OpenAI will test ChatGPT image ads this month. DV and IAS will check brand safety
OpenAI will test image ads in ChatGPT in the US later in October and is adding measurement and brand safety partners.
Photo: Sanket Mishra / Pexels
OpenAI said on 5 October that it will start testing ChatGPT image ads in the US later this month. The ads will appear while a user is creating an image. They will carry a label and stay separate from the picture ChatGPT makes. Only a small group of advertisers will join the first test.
The same post adds new measurement partners and two brand safety pilots. It answers two complaints buyers have raised since ads went live in February: few formats and weak measurement. In recent weeks, agencies kept ChatGPT budgets at test level because they could not prove results.
OpenAI links ChatGPT Ads to LiveRamp, AppsFlyer and Triple Whale
Advertisers can now send their sales data through Hightouch, Tealium and LiveRamp. For attribution, OpenAI named AppsFlyer, Adjust, Branch, Singular, Kochava, Airbridge, Tenjin, Northbeam, Triple Whale and DV Rockerbox. These tools track sales and sign-ups on websites and in apps, not only the last click.
For bigger questions, Fospha, Measured and INCRMNTAL offer full-funnel models. Haus, Measured and WorkMagic run geo tests. In a geo test, ads run in some regions and not in others, and the gap in sales shows what the ads added. Digiday reported that this part is still at an early stage.
WeightWatchers paid 15.3% less per customer, OpenAI says
OpenAI shared three results from advertisers, each measured by a partner. WeightWatchers paid 15.3% less per new customer than in its paid search benchmark, based on DV Rockerbox data. At Dose, 67% of the extra purchases came from new customers, according to WorkMagic. Portland Leather said 93% of visitors from ChatGPT ads were new to the store, based on Triple Whale. OpenAI picked these examples, so they are not an independent study.
DoubleVerify and IAS test ChatGPT without reading chats
DoubleVerify and Integral Ad Science will run separate pilots. They will judge brand safety in a controlled setting. OpenAI says neither firm will see private conversations. Some advertisers also get “negative phrases”, a list of topics their ads should stay away from.
Supply is the third problem, and it is not solved. A Zeno Group executive told Digiday that ChatGPT gets about 2.5 billion prompts a day, against about 15 billion Google searches. OpenAI says ChatGPT reaches 1.2 billion people a week. Its ad business passed a $1 billion annual run rate earlier this year.
Before the image test opens, check which of these tools your team already pays for. A brand that uses AppsFlyer, Triple Whale or LiveRamp can send ChatGPT sales data without a new contract. It is also worth writing the list of topics for the negative phrase filter now, not after the first campaign runs.