Criteo Onboards 2,000 Advertisers to OpenAI Ad Partnership

Full analysis

Criteo says it wired 2,000 advertisers into OpenAI's platform in four months, and that 80% of the traffic OpenAI sends to its retail partners is net-new to those brands. Operators don't need to decide whether Criteo's stock deserves a bounce. They need to decide whether conversational AI is a genuinely new demand channel that publishers, retail media networks, and DSPs need a plan for, or whether this is an early-adopter mirage that regresses the moment the ad surface matures.

This is a Type 2 decision for most operators. Nobody has to bet the roadmap on it today. You have to decide how much attention to spend, and whether to start building feeds and attribution for a surface that barely exists. Forcing function: Cannes noise fades, and by year-end there will either be revenue numbers or there won't.

The Skeptic. Four months, 2,000 advertisers, one Todd Parsons quote at Cannes. That's the whole evidence base. The 80% net-new number is measured on a tiny self-selected pool of early adopters, on a surface where novelty is the entire behavior. People poke at a new toy, so of course they land somewhere new. That regresses hard once OpenAI builds real retargeting pools and the curious-tourist traffic normalizes. "Co-developing how advertising works" is what you say when you have API access and a press slot. For an informed outsider: this is a partnership announcement, not a revenue report, and the exciting stat is the least trustworthy part of it.

The Market Analyst. Wall Street still prices Criteo as a retargeting business in slow decline, terrified of cookie deprecation. This channel is cookie-independent by design, so it sidesteps the exact fear the shorts are anchored to. That's the interesting mismatch. If OpenAI monetization is real even at modest scale, there's a re-rating case that consensus isn't carrying. But note what actually moved: Criteo got the OpenAI logo first, which matters more than the 80% stat. In plain terms, being the incumbent's chosen partner on a new surface is worth more than any single early metric, because it's hard for a rival to dislodge once merchant feeds are plumbed in. The risk is OpenAI eating the toll booth later.

The Operator. Two thousand advertisers in four months is real pipeline velocity, and the demand team now has a live proof point to accelerate enterprise sales. Then Tuesday morning arrives. OpenAI's conversational surface doesn't map onto product-feed and retargeting schemas. Bid logic breaks first: CPCs tuned for high-intent search will misfire on a surface where intent is still forming. Attribution breaks second, and worse. Last-click models will systematically undercount a discovery channel, so the finance team sees spend with no credited conversions and kills the budget before the channel gets a fair shot. For a non-specialist: the plumbing built for "people who already want to buy" doesn't fit "people just starting to look."

The Customer / End User. Put yourself at a retail media network or a mid-size retailer. You didn't ask for an OpenAI channel. You asked for incremental sales you can measure. The 80% net-new claim is genuinely attractive if true, because incremental reach is the one thing retargeting can't sell you. But you've been burned by "new channel, trust the pixel" before. What you actually need is a clean read on whether these net-new visitors convert at a rate that justifies the CPC, and nobody can give you that yet. So you run a small test, cap the spend, and wait. That caution is the real governor on how fast this scales, not Criteo's sales team.

The Strategist. The durable asset here is Criteo's tens of thousands of retail merchant integrations, not any algorithm. OpenAI can spin up an ad product; it can't quickly replicate structured product feeds and merchant relationships. Whoever owns that layer when conversational commerce matters owns a toll road, even if conversational search only ever reaches a fraction of Google's scale. The two-to-three year threat is obvious and serious: OpenAI builds its own commerce graph and goes direct to merchants, or Amazon and Google close equivalent loops faster inside surfaces they already own. Partners are useful until the platform learns the route itself.

Where the council splits

Three real disagreements.

First, the 80% number. The Skeptic says it's novelty on a rigged base and will collapse. The Market Analyst and Strategist say even a regressed version of it is a new demand channel that doesn't exist elsewhere. Both can't be right, and the tell will be whether net-new rates hold once the early-adopter pool washes out.

Second, who has the leverage. The Strategist says Criteo's merchant integrations are the moat OpenAI can't rebuild fast. The Market Analyst's own risk case says OpenAI disintermediates exactly that layer. The whole partnership lives inside that tension: Criteo is teaching a platform the route to its own merchants.

Third, does this scale on its own or does attribution strangle it. The Operator says last-click undercounting kills budgets before the channel proves out. The Customer confirms it from the buy side: capped tests, slow trust. Velocity of sign-ups says nothing about velocity of spend.

What it hinges on

Two beliefs, and everything else is noise. One: does conversational AI actually deliver incremental buyers at a rate that survives contact with a mature ad surface, or is 80% a launch-window artifact. Two: does the platform need the partner, or just the partner's data and merchant list, over a two-year horizon.

The council leans skeptical on the number and wary on the leverage, but respects the positioning. Being OpenAI's chosen commerce partner is worth having even if the current metric is soft, because that seat is hard to win back once feeds are plumbed. What to de-risk before you spend real money: demand an attribution model built for discovery, not last-click, and run a capped incrementality test with a real holdout group before you believe anyone's net-new stat, including your own.

For operators who aren't Criteo: this is the signal that conversational AI is becoming a demand channel with a real toll booth, and the retail media networks and DSPs without a merchant-feed strategy for it are the ones who'll be renting access later.

Prediction: Criteo will not disclose specific, audited OpenAI-attributed revenue or GMV on any earnings call through its Q2 2027 report, and the "80% net-new" framing will keep standing in for a revenue number it hasn't produced.

Confidence: Medium. The incentive to keep selling the metric over the money is strong, but OpenAI's timeline could force a number.

Why: The only figure Criteo has put forward is a net-new traffic percentage measured on a tiny early-adopter base, disclosed at a conference by a product exec rather than a CFO on a results call. That's the pattern of a company with a narrative it wants believed and a revenue line too small or too messy to show yet, because discovery-mode traffic breaks the last-click attribution Criteo's own systems run on, so crediting real dollars to the channel is genuinely hard as well as unflattering. The traffic stat is safe to repeat and impossible to check; a revenue number invites the "is that all?" question. The opposite outcome, a clean audited OpenAI revenue disclosure, only happens if the channel scales fast enough that the number flatters the story, and four months of self-selected early adopters rarely gets there by mid-2027.

Revisit by 2027-08-15: We're right if, through Criteo's Q2 2027 earnings, management keeps citing engagement or net-new traffic metrics for the OpenAI partnership without a specific attributed revenue or GMV figure. We're wrong if Criteo reports a concrete OpenAI-attributed revenue or GMV number on any earnings call before then.