OpenAI plans visual ads during ChatGPT image generation and expands campaign measurement

OpenAI plans to test a visual advertising format during image generation in ChatGPT later this month in the United States. The company says ads will be labeled and separate from the image a user creates. The initial test involves a group of advertisers rather than a general launch of the format.

The October 5 announcement also expands conversion-data integrations and attribution partnerships. OpenAI is developing brand-suitability evaluation pilots with DoubleVerify and Integral Ad Science, bringing questions about campaign results and placement safety into the same release as the new creative format.

Images beside the creative workflow

OpenAI's visual ads and measurement announcement describes ads that show product inspiration, product use or the experiences a service could provide. The initial placement is within an image-generation workflow, where a person may already be considering how something should look.

The company says advertising does not influence ChatGPT's answers. That assurance and the separation between an ad and the generated image are central to understanding the proposed experience. A paid image beside a creative task should remain distinguishable from the output the reader asked the assistant to make.

Testing has yet to begin under the announced timetable. The company specifies later this month, the US and an initial group of advertisers; it does not establish a completed rollout across countries or a result from the new format. Marketers evaluating the announcement should keep those availability limits separate from the broader ChatGPT Ads platform.

Conversion integrations and attribution

Integrations with Hightouch, Tealium and LiveRamp let advertisers send conversion data from existing systems to ChatGPT Ads. OpenAI also names attribution partners supporting web and app measurement, including AppsFlyer and Triple Whale. The point for a business is to connect ad activity with its own measurement process rather than treating an impression as the end of the analysis.

OpenAI distinguishes attribution from its work on incrementality. It says the latter is at an early stage and names Haus, Measured and WorkMagic as partners exploring geographic experiments to assess causal advertising impact. That distinction is useful: assigning credit to an ad interaction and demonstrating that an ad caused additional business are different questions.

The announcement cites favorable customer findings from measurement partners. Those examples describe specific campaigns and methods; they do not establish that an advertiser joining the visual-format test will receive the same result. A business still needs its own objective and a consistent way to compare spending with outcomes.

Brand suitability without private conversation access

OpenAI says its placement safeguards assess whether a conversation is appropriate for advertising. It describes automated review, human oversight and ongoing monitoring, with Negative Phrases available to qualifying advertisers for narrow requirements tied to their own brand policies.

The DoubleVerify and Integral Ad Science pilots are intended to evaluate how OpenAI's brand-safety standards work in a controlled testing environment without accessing private user conversations. These are evaluation pilots in development, rather than a completed independent certification of all placements.

That design creates a specific question for the pilots to answer: whether a controlled evaluation provides advertisers enough evidence about where their ads can appear while preserving conversation privacy. The announcement identifies the proposed method, but does not provide finished pilot results.

Businesses can sign up for ChatGPT Ads now, according to OpenAI. For the new visual format, the next evidence to watch is the US test itself: how the labeling and image separation work in practice, which advertisers can participate, and what measurement reveals after a campaign runs. The announcement gives marketers a format to evaluate; it does not remove the need to judge performance against their own goals.