Criteo-OpenAI Partnership: ChatGPT Drives 1.6x CPG Product-Page Visits vs. Search

Full analysis

Criteo just told the market that ChatGPT is now a real traffic source for consumer packaged goods brands. The claim: from January through July 2026, ChatGPT drove 1.6 times the share of product-page visits that search engines did on average, and in July 86% of Criteo's CPG clients booked at least one sale referred by ChatGPT. It is marketing collateral from a Criteo-OpenAI partnership, so read it as a pitch, not a print.

What's actually being decided here: not "is Criteo's number right," but whether ad-tech operators should start building for AI chat as a commerce referral channel now, before the measurement is settled. That's an easy-to-undo call. Updating your attribution to tag ChatGPT traffic costs little and reverses cleanly if the volume turns out to be noise. No hard deadline, but the CPG planning cycle for 2027 is the clock that matters.

The Market Analyst. Criteo's stock has carried a cloud for years because its core business leaned on third-party cookies, the tracking that lets advertisers follow users across sites, and that tracking is dying. So any story that says "Criteo is an AI commerce company now" is a cheap way to re-rate the shares without doing the hard work. This data does not do that work. It is first-party, unaudited, and the baseline for the 1.6x is undisclosed. The test is whether a Procter & Gamble or a Unilever says on its own earnings call that AI chat drove measurable sales lift. Until a buyer corroborates, this is Criteo grading its own homework. In plain terms: a company talking up its own numbers is not evidence until a customer repeats them.

The Skeptic. Pull the two numbers apart and they shrink. "1.6x the share of visits versus search, on average" compares a raw referral count against some normalized average, and Criteo did not say how it built the baseline. "At least one transaction" for 86% of clients is the lowest bar there is. One sale across a client's entire book clears it. That tells you nothing about volume, repeat purchase, or whether these were sales that would have happened anyway through search. And the 40% comfort with AI reordering personal care products is a survey answer, not something anyone did. For this to matter, ChatGPT referrals have to scale without stealing conversions that search was already closing, and the attribution has to survive clean measurement. Neither is shown here.

The Operator. Tuesday morning, the real problem is plumbing. Most analytics stacks dump ChatGPT traffic into "direct" or "other" because the referral string does not resolve cleanly. So the 86% figure is almost certainly undercounted inside your own reports, which cuts both ways: the channel may be bigger than your dashboard shows, and you cannot prove it to a client who asks. Fix the tagging now. Update your UTM parameters, the little tracking codes stapled to a link, and your last-touch rules so chat referrals get their own bucket. The second-order effect at 90 days: CPG procurement teams start asking for AI-referred intent in the renewal conversation, and the vendor who can show it in clean reporting wins the contract from the one who shrugs.

The Customer / End User (the CPG brand). A brand manager at a personal-care company reads this and feels two things. First, mild relief: here is a growth channel that is not Amazon's walled garden or Google's rising prices. Second, suspicion, because the brand has lived through every ad-tech vendor overstating a new channel. What the brand actually wants is incrementality, meaning sales it would not have gotten otherwise, not a reshuffle of credit from search to chat. Nobody has shown that yet. The 40% who say they would let AI reorder their shampoo is interesting, but saying it in a survey and handing a model your card are different acts, and the gap between them is where most "AI commerce" forecasts go to die.

The CFO. The cost of acting on this is low, which is the whole point. Re-tagging referral traffic is a few sprints of analytics work, not a platform rebuild. The expensive mistake would be the opposite: building a dedicated AI-commerce product line, pricing it, and staffing it against a channel that turns out to be 2% of referrals that nobody audited and everybody overstated. Spend on measurement, not on a new business unit. The payback on better attribution is immediate because it protects renewals. The payback on an AI-commerce bet is unknown because nobody has sized the channel with audited numbers.

Where the council splits. The Strategist view in the room says Criteo is smart to plant a flag early, because the real prize is owning the product-catalog feed that AI models pull from to make recommendations, and Criteo already runs large product feeds for retargeting. The Market Analyst and the Skeptic say that arc is seductive and unproven, and the two-year risk is that Amazon and Google build their own AI commerce loops and cut the middleman out entirely. The second disagreement is simpler: the Operator wants everyone building now because the plumbing takes a quarter, while the CFO and Skeptic want proof of incrementality before a dollar moves beyond measurement. Both can be right. Fix the tagging cheaply, withhold the big bet.

What this hinges on. Two facts, neither established. Is ChatGPT referral traffic incremental or just recredited from search? And does the volume scale past a rounding error? The council leans skeptical on the specific numbers and constructive on the cheap, reversible move: update attribution now, treat the product bet as unfunded until a buyer corroborates. What to verify before committing real money: pull your own ChatGPT referral share from clean logs, and watch whether any major CPG advertiser names AI-chat sales lift in its own reporting.

Prediction: No large CPG advertiser (Procter & Gamble, Unilever, Colgate-Palmolive, or Kimberly-Clark) will cite ChatGPT or AI-chat referral as a material sales driver on a quarterly earnings call before the Q4 2026 earnings season ends in February 2027.

Confidence: Medium — the channel is real but too small to name, and the number nobody can audit stays convenient for Criteo.

Why: Criteo's own framing gives it away: the strongest stat it could publish was "at least one transaction" for 86% of clients, which is the floor, not a volume claim, and the 1.6x has an undisclosed baseline. A channel that drove material revenue would be measured in dollars, not in the lowest-possible threshold. Big CPG advertisers put specific channels on earnings calls only when the dollars move the quarter, and nothing here suggests AI chat clears that bar yet. The opposite outcome, a P&G or Unilever CFO crediting AI-chat commerce with real lift, would require the volume to be large and the attribution clean, and the vendor pushing the story would not be hiding behind "at least one transaction" if either were true.

Revisit by 2027-02-28: We're right if no top-four CPG advertiser names ChatGPT or AI-chat referral as a material revenue or sales driver on a Q3 or Q4 2026 earnings call. We're wrong if any of Procter & Gamble, Unilever, Colgate-Palmolive, or Kimberly-Clark does so.

The useful move is still the cheap one. Tag the traffic, size it from your own logs, and let the brands' own numbers decide whether this is a channel or a press release.