Retail took 39% of ChatGPT ads on 24% of US queries, Adthena finds

Retail and Fashion advertisers took 39% of the ChatGPT ad items Adthena observed in the United States while accounting for only 24% of query volume, and that concentration is the closest thing to a forecast available for the 31 European markets where sponsored placements began appearing on 24 August 2026. Adthena published the analysis on 21 August 2026, written by Serra Benstead, drawing on almost 850,000 AI search queries collected in the United States and the United Kingdom. PPC Land reported the findings on 24 August 2026. Adthena is explicit that the numbers predict European behaviour rather than report it, because they were captured before Europe opened. Thailand is not among the markets named anywhere in the analysis.

What OpenAI confirmed, and what actually switched on

OpenAI confirmed on 18 August 2026 that ChatGPT Ads was coming to 31 European markets, and described the move as its largest expansion to date. The same product update email that carried the European schedule also documented a campaign-level platform targeting control and confirmed that Brazil and Mexico had already activated. Six days later, on 24 August 2026, the placements themselves became visible across the 31 markets.

Three structural facts shape how the opening weeks behave, and all three are worth writing down before anyone reads a benchmark number. Access is gated at the start, routed through OpenAI's ads team, through agencies and through technology partners, with a self-service Ads Manager described as arriving later. Delivery is confined to the Free and Go subscription tiers, which means Plus, Pro and Enterprise accounts see no advertising at all, a boundary that has held since the pilot's first day. And OpenAI still characterises the platform as early in its development, despite the conversion optimisation, geographic targeting, custom audiences and pixel-based measurement it has accumulated since the United States pilot launched on 9 February 2026.

The gate matters more than it looks. A market where advertisers are approved one by one is a market with a known, countable and initially small set of bidders, and that is the condition every figure below depends on.

Why 39% of the slots went to 24% of the queries

The mechanism behind the headline number is a divergence between share of demand and share of supply. Users ask about a very wide range of things, but only a narrow set of advertisers has been approved to answer them commercially, so the brands whose catalogues intersect the largest number of conversational prompts end up occupying a disproportionate share of the slots that exist. Retail and Fashion ran at an ad frequency of 6.55% in the United States against a US average near 4.5%, and ads reached 4.47% of US queries overall.

Adthena identified Expedia as the first confirmed live ChatGPT ad, triggered on a single travel-planning prompt, with Best Buy close behind and landing two placements inside one response. Both fit the same profile: broad catalogues, comparison-heavy intent, and a query surface wide enough to touch a large share of conversations.

That is a claim about market structure rather than about consumer preference, and it is the part of the analysis that transfers to any market that opens later. Concentration at launch is a function of who has been let into the auction, not of what people are asking. It should compress as the advertiser field fills out, and Adthena's argument is that the brands holding position while the field is still thin are the ones that later become expensive to displace. That framing comes from a company that sells competitive monitoring for this channel, which does not invalidate the measurements but does sit alongside them.

Frequency and concentration are two different measures

The categories that ran hottest by frequency in the United States were not the categories that took the most slots, which is a useful reminder that the two figures answer different questions. One measures how often any sponsored unit appears at all inside a vertical. The other measures who occupies the units once they appear. The United States readings Adthena published were as follows.

Logistics, Home and Garden and Beauty all sit above Retail and Fashion on frequency while sitting below it on share of observed placements. Adthena does not explain the split, and the analysis does not offer a reason for it, so the honest reading is that a vertical can be well served by ads without being dominated by a few large advertisers.

One slot per answer, not a carousel

The format constraint is the second load-bearing figure. Across a March to May 2026 measurement window, ChatGPT served ads on 4.47% of US queries, and the average ad-bearing answer carried 1.06 ad items. There was no row of three and no shopping carousel. A single sponsored unit sat below the response. Google's AI surfaces averaged 3.53 items per ad-bearing answer over the same kind of measurement, behaving much closer to a carousel.

One slot per answer changes the arithmetic of the auction. The number of winners per query is one, so the pool of approved bidders competing for that one slot sets the clearing price directly. Adthena's position is that a shallow pool plus a single slot per answer keeps costs suppressed at the start, and that both conditions erode as more advertisers are approved in each market. Nothing in the analysis puts a date or a price on that erosion.

For anyone planning budget, the practical consequence of one slot per answer is that share of voice in this channel is closer to a binary than to a curve. In a carousel you can buy a lower position. Here you either take the slot or you are absent from the answer.

What does not travel from the United States benchmark

The read-across assumes the European product behaves like the American one, and on at least one documented dimension it does not. Personalised advertising is not initially available in the European Economic Area or Switzerland. Selection at launch draws on the current conversation topic plus general location and device type, a narrower signal set than the past-chat and prior-engagement inputs that inform delivery in the United States. Every US frequency and concentration figure in Adthena's dataset was produced by a system that had those additional signals available to it. A market where selection runs on live thread context alone is not the same auction, and the direction of the difference is not predictable from the American numbers.

The legal footing sits in Dublin. OpenAI Ireland Limited is the controller named for users in the European Economic Area and Switzerland, which places lead supervisory authority with the Irish Data Protection Commission. Measurement defaults add to the picture: automatic advanced matching, which transmits hashed customer information collected from website forms, became the default on existing web pixels on 17 August 2026 after a ten-day opt-out window, and the ChatGPT pixel initialises consent as true unless it is explicitly set to false or a stored denial is found. In jurisdictions where prior consent is a precondition for tracking, the lawful basis for that transmission rests with the advertiser rather than with the platform, whatever the default says.

One more exclusion holds across every market in the test. Dating, health, financial services and politics have been outside it since it began, which removes a large slice of advertising demand from the auction before it opens.

The controls keep arriving before the documentation

Adthena reports observing a negative phrases field inside ChatGPT Ads Manager, a search-style exclusion control, before any public notice of it appeared. That is not an isolated case. View-through conversion reporting surfaced through documentation revisions on 18 and 19 August 2026 rather than through an announcement, and automated bidding arrived preselected in new ad groups with a help centre page dated before the email that told advertisers it existed.

The operational implication is unglamorous and worth stating plainly. On this platform the interface is ahead of the changelog, so the release notes are not a reliable inventory of what your account can do or of what it is already doing on your behalf. Anyone running the channel should be opening the campaign settings and the ad group defaults on a schedule and reading them as the source of truth, rather than waiting to be told a control exists. A default that is switched on before it is documented is still a default you are accountable for.

The readings Adthena has not reconciled

Two numbers inside Adthena's own published record sit awkwardly against the current analysis, and a reader should know about both. Adthena's Monthly AI Data Pulse for June 2026 recorded a 0.00 percent ChatGPT ad frequency across 169,560 United Kingdom scrapes, and zero UK ad items among 29,237 catalogued across all markets that month. The current analysis presents the United Kingdom, live since 6 June 2026, as the closest available proxy for the 31 new markets, without reconciling that description against a June reading in which its own monitoring found no paid placements in the market at all.

The second is a frequency gap. Adthena counted 7,378 distinct ChatGPT advertisers during the week of 13 to 20 July 2026, with the United States accounting for 60.1 percent, and reported US ChatGPT ad frequency of 24.7% in that window. The figure published for the March to May period is 4.47%. Both are described as United States ad frequency, and the analysis sets out neither the methodology difference nor the trajectory between them.

Independent measurement does not settle it either. Similarweb published a dataset on 17 August 2026 recording sponsored placements in 26 percent of ChatGPT responses at a 0.50 percent click-through rate. Different panels and different denominators produce numbers that cannot be laid against each other cleanly, and there is no third-party measurement standard for conversational advertising surfaces to arbitrate between them.

What this means for Thai marketers

Thailand is not among the markets in the source, and the analysis does not say when or whether ChatGPT Ads will reach Thailand. There is no stated APAC timeline beyond Japan and South Korea activating on 22 June 2026, so anyone reading a Thai launch date into this story is inventing one. What follows is Relevant Audience's reasoning, not a finding from Adthena or from OpenAI.

The transferable part is the structure, not the percentages. If and when a new market opens under the same gated model, the same shape should be expected: a small approved field, a single slot per answer, broad-catalogue advertisers taking a share of placements well above their share of queries, and regulated categories absent. For a Thai advertiser that is an argument for preparation rather than for waiting. The assets that would matter on day one are the ones that already matter for ecommerce marketing: a clean product feed, catalogue breadth that intersects real questions, and landing pages that answer a comparison prompt rather than restating a category page.

The second implication is about the organic side of the same surface. A single sponsored unit sitting below an answer does not change the fact that the answer itself is the thing users read, and the work of being cited inside that answer is generative engine optimisation rather than media buying. In a channel where paid inventory is one slot per answer, the unpaid part of the response is the larger surface by a wide margin, and it is available in Thailand today whether or not the ad product ever arrives. Brands that treat GEO as the standing workstream and paid placement as an option that may open later have the ordering right.

The third implication is measurement discipline. The consent question raised by an advertising pixel that initialises consent as true is a question for any advertiser operating under a prior-consent regime, and Thailand's own data protection framework is a reason to settle that question before adopting a tag rather than after. That is a legal and compliance decision for each business and its advisers, not something this article can resolve.

FAQ

Are ChatGPT ads available in Thailand?

The source does not say. Thailand is not among the 31 European markets that activated on 24 August 2026, nor among the earlier markets listed, and no timing for Thailand is stated anywhere in the analysis.

Do paying ChatGPT subscribers see these ads?

No. Delivery is confined to the Free and Go subscription tiers, so Plus, Pro and Enterprise accounts see no advertising, and that boundary has held since the pilot's first day.

How many ads appear inside a single ChatGPT answer?

Approximately one. Across Adthena's March to May 2026 window the average ad-bearing answer carried 1.06 ad items, against 3.53 items per ad-bearing answer on Google's AI surfaces.

Can advertisers in Europe use personalised targeting?

Not at launch. Personalised advertising is not initially available in the European Economic Area or Switzerland, and selection draws on the current conversation topic plus general location and device type.

Do Adthena's United States figures predict what Europe will do?

Adthena says they predict rather than report, and states that the data was captured before the European launch. The figures describe the United States and the United Kingdom only, and the European product runs without the personalisation signals that produced them.