OpenAI готовит тест визуальной рекламы при генерации изображенийOpenAI plans a visual ad test during image generation

OpenAI announced a new visual advertising format on October 5. Its initial test is planned for later this month in the United States with a limited advertiser group, during image generation in ChatGPT. The company says ads will be labeled and kept separate from the image being created. This is a pilot announcement, rather than a release to all users.

The placement is the interesting part: someone is making an output while a commercial suggestion appears alongside the process. Our editorial question is how easily a user will distinguish paid material from the assistant's response. It is too early to assess the interface or audience reaction before the test. The stated principles are commitments to examine against the product's behavior.

In a separate measurement explanation, OpenAI describes event signals through its Pixel and Conversions API, covering purchases, leads, installs and registrations. Advertisers can select a click attribution window and include a one-day window for conversions following an impression without a click. The company explicitly separates these reporting settings from campaign optimization. That matters because different counting rules can change how the same purchase appears in a report.

The company also describes early work with Haus, Measured and WorkMagic on geographic experiments. Attribution asks which impressions or clicks a reporting system credits with an outcome. Incrementality asks how much additional business advertising caused. Our interpretation is that a purchase soon after an impression does not settle the second question. A shopper may already have intended to buy, and that contribution cannot be separated without an appropriate comparison.

OpenAI's public advertising policies, updated on September 10, require ads to be distinguishable from the ChatGPT interface. Review covers the advertiser, creative and landing page, and placement context, using automated systems with human oversight. The document also anticipates that an ad violating policy could get through and require investigation and response. A published rule is consequently a different claim from error-free enforcement.

The new announcement names developing brand suitability pilots with DoubleVerify and Integral Ad Science. OpenAI says partners would evaluate a controlled environment without accessing private user conversations. The announcement does not provide an independent report of a completed audit of those pilots. It should be read as the platform's description of its intended approach, with the future work kept in the future tense.

Why it matters

Users and advertisers have different tests of success. Users need to recognize a commercial placement and understand why a suggestion appears. Advertisers need to separate credited outcomes from demonstrated additional sales. Those questions deserve separate checks: readable labeling does not prove campaign effectiveness, and a strong campaign metric does not prove that an interface was clear. The October pilot could produce evidence for that discussion; today's announcement establishes its starting conditions.