Criteo, for now, is mostly a familiar interface layered over an unfamiliar product.
The real test is the renewal
Truist analysts called 2026 an "inflection year" for LLM-powered ads and estimated OpenAI will generate under $1 billion in ad revenue this year. By 2030, that figure could exceed $30 billion if the thesis holds. Those projections assume the measurement story improves and the inventory scales, neither of which is certain.
Selling a first ad campaign isn't the hard part. Agencies were eager to test; the FOMO alone guaranteed initial commitments. The hard part is getting the renewal. And to get the renewal, advertisers need to show their CMOs something more convincing than an impression count on a spreadsheet emailed once a week. Right now, they can't do that.
OpenAI's ad pilot runs through the end of March. What happens after that, whether budgets scale or quietly get redirected to Google and Meta, will say more about ChatGPT's advertising future than any analyst projection.
