OpenAI has figured out what to do with the seconds users spend waiting on ChatGPT. The company announced Oct. 5 that it will test visual ads during image generation in the U.S. later this month. The ads will be labeled and kept separate from the image the user requested. OpenAI says they won't influence its assistant's answers.
Call them pause ads for chatbots. In CTV, a pause ad appears when a viewer pauses a show. ChatGPT could apply similar logic to the seconds its model takes to generate an image, using a lull in the experience to place a brand. OpenAI calls them "visual ads," not "pause ads," but the comparison helps explain the format.
One difference matters. Here, the user hasn't chosen to stop what they're doing. They're waiting for the service to finish. OpenAI is turning a product constraint – compute time – into a chance to serve ads. The idea is to make that wait help pay for the service, though how much of the cost ads can cover is still unclear.
From purchase intent to inspiration
What could make the format valuable is when the wait happens. Someone asking an assistant to visualize a project has already said what they want, what constraints they're working with and sometimes what it's for. Brands could get in before that interest turns into a product search.
Say a user generates a layout idea for their living room. For a furniture advertiser, that could be the right moment to show up. That's a hypothetical, though. The announcement doesn't detail any targeting mechanism.
That would widen ChatGPT's commercial reach. The assistant could appeal to brands at the stage where preferences form, not only when users are already comparing products. But a creative request isn't commercial intent. Generating a living room, a car or an outfit doesn't necessarily mean wanting to buy one.
The same caveat applies to attention. A user waiting on an image may watch the ad, switch tabs or put down their phone. Generation time creates a chance for exposure. Its media value has yet to be proven.
Measurement has to justify the budget
That's why the measurement news that came with the format matters so much. Customer data platforms Hightouch and Tealium are joining LiveRamp to help advertisers pass conversion data back. OpenAI is also expanding integrations with attribution vendors including AppsFlyer, Adjust, Triple Whale and Rockerbox. Geo experiments are being explored with incrementality testing firms Haus, Measured and WorkMagic. Research firms Kantar and Cint are taking part in early brand measurement tests.
Each piece answers a different objection. Matching sales to ads makes the channel workable. Seeing it in the measurement tools buyers already use makes it comparable. Measuring its incremental lift should show whether it deserves extra budget.
That's especially important for an assistant that can shape a decision without generating an immediate click. OpenAI also lets advertisers include a one-day view-through attribution window in reporting. The company says these reporting settings don't change how campaigns are optimized.
Buyers will have to resist a familiar shortcut. A sale that follows an exposure doesn't prove the exposure caused it. Wider attribution can better reflect the customer journey, but it can also inflate claimed conversions. Incrementality tests will have to tell the two apart. OpenAI acknowledges that work is still early.
Verification looks at the system, not the impression
Brand suitability is the other key issue. A "Negative Phrases" feature lets eligible advertisers specify placement restrictions. In pilots, ad verification firms Integral Ad Science and DoubleVerify will assess how OpenAI's safeguards are applied in a controlled environment. They won't have access to private conversations.
Buyers should be clear about what that covers. Stress-testing a system against different scenarios provides useful assurance. On its own, it can't reveal the exact context of each impression actually served.
Agencies will have concrete questions. What does the evaluation cover, how often is it renewed and which results can the advertiser see? Having a familiar name involved helps the conversation. But the protocol is what determines what anyone can conclude.
OpenAI is working on the two things its ad business needs to grow: more places to run ads, and a way for brands to justify buying them. The waiting time is there. Now OpenAI has to prove it's worth a line in the media plan.
