Every few years, a new platform is launched and brands that usually take six months to approve a creative brief suddenly get an emergency budget. We’ve seen it with TikTok, which is influential, with connected TV. Now it’s possible with OpenAI ads. We’ve shared our thoughts on this with Knowledge recently, and the discussion should continue here.
Clients who have been wary of testing proven formats are ready to move quickly to something that, in many ways, has more unknowns than anything that came before it. The motivation is understandable. The need for search has been soft for a long time, and OpenAI feels like the place where that goal is going. But the sentiment of treating it like a like and replacing like with it should be tested before any budget moves.
Fear that drives you
Most of the urgency we see from customers right now comes from search concerns. They’ve watched their organic and paid search volumes flat or decline, they’ve heard about AI overviews eating away at click rates, and now there’s a platform where people can clearly ask questions and make decisions. The logic writes itself: get in front of them there.
The problem with that framework is that it takes work to repeat the search, just in a new location. That however sells the opportunity short and sets unrealistic expectations. The intent in the chat area is real, but it’s not the same as a search query. The context is different, the format is different, and the relationship between the user and the platform is different in ways that are important to how the ads will actually perform.
Why is this face really different
One thing that separates ChatGPT from Meta or Google, at least emotionally, is how personal the experience feels. People use it to think about decisions, solve problems, have conversations that they might not have with other tools. That quality of one person did not happen by accident. It’s a product. And it creates a dynamic that brands that advertise there will need to think about.
On a practical level, the ad space is not limited. The same question can yield a different answer depending on a thousand variables, and there are limited tools currently available to control the context in which an ad appears. Product safety, which many marketers take for granted and set strict limits on, is harder to ensure here than on any other platform. The ad may appear alongside relevant content but off, and the personality and tone of the chatbot at that time will change how the ad comes across. That’s a layer of uncertainty that isn’t in display or search.
Three things to consider before making a budget
CPM statistics are not static yet. At its core, this is a native, context-oriented format. That’s not a new type of media, and the industry has been buying similar placements for years, often at a fraction of OpenAI’s current prices. If the same inventory is available at a much lower cost, the question for any effective marketer is: what else do we get for the premium? Right now, that’s really hard to answer.
Rating is another big blocker. Without tools to quantify efficiency, ensuring investment becomes more difficult. A less effective method is to evaluate the increase in delay: divide the audience randomly, hold the control audience, spend enough time to get valuable data, and measure how the values differ between the two groups. That way you get a read that isn’t protected from what the ads are doing. It’s not the kind of indicator that most clients expect before measuring spend, but it’s the most reliable signal available right now.
Audience clarity is also important, and you never know whose intent you are buying. “Users with high intent” is a compelling pitch, but intent is only useful if you know whose intent it is. ChatGPT’s user base is not a monolith, and neither is the entire Internet. There is already a logical divide across LLM platforms, and the demographics can vary greatly between ChatGPT, Gemini, and Claude. Thinking you’re reaching a broad, high-value audience without that data is a leap.
How to test properly, if you’re going to test
The most logical framework now is to treat this as an experiment with a clear hypothesis, not an activation with an expected return. That means going in with some questions you want answered: which specifics work, where in the funnel does this format fit, what does CPM justification look like compared to comparable placements. It also means not letting internal structures get in the way. Whether this stays in the search budget or the planned budget is less important than making sure the right people are there and the tests are set up to generate learning.
Big picture of the LLM classification
One thing to keep in mind as OpenAI continues to build its ad product: it doesn’t have the market presence that Google and Meta had when they launched their ad platforms. There are real competitors, with great services with growing user bases, and the audience is already diverse on social media. The total audience that can be addressed for any one LLM is the set that it may appear to be, and that segmentation will continue.
That also means that the ad campaigns these platforms run to attract and retain users will be more fun to watch. How OpenAI, Google, and Anthropic each position themselves for different audiences will tell marketers a lot about where certain audiences are actually going.
The right position right now
There is a version of this that works well for brands, and it manifests as clear visions, realistically measurable expectations, and a genuine willingness to learn from what the data shows. It doesn’t look like redirecting search budgets in hopes of recapturing lost volume, or committing to a platform before understanding who’s really on it.
The opportunity is real. The timeline for becoming a proven, scalable channel is not as short as the current momentum might suggest. Going in with your expectations, rather than what’s being sold, is probably the best bet.
