OpenAI is done asking nicely.
The company has told its advertising partners that the pilot program running inside ChatGPT will continue past April, and it wants real money committed in return. Advertisers are now being asked to sign insertion orders — binding spend agreements covering two-month periods — a sign that what began as an experiment is being reclassified internally as a business.
"We are being asked for an IO commitment, and the expectation of the IO going over a two-month period," said Jai Amin, chief strategy officer at Jellyfish, which was briefed directly by OpenAI on the new terms. The shift matters. An insertion order is not a test. It is a guarantee.
OpenAI confirmed the ad program extension in a blog post, also telling AdWeek that it is expanding the footprint beyond the United States to Canada, Australia, and New Zealand in the coming weeks. That international push arrives alongside a surge in volume. Sensor Tower data shows a roughly 600 percent jump in impressions between early and mid-March. Dentsu has reported consistent week-over-week growth. Omnicom has placements running for more than 30 clients.
The numbers sound like momentum. The numbers are incomplete.
The critical metric that every advertiser cares about — what happened after the ad was served — is broken in critical ways. A technical glitch in OpenAI Ad Manager has been preventing advertisers from accessing their own campaign data. One advertiser who spoke to AdWeek about the issue had burned through just 3 percent of a 250,000 dollar budget and had no functioning dashboard to assess whether any of it had worked. The campaign was running, the money was flowing, and the advertiser was flying blind.
This is not a minor backend inconvenience. Measurement is the mechanism by which an ad product proves its worth. Without it, the $60 CPM that OpenAI is charging — a figure consistent with what was reported in an earlier type0 story — cannot be evaluated against alternatives. Advertisers cannot calculate return on ad spend. They cannot optimize creative. They cannot know whether the audience they reached was the right one.
The click-through rate compounds the problem. Some advertisers reported click-through rates as low as 0.91 percent, according to figures reported by AdWeek — a floor figure for the worst-performing campaigns, not a representative average. Google search product averages around 6.4 percent. The gap reflects something structural about how people use ChatGPT. Users are there to complete a task: answer a question, write something, work through a problem. They are not browsing. They are not in a discovery mindset. The advertising format assumes a passive audience; ChatGPT has an active one.
The current ad test is running for logged-in adult users on the Free and $8-per-month Go subscription tier. Plus, Pro, Business, Enterprise, and Education accounts will not have ads in this test. That is a meaningful constraint. The most engaged, most frequently using users are off-limits. The audience that remains is self-selected by willingness to pay for a lower service tier — which raises its own questions about what kind of user accepts an ad in exchange for a free or discounted product.
None of this means the ad business is failing. Volume is growing, the pricing has held, and the IO ask signals that OpenAI commercial team believes it can now sell certainty rather than possibility. The company needs the revenue. OpenAI is burning cash at a significant clip, according to widely reported industry estimates — a financial reality that makes guaranteed ad minimums, even at a high CPM with poor measurement, more attractive than a per-impression model that advertisers can abandon at any time.
But predictable revenue and valuable revenue are different things. Agencies are placing clients into the product — Omnicom 30-plus active placements is a real number — but several of those same agencies are also waiting for the reporting layer to be fixed before they recommend increasing budgets. An insertion order is a commitment to spend. It is not a commitment to understand what the spend accomplished.
The pilot has extended. The audience is expanding. The measurement is not.
