ChatGPT Ads: what early testers are reporting, and which countries are next

Advertising in ChatGPT is no longer a hypothesis. OpenAI launched the pilot in the United States on February 9, 2026, with clearly labeled "Sponsored" results shown below the answer, not inside it, and only for logged-in users on the Free and Go plans. The higher paid tiers (Plus, Pro, Business, Enterprise and Edu) remain ad-free.

The expansion was aggressive. On March 26, according to Reuters, the pilot crossed 100 million dollars in annualized revenue in just six weeks, and OpenAI simultaneously confirmed pilots in Canada, Australia and New Zealand. On June 6, ads went live in the United Kingdom, the first market outside North America and Oceania. The May 7 announcement made the next markets explicit: Japan, South Korea, Brazil and Mexico "in the coming weeks", with a promise of "many more markets" by the end of 2026.

Behind the rollout, OpenAI built the plumbing of a serious performance channel. On May 5 it launched the self-serve Ads Manager with pixel-based measurement, an identifier called oppref stored in a first-party cookie, plus a server-side Conversions API to send events from the advertiser's own infrastructure. On June 5 it began rolling out conversion-optimized campaigns. Announced technology partners include Criteo, Kargo, Adobe, Pacvue and StackAdapt. The economics moved fast too: the minimum spend started at 200,000 to 250,000 dollars, dropped to 50,000 on April 13, and was removed entirely on May 5 with self-serve access. The reported CPM fell to a range of 25 to 60 dollars depending on the source, with a suggested CPC of 3 to 5.

The paradox: low CTR, high conversion

The number that traveled the most is conversion. On May 5, Criteo reported that traffic referred from ChatGPT converts at roughly 1.5 times the rate of other referral channels, with peaks close to twice the rate of search in categories like consumer electronics, lifestyle and home. It is worth being precise about what that figure is: it is a single Criteo report, based on a sample of 500 US retailers and more than 1,000 active brands, presented in two cuts (the 1.5x average and the 2x peak), not two separate studies. The counterpoint is CTR: estimates from Ahrefs put ChatGPT's click-through near 1.3%, against 29.2% for Google Search.

At first glance those two numbers fight each other. In reality they describe the same mechanic. ChatGPT is a conversational environment designed to keep the user inside, not to push them out, and that depresses CTR by construction: most people do not click because they keep talking. But whoever does click arrives far more qualified, because the ad appeared in response to a specific, contextual query. Fewer clicks, hotter ones. The operational consequence is direct: evaluating this channel by CTR is reading the wrong thermometer. The right unit is cost per conversion and value per click, never traffic volume. This is the measurable version of the shift we already described in Zenda's note on Answer Engine Marketing: influence is measured differently than the click.

The problem is not the number, it is where it comes from

The deeper problem is not the size of the numbers, it is where they come from. OpenAI reported the revenue. Criteo reported the conversion, and Criteo is not only a source but a commercial partner of the channel. Specialist coverage itself acknowledges that no independent verification at scale has been published. That does not make the data false, but it downgrades it from "evidence" to "interested-party signal", and that changes how it should be used. An interested-party signal is enough to justify a controlled test you measure yourself. It is not enough to justify reallocating budget.

Two tensions that decide entry

The first tension is between first-mover advantage and the immaturity of measurement. Entering now can mean cheaper CPMs and real learning, but the channel's attribution still rests on tools OpenAI launched weeks ago, and the reporting it returns is aggregate only: impressions, clicks, spend, average CTR, CPC and CPM, and conversions, never at the user level. A large team used to analyzing cohorts in Meta or Google will hit the ceiling quickly. The way to resolve that tension is not to trust OpenAI's dashboard or Criteo's number, it is to build your own incremental read: measure what happens with and without the channel, with server-side conversion under your control.

The second tension is new and underestimated: brand safety by adjacency. A well-labeled "Sponsored" tag solves disclosure, but it does not solve what the ad appears next to. In a traditional search engine, the ad sits beside a list of links. In ChatGPT it sits beside a generated answer that may be wrong, outdated or sensitive depending on the query. The reputational risk here is not that it looks like advertising, it is the generated context surrounding it, and to date the advertiser's control over that context is limited. For a corporate brand, that alone is not a detail: it is the question that decides whether the channel is in or out.

Who it's for (and who it isn't)

Geography rules. If you operate in Mexico or Brazil, the channel is imminent and the preparation work is now, not after launch. If you operate in Argentina or any other market in the region without a confirmed date, the move is not to prepare blindly: it is to watch the early adopters in Mexico and Brazil closely and learn from their data before spending a peso.

Team size defines the scope of that preparation. For a large operation with engineering resources, the standard is to have the server-side Conversions API configured before launch, because that is the only way to evaluate the channel by its real metric from day one. For a small team with no developer available, the reasonable MVP is to start with the pixel only and not commit budget until you see your own signals; heavy preparation is not justified for a channel that has not reached your market yet. In both cases the evaluation metric is the same and non-negotiable: cost per conversion and return, not CTR.

What we still don't know

As of today there is no conversion data verified by an independent third party: everything available comes from OpenAI or Criteo. There is no date for Argentina or for most of LATAM beyond Mexico and Brazil. It is unclear how the channel performs outside retail and e-commerce, which are precisely the categories where the reported numbers were best. The advertiser's control over brand adjacency is not yet documented in detail. And both pricing and the country list change month to month: every figure in this note is dated and can age quickly.

This content was developed with AI assistance and reviewed by the Zenda team. The bad ideas are 100% ours.