The time for research is over. People who used to find your product through a search box are now finding it through chat, and ChatGPT is where most of those conversations are happening. A few months ago, almost all client conversations began to end in the same way: “should we be investing in ChatGPT ads?” My answer has always been the same: the format is interesting, but let’s wait until we know how to talk to users this way. We started exploring so we could have something tangible to take back to them – and here’s what we found.
The question about advertising on ChatGPT is no longer whether products should exist. That’s what the format does once you use it. We’ve been testing it with customers, and the results are clear enough to share.
The prize is big enough to justify the test. ChatGPT sees more than 800 million active users per month, nearly one-tenth of the world’s, and processes nearly two billion messages daily. About 70 percent of all AI assistant traffic originates from ChatGPT. Of those who use it, 63 percent say they use AI for research and discovery, and 53 percent use it to compare products and prices. Considerations move up the funnel, in a place where brands didn’t have a way to appear until recently.
The unit is simple, targeting is not easy
The ad appears below the ChatGPT answer in the box labeled “Sponsored”: one 1×1 image, title, description, and click to the product site. The targeting is duller than most buyers expect. It is only contextual, locked to the user’s knowledge rather than what the model implies. There is no audience targeting, no crowdfunding, and very little country control. Keeping up with the information is the only thing that determines whether it helps.
Niche, quickly aligned art wins on price
The price is where testing has changed the way we think about the channel. OpenAI uses an implicit relational model and a proto-auction, where in practice it creates two values. Apply immediately with no competition and you pay around $15 CPM. Compete quickly and win, and pay up to $60. For all brands, CPMs averaged between $25 and $35 depending on the message, and our test reached $40.
The lesson is that the channel rewards accuracy. The more accurate your creative maps are in information, the better your cost effectiveness. CPC is not worth it. The methodology is too vague to trust, and we see a better cost per click by bidding on CPM and letting the clicks come.
Measurement is a real limitation
Today you get impressions and clicks, and little else. There is no direct conversion optimization yet, so you can’t tell the platform to chase results. What you can do is measure click results and answer a key question: do clicks from ChatGPT show higher intent than native, display, video, social, or search, judged by what people do once they arrive on the site? Pixel and conversion data can be shared with partners for assistance.
Early quality reading is encouraging. CTRs come in comparison to native ads, a strong signal for such a format. The first thing you need to do is create an evaluation framework. Decide how you will compare the intent of ChatGPT against site-based interactions from your other channels before spending a dollar.
Three ways to get in, and how much each one costs
There are three routes. Criteo is commerce-focused and can handle product feeds, making it ideal for CPG and retail clients because you can run thousands of ads at once. It recommends $50,000 to $100,000 per month. StackAdapt recommends around $50,000 per month and adds some incentives and geo-targeting options. You can also go directly to OpenAI, although that carries a minimum of around $250,000 per month.
Which route is best depends on what you are trying to drive. There is little data so far on which format is best suited, but it reaches people at a time of high consideration. So the questions you should ask yourself are how you want to lead those people to your site and how you are going to measure the knock-on effect. Your answers should set the budget, not the other way around.
Bottom line
The format is new, the targeting is unclear, and the rating is small. None of this has stopped performance in testing, and considerations still enter the debate as to whether or not the products are ready for you. The question now is not whether it should be checked but how much money should be set after the inspection while the laws are being written.
The following can be predicted if you watch the development of the program. Audience orientation will come; OpenAI has valuable data about its users and leaving it unused does not make business sense. Conversion development will follow. CPMs will increase as more brands appear. The auction will be very transparent. And at some point, the format will look a lot like the channels you already use.
Product testing will now begin when the format matures. That gap closes quickly when CPMs rise and competition arrives.
