ChatGPT is testing ads inside image generation

OpenAI will begin testing a new visual ad format during image generation in ChatGPT in the US later in October. The ads are supposed to be clearly labeled and kept separate from the image being created. Only an initial group of advertisers will join the first test. The company also says advertising will not influence ChatGPT's answers.

OpenAI says ChatGPT reaches 1.2 billion people each week. That makes the visual format more than another banner. It enters the moment when a user is actively forming a visual idea of a product, place or experience.

OpenAI is placing ads inside the act of creation

The new format is intended to show product inspiration, usage or the experience a service can provide. OpenAI will first place it during image generation rather than across every conversation. Users outside the US are not included in the announced test.

The company is also adding integrations for conversion data and partners for attribution, incrementality and brand measurement. Its list includes Hightouch, Tealium, LiveRamp, AppsFlyer, Triple Whale, DoubleVerify and Integral Ad Science.

Ad budgets follow proof that a conversation led to a sale

For advertisers, a new image beside an answer matters less than the ability to connect a purchase to a campaign. ChatGPT is becoming an ad system with the same kind of data pipes marketers already use across the web and mobile apps. A budget decision can therefore pass through familiar attribution tools instead of living as an isolated experiment with incomparable results.

The setting is different. The ad appears around a private conversation rather than next to an article or video. OpenAI is therefore developing brand suitability pilots with DoubleVerify and Integral Ad Science in a controlled environment without giving them access to private conversations.

Three attractive numbers do not make a market benchmark

OpenAI says WeightWatchers recorded a 15.3% lower attributed cost per acquisition than its blended paid search benchmark. For Dose, 67% of incremental purchases reportedly came from new customers, while 93% of Portland Leather visitors from ChatGPT Ads were reportedly new. These are partner results selected by the company, not an independent study. The announcement does not provide a common methodology, sample sizes or a range of outcomes across campaigns.

The brand suitability pilots are similarly preliminary. The promise that partners will not see private conversations matters, but there is no public account yet of exactly what they will test or whether the results will be released.

Public methods and placement controls will decide trust

The first useful signal will be campaigns with published methods, spend and comparable control groups. The second will be a report from the suitability partners showing how the system handles a sensitive conversation without exposing it to the advertiser.

If measurement remains a closed collection of selected case studies, ChatGPT will have an ad format without a trusted market. If OpenAI opens the methods to independent scrutiny, it could build a new advertising channel around conversation without turning the conversation itself into a shop window.

Lilith's verdict

OpenAI is hanging a price tag beside an unfinished image while building a register to trace the sale. Trust will depend on whether outsiders chosen by no one at OpenAI can inspect the ledger.