OpenAI's ChatGPT Ad Pilot Falls Short: Agencies Left With Unspent Budgets

OpenAI entered advertising with a bold pitch: access to hundreds of millions of users in a uniquely intent-rich environment. Advertisers lined up. The results were something else entirely.

Since the pilot launched in February 2026, brands have been dealing with what the industry calls underdelivery. Campaigns fell well short of their targeted impressions, budgets sat unspent, and results were nearly impossible to justify.

One advertiser spent US$2,500 of a US$250,000 commitment

That is not a typo. One early participant in OpenAI's ChatGPT ad pilot spent just US$2,500 of a US$250,000 commitment over four weeks. Another agency described going almost a full month after launch with barely anything happening.

A senior exec at a major agency told Digiday that OpenAI could only push through roughly US$100 per client per week until supply started opening up around mid-April.

The minimum commitment made things worse. Ad reps told clients the money should come from innovation budgets, funds set aside for tests that do not need to deliver a return. But a US$250,000 floor set expectations the inventory could never match. One agency called it "smoke and mirrors in hindsight. There was never enough inventory at launch to absorb that level of spend."

The cultural collision nobody warned agencies about

What happened next revealed something more fundamental than a slow rollout. When advertisers wanted their unspent budgets returned, OpenAI's team simply did not understand why that was a reasonable request. Instead of unwinding commitments directly, they shuffled clients in and out of the pilot to free up the spend.

Compare that to how Netflix handled its own rocky ad launch. When Netflix's ad-supported tier underdelivered in its early months, the company proactively offered money back. It became a case study in how to manage a difficult start and earned the platform significant goodwill.

OpenAI did not take that approach. Not because of bad intent, but because its executives genuinely did not see why they should. This is a company born in software subscriptions and cloud services. In that world, underdelivery is a product bug to fix in the next release. In media, it is a financial obligation requiring immediate action.

Ad tech analyst Shirley Marschall put it plainly: "It's the kind of mismatch you'd expect from a company that is still, in many ways, an advertising baby."

Delivery is improving. Trust is not moving as fast.

Things have gotten better operationally. Four of seven agency executives who spoke to Digiday said fill rates climbed 30 to 50% from what they were at launch. When ads did land, the results surprised people. One agency's client tracked conversions and found efficiency close to Google's non-brand search, which is a meaningful benchmark.

But a number of blue-chip advertisers have already told agencies they will not return to ChatGPT advertising without a trusted intermediary in place. They cited both the underdelivery and reporting that failed to meet basic standards. Better delivery numbers alone will not fix that.

The pilot was a cold-start situation: one ad format, limited inventory, and basic reporting. Advertisers who understood they were testing unknown territory had different expectations than those who assumed they were buying a primitive version of Google or Meta. Most were not expecting miracles. They were expecting more than US$100 a week.

OpenAI has since added CPC bidding, conversion tracking, and removed the minimum commitment when it opened a self-serve ad manager in May. Whether those moves rebuild trust with the brands that walked away is an open question.

As Marschall put it: "There are simply too few alternatives, and way too much curiosity about what comes next." The platform may find that fixing delivery is the easy part.