Advertisers treated OpenAI’s ad pilot like a bet on the future. The question now is how long they’re willing to wait for it to pay off.
So far, it has not been an easy wait. Since the pilot launched in February, advertisers have reported chronic underdelivery — campaigns falling well short of their targeted impressions, leaving budgets unspent and results hard to justify, according to several ad execs, who all asked to remain anonymous due to the sensitivities around the test.
One of them spent just $2,500 of a $250,000 commitment over four weeks, receiving 200 clients in the process. The others described similar experiences, with another agency exec saying that when their first ads went live in February, almost nothing happened for close to a month. The inventory was, to put it plainly, barely there.
The frustration came into sharp relief at last month’s Digiday Programmatic Summit in Palm Springs.
A senior exec from a major agency told Digiday that OpenAI could only push through roughly $100 per client per week until supply began opening up around mid-April. In a town hall session, attendees were more blunt. “I think they got revenue blind and did a lot of things before they were ready,” said one. Another put the problem in structural terms “Less than 10% of the eligible base was part of the pilot. We just had massive scale issues in general.”
The $250,000 minimum commitment didn’t help. For all that OpenAI’s ad reps were clear that the money should come from innovation budgets — funds set aside for tests that don’t need to deliver an immediate return — the number itself set expectations that the inventory was never going to meet. One agency in the pilot from the start called those minimum spend commitments “smoke and mirrors in hindsight. There was never enough inventory at launch to absorb that level of spend.
That is starting to change. Over the last six weeks, four of the seven ad execs who spoke to Digiday said they were seeing more ads served in the app and, as a result, spending more. The pattern they described was consistent: a user asks ChatGPT a question, gets an answer and an ad. Ask a follow-up, get another. The fill rate is climbing — as much as 30% to 50% to what they were at the launch, said a representative of one ad agency.
When the ads do land, the performance has surprised people. One agency’s client, which tracked conversions via URL redirects before OpenAI’s own measurement tools were available, found efficiency close to Google’s non-brand search. For a CPM-only product in its first weeks, that was not the outcome anyone had modeled for. The intent environment, it turns out, is genuinely different. Users aren’t searching and clicking the way they would normally. They’re mid conversation weighing options and receptive in a way that passive browsing rarely produces.
It’s data like that OpenAI ad execs can point to when making the case for patience. For them, the underlivery is the cost of doing this properly. Before an ad is served, the platform filters for sensitive conversations, then for contextual relevance and only then considers the auction dynamics. The bar at each step is intentionally high. Get it wrong and OpenAI doesn’t just lose advertisers, it could lose users, and with them the entire premise of the platform.
Whether marketers are sympathetic to that depends on how clearly they see what’s going on. Building an ad business without breaking the trust that makes the product worth advertising in the first place is genuinely hard. The advertisers who understood that went in with different expectations than those who didn’t. Most weren’t expecting miracles. But they were expecting more than this.
For some, the gap between the two was enough to walk away. A number of blue-chip advertisers have already told agencies they won’t return to OpenAI advertising without a trusted intermediary in place, citing both the underlivery and reporting that failed to meet basic expectations, said one ad exec who works with them. For hardcore performance advertisers — those for whom every dollar has to work as hard as possible — the pilot was never the right fit.
The pilot was still effectively in a cold-start phase when it started. One single ad format, limited inventory and basic reporting — meaning advertisers were very much testing the unknown, rather than having the basic mechanics that make performance marketing provable and deliverable.
There was a bit of an expectation mismatch; advertisers were repeatedly being told it’s a test and were treating it almost like an early-stage version of Google or Meta, while OpenAI was still very much building out the basics in real-time. Could this have impacted early perception? Sure. Underdelivery and limited transparency are not easily forgotten when advertisers want their dollars working hard. For some, it then sits in the bucket of “not ready” or “not worth it”, even if OpenAI has been very clear from the start.
None of that is lost on OpenAI. CPC bidding, conversion tracking and a pixel-based measurement API have all been added since launch, with CPA still to come. The self-serve ad manager, now open in the U.S., removed the minimum commitment altogether. Persuading advertisers of a clear direction rather than a company still working out where it wants to go, is the harder task.
Harder still because OpenAI’s own actions suggest it is still learning how the ad business actually works. When advertisers wanted unspent budgets released, OpenAI struggled to understand why that was a reasonable ask, according to the senior exec at a media network. Rather than unwind commitments directly, they had to quietly shuffle clients in and out of the pilot to free up the spend.
It was an avoidable friction, and a revealing one.
Netflix faced a similar situation when its own ad fell short of viewership guarantees in its early months, it proactively offered money back, earned significant goodwill and the gesture became something of a case study in how to handle a rocky start. OpenAI didn’t do that. Not out of bad faith, sources say, but because its execs simply didn’t grasp why they should.
For media companies, offering money back is almost muscle memory. OpenAI, by contrast, born from software subscriptions and cloud economics, approached the situation more like a platform trying to reshuffle the spend than a media company managing an underdelivery, said ad tech expert Shirley Marschall. “It’s the kind of mismatch you’d expect from a company that is still, in many ways, an advertising baby.
“Besides, with ads in chats becoming the industry’s latest obsession, this is neither the time nor the market for buyers to hold grudges over the awkwardness of an early pilot,” Marschall said. “There are simply too few alternatives, and way too much curiosity about what comes next.”
OpenAI did not respond to Digiday’s request for comment.