By Marjorie Bicaïs, Head of SEA/Média CyberCité
In the small digital world, this has been THE subject for several weeks: whether you are an advertiser, agency or even a simple user of the famous LLM, you could not miss this novelty: ChatGPT has opened its interface to advertising. Agencies like CyberCity have begun to test the management, the first feedback is widely published on social networks and the question from many advertisers is now “when are we going there?”.
To understand why this shift is happening now, we must first look at what made it possible: how the use of LLMs has changed in nature over recent months, the exponential growth of users and above all the purchasing intention which is now expressed in full conversation. Decryption of this mutation.
The shift in use that changes everything
First of all, let us point out that search engines remain the daily entry point for the vast majority of French people when they do a search. According to an Ipsos study published in May 2026, 85% of us use it every day, and Google alone brings together around 46 million unique daily visitors in France according to Médiamétrie. He therefore remains the undisputed leader.
However, we all see it around us, a second reflex has taken hold alongside it. At the start of 2026, 59% of French people were using generative AI and this use is becoming more widespread: according to Ipsos, 1 in 2 French people say they use a generative AI every week, compared to 29% a year earlier. This is certainly not a replacement for Google because, as we saw above, its use remains largely in the majority, but it illustrates a shift from part of research to conversation.
In detail, 20% of French people already use AI to get information and 18% to compare offers, i.e. two to three times more than a year earlier. This shift thus affects the phase where the decision is formed, and, a real consequence on the business level, 55% of AI users say they have made a purchase after asking an AI for advice.
The purchasing intention is now also expressed in full conversation, and this is precisely what OpenAI has chosen to open up to advertisers.
The mechanics of the launch
Announced in January, the OpenAI advertising pilot started on February 9, 2026 in the United States. He then won Canada, Australia and New Zealand in the spring, then the United Kingdom, Japan, South Korea, Brazil and Mexico. On August 24, it’s the turn of France and 30 other European markets, all open on the same day.
The tool matured at the same pace: the American pilot started at around $60 CPM, with a minimum commitment of around $200,000. In May, a self-service manager opened to all US advertisers with CPC bidding to which OpenAI has since added conversion optimization. Many marketers have already run ads, and with around 40 markets open, this is no longer a local experiment but a large-scale rollout.
The perimeter, however, remains very limited. Only major users of the Free and Go offers see ads, with OpenAI explaining that advertising finances free access and ensures that it does not influence the model’s responses and that conversations remain private from advertisers. These commitments are especially important since 26% of users in the Havas study would doubt the neutrality of the responses if ads appeared there.
It remains to be seen how many French people are really affected. ChatGPT brings together around 7.8 million unique visitors every day in France according to Médiamétrie. OpenAI does not publish a country breakdown, but globally only about 5.5% of weekly users are subscribers (50 million out of 900 million in February 2026). If France follows this average, most of the audience is therefore exposed to ads.
One point deserves the attention of advertisers: professionals equipped by their employer with Business or Enterprise offers are out of scope and consequently, the audience exposed is more that of personal use than that of professional use. This is good news for consumer brands who find their targets there, but for B2B it is necessary to think about its presence differently, which can be complex.
What the platform actually allows
On the format side, OpenAI has chosen discretion: a sponsored card is displayed under the ChatGPT response clearly identified as such and separated from the text generated by the model, so that the user always distinguishes what the assistant says from what the advertiser offers.
Temporary conversations do not display them and sensitive contexts are excluded: no ads appear near topics related to personal health, mental health or politics, and political advertising is not accepted.
On the targeting side, another way of thinking about distribution is established: there are no keywords or demographic criteria, and it is the subject of the current conversation that triggers the ad. The advertiser describes the situations in which its offer is relevant, using “context indications” which guide the distribution. OpenAI also added geotargeting, custom audiences from 1st party data and conversion measurement using the OpenAI Pixel and the Conversions API.
Targeting can be done by medium, web or application, provided that the campaigns are separated, which we recommend.
Each campaign is built around one of the three classic objectives of digital marketing, coverage to work on notoriety, clicks to generate traffic or conversion, with a minimum budget of 15 euros per day and per campaign. E-commerce advertisers can also import their product flowbut the operation remains operationally complex and OpenAI support is still inexperienced, which reminds us that we are dealing with a young platform, where the experience acquired by the agencies takes on its full value.
Our first test campaigns show a CPM around 11 euros and a CPC comparable to that of Google Ads but the CPA is still lacking perspective. The most unclear point remains the measurable impact of the campaigns because users rarely accept the cookies banner, the quantity of data going back to the Analytics tools is low.
A game changer?
This launch is not an isolated phenomenon: Google also brought France into the era of AI search this summer with the arrival ofAI Overview and AI Mode and these responses are already displayed on a significant proportion of informational requests. Meanwhile, the advertising space at the heart of the response remains closed to French advertisers at Google to this day, while OpenAI is already opening it to them. Alongside Google Ads and Meta Ads, a new advertising field is opening up: that of conversation.
This development does not only concern advertising, because it also affects natural referencing: when a generated response is displayed at the top of the page, the Internet user clicks less on the classic results. In the United States, the presence of an AI preview reduces the click-through rate by almost half. Conversely, the brands cited in these responses perform better, both in natural clicks and in paid clicks. A brand can thus be cited in the response, or sponsored just below, and the subject is not to choose between referencing and advertising but to take care of its overall presence.
Game changer, then? Not for the sole reason that the ad is triggered on an intention rather than on a keyword, since Google Ads is also going in this direction with AI Max. The real break lies in the location: the ad appears within a conversation where the user has detailed his needs and the constraints that matter to him. In terms of its volumes, on the other hand, this is not yet the case, since the audience remains lower than that of Google and limited to free offers.
So, do you need to be present? It depends on each business model and sector of activity but given the modest entry ticket and the fact that the auctions remain little contested, the learning phase costs less than it will cost in a few months. It is in this spirit that CyberCity is now supporting its advertisers on ChatGPT Ads: as with each of the levers we manage, we rely on an A/B testing and consulting approach to maximize the performance of each campaign.
Let’s return to the image of the earthquake, which is better read as a movement of the ground than an upheaval. Users’ attention shifts towards the conversation, and part of the budgets will shift with it. For brands and agencies, it is not a question of disrupting everything but of testing methodically and learning quickly, while the ground is still little contested.
