OpenAI is selling ads it can't measure, and advertisers are paying for vibes.

The Summary

  • OpenAI's first ChatGPT advertisers say they can't prove the ads work, with agency execs reporting zero measurable business outcomes from early campaigns.
  • The ad buying process is described as "low tech" with minimal performance data provided to advertisers.
  • OpenAI is scrambling to fix this before opening ad sales to more marketers next month, exploring partnerships with ad tech firms or building its own ad management system.

The Signal

This is what happens when you bolt a Web2 business model onto a Web4 product. OpenAI built the most sophisticated conversational AI in history, then slapped banner ads on it like it's 2005 Yahoo. The result is predictable: a measurement disaster.

Here's the tell. Two agency executives working with early ChatGPT advertisers admit they haven't proven any measurable business outcomes. Not weak outcomes. Zero outcomes they can measure. In 2025, when every pixel of the web is tracked six ways, OpenAI is asking brands to spend on faith. That's not a tech problem. That's a business model problem.

The ad tech infrastructure that makes Google and Meta money printing machines took a decade to build. OpenAI is trying to speedrun it in months because they need revenue, fast. But conversational AI breaks traditional ad metrics. There's no click-through rate when the interaction is a dialogue. There's no view-through conversion when the user asks follow-up questions across multiple sessions. The KPIs that matter in chat, intent signal strength, conversation depth, query refinement patterns, aren't things OpenAI has productized yet.

So they're selling ads the old way while figuring out measurement the new way. Advertisers are guinea pigs. Some will pay for early access and brand halo. Most will demand proof before the second insertion order. OpenAI knows this, which is why they're rushing to build or partner on ad tech before the broader rollout next month. The question is whether they can build attribution systems for agentic interactions faster than advertisers lose patience.

The Implication

If you're a marketer being pitched ChatGPT ads right now, ask for measurement frameworks upfront. What constitutes a conversion in a multi-turn conversation? How do they attribute outcomes when an agent researches on behalf of a user? If they can't answer, you're paying to help them figure it out. That might be worth it for brand positioning, but call it what it is: R&D spend, not performance marketing. For everyone else, watch how OpenAI solves this. The company that cracks attribution for agentic interfaces unlocks the business model for Web4. Right now, nobody has.

Source: The Information