French programmatic agency Gamned! moved fast. On Aug. 25, the day after OpenAI opened ChatGPT Ads to France and 30 other European markets, the agency launched a first test campaign aimed at French users. It ran the campaign through StackAdapt, the Canadian DSP that ranks with Criteo among ChatGPT Ads' main tech partners.
Gamned! wasn't trying to decide in one day whether ChatGPT Ads would become a new performance channel. The channel isn't cheap either, with CPMs quoted at €15 to €50. The agency mostly wanted to see how the whole thing actually works: campaign setup, targeting, creative delivery, optimization and conversion reporting. None of that is easy to judge from the platforms' sales materials alone.
The ads ran as Chat Cards. The format sits directly inside ChatGPT's conversational interface and pairs a visual, a brand message and a call to action. The ad shows up as a continuation of the conversation. It stays clearly separate from the answer ChatGPT generates.
Conversational targeting goes beyond keywords
Gamned! built more than 50 groups of keywords and phrases to steer delivery. They cover queries about choosing a trading desk, programmatic agencies, CTV and B2B audiences. They also cover comparison searches and more technical topics (DV360, The Trade Desk, Amazon DSP, retail media, data strategy and so on).
But this isn't exact-match keyword targeting, the way a search engine works. The phrases describe contexts that the platform interprets. OpenAI calls them "context hints." Remember the term, because you'll be seeing a lot more of it.
The same intent can be phrased many ways, and it can trigger ads beyond the list the advertiser supplied. That room for interpretation is both the product's appeal and its main gray area. It can catch commercial intent buried in a long, complex conversation. It can also make targeting less precise.
Without detailed reporting on the conversations that triggered the ads, it's hard to know how closely the contexts bought match what users actually asked.
Gamned! uses its own in-house tools, built on several large language models, to rephrase contexts, check how they're understood and spot variants likely to trigger delivery.
The agency also tries to avoid paying for queries where ChatGPT would already mention the brand on its own. Over time, targeting could become less about picking the right keywords and more about trading off organic visibility against sponsored placement.
Three buying strategies, one day
Gamned! ran three strategies at the same time:
standard CPC;
CPM;
conversion-optimized CPC, or C/oCPC.
With the last one, the algorithm gets more leeway to favor the impressions most likely to drive the action the advertiser wants.
After the first day, conversion-optimized CPC posted a 2.70% click-through rate, compared with 1.79% for standard CPC and 1.18% for CPM buying. Gamned! says that strategy also drove the most conversions.
The algorithm appears to have found contexts more likely to generate clicks during the test. That doesn't make the strategy a lasting winner. Several things could skew the comparison: the volume available to each buying mode, price gaps, how audiences were split and how long the algorithm needs to learn.
Gamned! hasn't disclosed impressions, budgets, or average cost per click or per conversion. A high click-through rate on low volume doesn't carry the weight of a result that holds for several weeks across different advertisers.
It also isn't clear what counts as a conversion. A site visit, a page view and a form fill say very different things about the value generated. Click-through rate, meanwhile, measures how well the format prompts an interaction. It says nothing about incremental sales or leads.
StackAdapt's edge will be cross-channel
OpenAI is expected to launch its own Ads Manager at some point, so advertisers can buy ChatGPT Ads self-serve. "That option will naturally suit campaigns dedicated solely to ChatGPT," said Baptiste Pechery, Gamned!'s head of programmatic.
StackAdapt will still matter for campaigns that combine ChatGPT Ads with display, video or other programmatic inventory. "Using the same pixel makes it easier to read conversions across channels. It could also make it possible to retarget users reached in ChatGPT on the open web, or instead to limit some of the overlap," Pechery said.
That capability still has to be tested at larger scale. But it's where StackAdapt can set itself apart from a campaign run entirely inside OpenAI's closed environment. "It reminds us a bit of the days when Meta wasn't a walled garden and you could buy it through AppNexus," Pechery said.
At the time of the interview, Gamned! was already preparing several campaigns for advertisers, with test budgets of up to a few tens of thousands of euros. Those campaigns should show how much the channel can deliver when several buyers go after the same contexts at once.
A playbook, not a benchmark
"This channel is shaping up to be a must. It felt natural for us to be among the first to test it in real conditions. That way we clear the operational hurdles ourselves before our clients hit them, and we can offer them launches they control rather than launches they're forced into," Pechery said.
So the first test yields more of a playbook than a performance benchmark. Figuring out where ChatGPT Ads really fits in a media plan will take longer campaigns, more advertisers and comparisons with other channels chasing the same goals.
Four things are still unknown: how much the channel can deliver, what it really costs, how good the contexts bought are and whether the conversions are incremental. The real test isn't a one-off spike in click-through rate. It's whether ChatGPT Ads drives extra actions at a competitive cost, rather than just picking up users who would have found the brand in the organic answer anyway.
