On August 18, OpenAI confirmed that it would open its advertising platform to 31 European markets, including France, starting August 24. The ads will take the form of labeled inserts, displayed below the AI’s response and visually separated from it, without altering the response’s content. It is, therefore, a traditional media buy in principle: a placement, a budget, and measurement.
The announcement itself boils down to roughly that. What we find more revealing is the approach they’ve taken: two design options, both well-documented, are enough to indicate that this medium won’t be planned like retail search.
Without personalization, the topic of the conversation is the only cue
At launch in Europe, targeting will initially be context-based. It will rely on the topic of the conversation and limited contextual information; any personalization will require explicit consent, and advertisers will not have access to conversation histories.
This changes the basic unit of a media plan. In France, more than three-quarters of the retail media budget is currently spent on search—that is, on keywords linked to product listings: €599 million out of €775 million in the first half of 2026, according to the 36th E-Advertising Observatory.
However, this feature does not exist in ChatGPT Ads. There are no keywords to purchase—only the topic of a conversation. This means you have to think in terms of intent rather than search terms.
Let’s see how this keyword performs in the food industry. On a retailer’s website, a customer types in “whole-grain pasta” or “low-fat milk”: a product search query for which a keyword is purchased. That same customer, in a conversation with a chatbot, rarely types that out. Instead, they ask what to make for dinner with whatever’s left in the fridge, how to feed four people for fifteen euros, or what to use as a substitute for an ingredient they can’t tolerate.
This is a viewpoint we have long advocated: grocery shopping starts with a meal or a craving, not a shopping list. Contextual targeting provides access only to this spontaneous expression of need. The planner’s question therefore becomes “What situations does my product address?” rather than “Which keywords should I bid on?”
Access is provided through six groups of agencies
Option 2: At the European launch, purchases will not be made via self-service. Purchases will be made through Publicis, Omnicom, WPP, Havas, Dentsu, and MediaPlus; OpenAI has announced a self-service platform for smaller organizations, though no date has been set.
So the question this fall isn’t “what budget,” but “by whom.” And above all: an ad network that isn’t yet self-service is one that’s still being put together. Formats, measurement methods, and the scope of the inventory will change in the coming months, as is typical for any advertising platform in its early stages. This makes a case for treating this fall as a period of observation—working with amounts we’re willing to lose—rather than as a channel to scale up immediately.
It’s worth noting that the ads appear only on free and Go accounts: Plus and Pro subscribers are not affected, as is typical in any freemium model. It’s simply a coverage limit to factor into the calculation—nothing more.
And what about organic products in all of this?
The ad appears below the response and does not alter it: purchasing it has no effect on what the model spontaneously generates. Whether the ad appears within the response itself—known as GEO, or generative engine optimization—remains an open question, one that depends entirely on something else: the quality, structure, and reliability of your product data and content.
The parallel with the traditional web holds up well: the arrival of Google Ads never exempted anyone from doing SEO. The two have coexisted for twenty years, with different teams, budgets, and time horizons—one stops the day the budget is cut, while the other builds up slowly and continues to produce results. There’s no reason for that logic to change here: the launch of a paid advertising inventory in ChatGPT simply means both efforts are now happening simultaneously, whereas many hadn’t yet started either one.
August 24 won’t change anything overnight. Above all, it opens a window of opportunity to learn about a medium that is still malleable, before its conventions become set in stone. If you’d like to apply these insights to your own context, we’d be happy to discuss them with you.
