Criteo Discloses ChatGPT Ads Performance: 80% New Customers and 1.5x to 2x Conversions Are Moving Ad Budgets
Criteo's CEO put numbers on ChatGPT Ads for the first time on September 8: 80% of paid referral traffic is new to brand, and lower-funnel conversions run 1.5 to 2 times better than search. What Japanese advertisers can buy today, which budgets move, and where measurement still falls short.
Key Takeaways
- On September 8, Criteo CEO Michael Komasinski put operating numbers on ChatGPT Ads at Citi's conference. In mid-funnel use cases, 80% of paid referral traffic is new to brand, and lower-funnel conversions run 1.5 to 2 times better than traditional search
- If those two figures hold, budget moves out of connected TV, social, and product listing ads. Komasinski named the sources himself
- Advertisers in Japan have been able to buy ChatGPT Ads inventory since June. At the same time measurement is immature, and retail's share of total ChatGPT ad spend has actually fallen. The numbers need discounting
The two-month feel of running the channel
AI advertising coverage has so far been either market-size forecasts or partnership announcements. What came out of Citi's TMT conference in New York on September 8 was neither. It was a description of how advertisers who are actually buying the inventory behave.
Criteo used Citi's 2026 Global TMT Conference to lay out a broad strategy built around commerce data, retail media and automated ad buying.
www.investing.comThe speaker was Criteo CEO Michael Komasinski, 18 months into the role. Criteo became the first advertising technology partner integrating with OpenAI's advertising pilot in March 2026, and has provided a buying path into ChatGPT Ads since.
The operating detail he offered was a duration: two months. An advertiser runs a campaign for a couple of weeks, evaluates, tweaks, runs it again, and takes roughly two months to reach a comfort level on how the channel performs against other tactics. That is why the first real milestone lands in the October to November holiday season, and the second in the January budget-setting cycle. Whether brands put ChatGPT into their always-on plans is the first signal the whole ecosystem gets.
What 80% and 1.5x to 2x actually say
The figure Komasinski described as one Criteo keeps sharing because it keeps showing up in the data is the new-customer rate.
For mid-funnel objectives, the stage between awareness and consideration, 80% of paid referral traffic through ChatGPT was new to brand, meaning first-time visitors. Criteo's June update published the same claim as "more than 80% of ad-driven ChatGPT Ads traffic comes from new customers," and the August European expansion release repeated the level.
The second figure concerns the lower funnel. In conversion-oriented use, ChatGPT Ads traffic converts 1.5 to 2 times better than traditional search. Back in March, Criteo phrased it more conservatively: LLM-referred users converted at roughly one and a half times the rate of other referral channels, based on a sample of 500 US retailers in February 2026. Six months later the top of the range has moved to double.
Why this happens becomes clearer once you look at where the ad appears. Someone asking a conversational AI about a product is past "what should I buy" but short of "which one." A person searching a brand name already knows the brand; a person asking for sunscreen for sensitive skin in their thirties has no shortlist yet. That produces both a high new-customer rate and, because the conversation carries genuine purchase intent, a high conversion rate.
Where the budget comes from
This is the part with the most practical bite. The moderator asked directly where the money would come from, and Komasinski named the sources.
| Funnel stage | What ChatGPT Ads delivers | Existing channels that lose budget |
|---|---|---|
| Mid-funnel (new customer acquisition) | 80% of paid referral traffic is new to brand | Connected TV, social advertising |
| Lower funnel (conversion) | Conversions 1.5x to 2x better than traditional search | Product listing ads (PLAs), branded search depending on CPCs |
| Possible net addition | Genuine incrementality would lift total ad spend | Not a transfer from existing channels but spend on top |
New customer acquisition has historically been the job of connected TV and social. If ChatGPT posts better numbers against the same objective, budget moves. In the lower funnel, product listing ads take the first hit, and depending on CPC levels some branded search spend could follow. He was careful to limit the claim: not all of traditional search, but certain tactics within it.
Komasinski also allowed for the opposite reading. If the channel is genuinely driving incremental outcomes, he said, there could be some expansion in overall ad spend as a percentage of GDP. Whether this is a reallocation or a net addition will not be settled until the holiday results are in.
Advertisers in Japan can already buy
One premise matters for Japanese e-commerce teams. After the United States, Australia, Canada, and New Zealand, advertisers in the United Kingdom, Japan, and South Korea have been able to access ChatGPT Ads inventory through Criteo since June 2026. The August expansion added 31 European markets, bringing Criteo's coverage to 51 markets. Japan is not a market that is coming; it is already open.
The buying entry point has widened too. The same June update made ChatGPT Ads inventory available through Criteo GO, Criteo's self-service cross-channel platform, which puts it within reach of small and mid-sized businesses. By vertical, apparel, home furnishings, consumer electronics, automotive, and beauty were named as the leading categories running ads in ChatGPT through Criteo.
What Criteo sells here is not the inventory itself. In his own framing, the job is optimizing the inputs and explaining the outputs.
What SKUs am I going to advertise? Is the data associated with those SKUs up to date? Is the metadata correct? Is it deep enough and rich enough to drive the relevancy signals?Source: Michael Komasinski, CEO, Criteo
The input side is SKU selection, metadata enrichment, and prompt-hint engineering. Criteo says it regularly ingests 25 billion raw SKUs and normalizes them into a usable catalog of about 5 billion, the work of recognizing that two slightly different listings describe the same vacuum cleaner. The output side is cross-channel measurement: how ChatGPT-referred traffic performed relative to other paid referral traffic.
Reasons to discount the numbers
Everything above is Criteo's account. The other side deserves airtime.
Criteo itself is cautious. Komasinski held back on guidance, explaining that while the client count is decent, budgets per client are relatively small. He expects a "meaningful" contribution in 2027, but the specific form that guidance will take, which metric and what level, has not been disclosed. Even the published figure of more than 2,000 advertisers dates from June; asked for an update, he said Criteo hopes to refresh it jointly with OpenAI sometime in the fall.
The sharper counterevidence comes from third parties. Sensor Tower research published on September 1 and reported by PPC Land found that shopping and retail's share of US ChatGPT ad spend fell from 37% in April 2026 to 21% in August, as financial services climbed from 2% to 13% and software from 13% to 16%. Retail is still the single largest category, but the ChatGPT ad market itself is shifting its center of gravity away from commerce. Read the channel purely as the front door to AI commerce and you will misjudge the competition for inventory.
Measurement gaps are the other complaint. In AdExchanger's September 3 reporting, buyers cited the absence of competitor data, limited conversion tracking and audience options, and no way to manage multiple advertisers in one place, with one practitioner calling the platform rudimentary. OpenAI's $1 billion run rate also falls short of the company's own earlier projection of $2.5 billion in ad revenue for 2026. On top of that, AI referral traffic remains a low single-digit percentage of most sites' total visits in Similarweb's 2026 data.
The honest summary of the current position: a small volume of high-value traffic. Ratios like 80% and 2x are also the kind that run large when the denominator is small.
What e-commerce teams should do first
The sequence follows fairly directly from the evidence above.
Product data comes first. Whether you can compete in ChatGPT Ads depends on whether your SKU metadata matches the context of the conversation. Criteo normalizes 25 billion records into 5 billion precisely because that is the precondition for relevancy. Before adding budget, tighten the product-information granularity covered in our piece on LLMO for e-commerce sites.
Second, budget two months for the test. A week produces no meaningful difference. If you want a read before the holiday season, working backwards means you should already be running.
Third, close the measurement gap on your own side. With platform-side tracking as thin as it is, you need traffic-source separation and an evaluation frame beyond last click, or comparison is impossible. As noted when ChatGPT Ads expanded to 31 European markets, paid placement and organic visibility compete inside the same conversation. Optimizing only one of them does not raise your odds of appearing in the answer.
Closing thoughts
What was disclosed here moves AI answer-surface advertising from "it seems to work" to "here is how it works." Whether 80% new to brand and 1.5x to 2x conversions are enough to move test budgets into always-on plans gets answered in October and November.
Two things to watch. The advertiser count Criteo says it will refresh this fall, and the retail share Sensor Tower is tracking. If the first rises while the second stops falling, AI answers really are becoming a commerce entry point. If only one moves, the wait-and-see period continues.
