OpenAI will test labeled visual ads during ChatGPT image generation in the US

OpenAI says it will test visual ads while ChatGPT generates images. Testing begins later this month in the United States with an initial group of advertisers. The ads will be clearly labeled and kept separate from the image being created. OpenAI said this in a format note paired with an ads post dated 5 October 2026. The posts give no calendar day for the first impression.

The format is images of product inspiration, of a product in use, or of experiences a service makes possible. OpenAI says that, as with its other ad formats, advertising does not influence the answers ChatGPT provides. Labeling, separation from the generated image, and answer independence are the company’s statements about how the test is built.

Digiday reported the same day that the unit advertisers have used for the past seven months carries the advertiser name, logo, headline, description, landing page, and an image. On that account, the October change is a visual placement during image generation. OpenAI’s line is that the ad stays separate from the image ChatGPT creates.

Measurement

The ads post expands in-house tools and partner measurement. Hightouch, Tealium, and LiveRamp are named so advertisers can send conversion data from systems they already use. The post says pixel and Conversions API coverage includes purchases, leads, installs, and sign-ups.

The format note and the ads post name the same attribution partners: AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge, and Tenjin. Both name Fospha, Measured, and INCRMNTAL for full-funnel work. Geo-based incrementality is early, with Haus, Measured, and WorkMagic. The ads post alone adds early-stage brand-resonance tests with Kantar and Cint.

OpenAI attributes the case studies to partners. One figure is its own early analysis.

• DV Rockerbox: WeightWatchers’ attributed cost per acquisition on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark.
• WorkMagic, on the format note: statistically significant lift for Dose, with 67% of incremental purchases from net-new customers. WorkMagic, on the ads post: a geo-lift study estimated 2.3x more incremental orders for Dose than last-click attribution captured. The two Dose lines measure different things.
• Triple Whale, on the format note: 93% of Portland Leather’s visitors from ChatGPT Ads were new.
• OpenAI, on the ads post: in an early analysis of global campaigns optimized for deeper-funnel outcomes, 52.7% of eligible one-day view-through conversions occurred within an hour of the matched ad impression.

The ads post says conversions already show up after an impression without a qualifying click, and that a one-day view-through window can be added in reports. That setting, OpenAI says, changes the report. It does not change how the campaign optimizes.

Brand safety and conversation access

Placement guardrails, OpenAI says, assess whether a conversation is appropriate for advertising, to keep ads out of emotionally vulnerable, sensitive, or otherwise unsuitable contexts. The posts do not publish how often that check blocks an ad. Qualifying advertisers can add Negative Phrases when a brand’s own rules are narrower.

Pilots with DoubleVerify and Integral Ad Science are for those firms to assess how the brand-safety standards are applied in a controlled testing environment. OpenAI says the pilots do that without accessing private user conversations. The 9 February 2026 post that opened the ads test said conversations stay private from advertisers, that ads are labeled sponsored, and that they sit visually apart from the organic answer. The October path for Hightouch, Tealium, and LiveRamp is conversion events advertisers send in.

Scope of the older ads test

The 9 February 2026 post limited the original US test to logged-in adults on Free and Go. Plus, Pro, Business, Enterprise, and Education were outside that test. A 26 March 2026 update said expansion beyond the United States would start in the coming weeks, beginning with Canada, Australia, and New Zealand. An 11 August 2026 update said ChatGPT Ads had launched in the United Kingdom, Mexico, Brazil, Japan, and South Korea. The October format note keeps the visual unit in the US, later this month. It does not restate subscription tiers for the new placement, and it does not name the advertisers in the initial group.

Audience size is OpenAI’s figure. The October posts say ChatGPT reaches 1.2 billion people each week, and they call it the world’s largest AI-native consumer platform. The February post said hundreds of millions of people.

People buying ChatGPT ads can treat the visual unit as a US test with an October window and no start date, and can treat the CPA, lift, and new-visitor figures as partner results OpenAI chose to publish.