OpenAI Is Paying Someone to Figure Out If Its Own Ads Work

OpenAI's ad rates are falling, and the company does not yet have a way to prove that matters.

Ten weeks after OpenAI launched its ads pilot inside ChatGPT on Feb. 9, the cost per thousand impressions has collapsed from $60 to as low as $25, according to Digiday. That is a 58 percent drop in under three months. The minimum spend required to participate has fallen from $250,000 at launch to $50,000, Digiday reported separately, widening the pool of advertisers while inventory stays underpriced.

The descent is faster than what Netflix experienced. Netflix entered at CPMs of roughly $55 to $65 in 2022 and watched them fall to $20 to $30 within a year as inventory scaled, Digiday noted. The difference is that Netflix had years of user behavior data and a subscription business to anchor its pitch. OpenAI is still proving the basics: that a user who sees an ad in a ChatGPT conversation is worth reaching at all.

Performance advertisers have been effectively frozen out of ChatGPT until now. The pilot ran on a CPM-only basis, which favors brand advertisers comfortable buying eyeballs in bulk. The switch to cost-per-click, confirmed by Digiday, is meant to unlock that market. But performance buyers will not move budget without measurement. They want to know: did the click cause a purchase, or was the user going to buy anyway?

The measurement gap is the most concrete sign of what OpenAI still does not have. A job posting for the company's first Advertising Marketing Science Lead, still live on the careers page, asks for deep expertise in multi-touch attribution, media mix modeling, incrementality testing, and geo experiments — all to be built from scratch. Reading the listing straight, the qualification requirements are a checklist of infrastructure that does not yet exist.

OpenAI has a conversion tracking pixel, described by Shopifreaks as live for select advertisers, that fires events including lead created, order created, page viewed, subscription created, and trial started. But the event model has been noted as somewhat outdated compared to what sophisticated performance buyers expect from Google or Meta. The gap between what OpenAI has built and what the market requires is the exact problem the Marketing Science Lead is being hired to solve.

The $3 to $5 per click that advertisers are being asked to pay only makes sense if the clicks can be connected to business outcomes. Until measurement infrastructure exists, performance dollars will not flow at the scale OpenAI needs to hit its $102 billion ad revenue projection for 2030.

The broader context is that roughly 95 percent of ChatGPT users are on the free tier. Only about 5 percent pay for Plus or Pro subscriptions. That majority, the free users, are the impressions being sold. They are also the reason the revenue target is not optional. Sam Altman's company is not a niche research lab. It is a business with infrastructure costs that scale with usage, and free usage is expensive.

A leaked advertiser deck reviewed by AdWeek showed StackAdapt serving CPMs as low as $15, well below the Digiday-reported average — a sign some buyers are getting inventory at a steep discount, possibly because OpenAI is desperate to show volume growth before the measurement story is complete.

If the measurement infrastructure lands and the CPM decline stabilizes, OpenAI has a credible case to performance advertisers: you can reach users in a high-intent context, during an active conversation, at prices that have already compressed significantly from launch. If it does not land, the CPMs keep falling, the performance dollars never materialize, and the $102 billion projection stays on a slide. The Marketing Science Lead will have a short window to prove the clicks are real. OpenAI's next earnings call is expected in late May — investors will be looking for advertiser retention data and any update on the measurement infrastructure timeline.