OpenAI launches visual ads in ChatGPT and hands measurement to the ad industry

OpenAI is adding a new format to ChatGPT Ads: a visual ad built around images of the product, its use, or the experience behind it. The format arrives together with the bigger move, which is measurement. OpenAI is integrating Hightouch, Tealium and LiveRamp so advertisers can push conversion data from their own systems into ChatGPT Ads, and it now supports a long list of attribution partners across web and app.

The company frames the bet with its own reach number: 1.2 billion people per week on ChatGPT, and ads as the way to monetize that audience beyond subscriptions.

The format: images, labeled, outside the answer

The format targets the moment of product discovery. Instead of a link buried in a text answer, the user sees an image that sells the possibility: the product in use, the outcome it enables. Per OpenAI, the point is to help people imagine the product fitting into their lives before deciding what to buy.

Testing starts later this month in the US with an initial group of advertisers. The rollout is contained:

• The format appears during image generation in ChatGPT, and only for users on the Free and Go plans.
• Ads are clearly labeled and kept separate from the image the user asked the model to create.
• Advertising does not influence the answers ChatGPT gives, the same rule the company states for its existing formats.

The list of measurement partners is the news

For an ad channel to make it into a media plan, the question advertisers ask is not "can I run ads here", it is "how do I measure them with the tools I already have". That is where this announcement spends most of its weight.

On the data side, the integrations with Hightouch, Tealium and LiveRamp let advertisers send conversion data out of their existing stacks. On attribution, OpenAI lists support across web and app with AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge and Tenjin, plus full-funnel and advanced measurement partners like Fospha, Measured and INCRMNTAL. Incrementality testing is at an earlier stage: geo-based experiments with Haus, Measured and WorkMagic.

OpenAI publishes three results, all reported by the partners themselves rather than by the company:

• According to DV Rockerbox, WeightWatchers' attributed cost per acquisition on ChatGPT Ads ran 15.3% below its blended paid search benchmark.
• WorkMagic reported a statistically significant lift for the wellness brand Dose, with 67% of incremental purchases coming from net-new customers.
• According to Triple Whale, 93% of Portland Leather's visitors who arrived through ChatGPT Ads were new to the brand.

The source matters. These are partner numbers, which is standard at this stage of a channel's life, but they are still early, and the real test is running them through your own stack.

Brand safety with outside reviewers

OpenAI describes placement guardrails that keep ads out of emotionally vulnerable or sensitive contexts, backed by automated review and human oversight. Brands with narrow placement policies can now use Negative Phrases. The newer piece is external review: brand suitability pilots with DoubleVerify and Integral Ad Science, where the two firms evaluate how OpenAI's safeguards hold up in a controlled testing environment, without access to private user conversations.

What to read out of this for a media plan

If you plan media, the practical read comes in two steps. The visual format is early and limited to image generation on the free plans, so it is a watch item, not a budget line. The measurement stack is the part you can act on now: if your conversion and attribution pipelines already run through any of those partners, ChatGPT Ads moves from "new channel" to "channel I can measure", and that is the gate incremental budget has to pass before it lands there.

Users on the paid plans, for their part, are not in the initial test. The ads that do appear are labeled, separate from the generated image, and they do not change the answers.