Advertising inside
ChatGPT. What we
know so far.
ChatGPT Ads went live across 31 European markets in the week of 24 August 2026 — six months after the first US test.12 Hungary, Czechia and Slovakia are all in — but self-service Ads Manager access is still listed as "coming soon" for all three, so the only route in today is OpenAI's Ads Solutions team, an agency partner or a technology partner.18 This is a working briefing on the audience, the format, the controls and the caveats, with every number linked to its source.
How it got here
Six months from US test
to full European rollout.
- 9 February 2026 First test of ads in ChatGPT, United States only.1
- 26 March 2026 Canada, Australia, New Zealand.1
- 7 May 2026 United Kingdom, Mexico, Brazil, Japan, South Korea.1
- 10 August 2026 OpenAI publishes its consolidated advertising policies.6
- c. 20 August 2026 Platform targeting — iOS app, Android app, web — added to campaign setup.7
- Week of 24 August 2026 — now 31 European markets open, Hungary, Czechia and Slovakia included. Buying runs through OpenAI's Ads Solutions team, agency partners and technology partners. OpenAI's own availability table still lists self-service Ads Manager as "coming soon" for Hungary, Czechia and Slovakia.2818
What that means for a CEE media plan
There is roughly one week of live data in this region. Every performance number on this page comes from US- and UK-led early campaigns, is advertiser-reported, and should be treated as a planning starting point rather than a benchmark. OpenAI has not published cross-advertiser CTR, CPC or conversion figures.9
The audience, on the other hand, is already measured locally — and it is large. That is the part of the case that does not depend on anyone's early-adopter anecdote.
Our read: treat Q3–Q4 2026 as a learning budget in a channel with almost no competitive density, not as a performance line item that has to beat your Google account in month one.
The audience
Hungary, Czechia, Slovakia.
Three markets, three different measurement systems — so the figures below are not directly comparable with each other. Each card states its own source, base and reference period.
- 1,907,876ChatGPT mobile app — monthly reach, July 2026
- 1,504,772ChatGPT web — monthly reach, July 2026
- 592,651ChatGPT app — daily reach
- 13:24average daily time in the app, per user (22:01 on web)
- 1,138,724daily app visits — roughly 1.9 sessions per daily user
NMHH, "Az online médiatér közönsége — 2026 július", published 5 Aug 2026, based on Egységes Digitális Mérés (Gemius) data. Base: Hungarian internet users 16–75. App and web reach are reported separately and are not deduplicated, so they cannot be added together.3 For context, Google Gemini web reached 525,640 monthly in the same period.3
- 32%of people aged 16+ used AI content-creation tools — official statistics office, Q2 2025
- 78%among 16–24 year-olds; 60% among the university-educated working population, 16% among those with basic education
- 34/30% men vs. women — one of the narrowest gender gaps of any digital behaviour ČSÚ tracks
- 20%of the working population use AI for work specifically
- 30%start an information search in an AI chatbot rather than a search engine (55%) or social (15%)
Czech Statistical Office (ČSÚ), ICT in households and ICT users, fieldwork Q2 2025, population aged 16+, published 11 Nov 2025.10 Search-behaviour split and ChatGPT's position as the best-known and most-used tool: Deloitte TMT Highlights Česká republika 2026, fieldwork Feb–Mar 2026, n=341 aged 15+.5 No Czech industry body publishes a deduplicated ChatGPT reach figure — NetMonitor models global services, but ChatGPT is not among the titles released.11
- 2.7MSlovaks aged 14–79 using AI tools — up 1.2M year on year (+79%)
- ~1.7Muse ChatGPT specifically — 62% of all AI users
- 533,000daily AI users, up from 140,000 a year earlier (+282%)
- 49.7%of AI users are women, up from 44.1% — the audience has stopped being male-skewed
- 700,000users aged 50+, up from 283,000 (+150%)
Median SK, MML-TGI survey, fieldwork 5 Jan – 29 Mar 2026, Slovak population aged 14–79; reported 17 June 2026.4 83% use AI to find or explain information quickly, 30% to write or edit text, 24% to generate images.4
Now subtract the subscribers
Ads only run on the free tiers.
That is still ~9 in 10 users.
This is the number most plans get wrong. "Ads may appear for users on the Free and Go plans. Plus, Pro, Business, Enterprise, and Edu accounts will not have ads."19 So every reach figure on this page has to be discounted by the paying base — but that base is smaller than people assume. OpenAI has reported more than 900 million weekly active users;22 reported paid accounts run to roughly 50 million consumer subscriptions plus around 9 million business seats.23 Note that Go is a paid tier that does show ads, so it stays in the addressable pool.
Weekly active users, global~900M
Less ad-free paid tiers~59M
Addressable share~93%
Applied to HU app reach
1,907,876 monthly~1.77M
Our arithmetic, not OpenAI's. The paid share is global and may differ in CEE, where willingness to pay for subscriptions is lower — so treat ~93% as a floor rather than a ceiling. OpenAI does not publish a free-versus-paid split by market.
European context: 32.7% of people aged 16–74 in the EU used generative AI tools in 2025, per Eurostat — 25.1% for personal reasons, 15.1% for work, 9.4% for formal education. Eurostat's country breakdown does not publish separate figures for Hungary, Czechia or Slovakia, which is why the national sources above are used instead.12
The format
One unit: the native
text Sponsored Card.
There is no display inventory here. The ad is a sponsored card, clearly labelled and visually separated from the answer, appearing below ChatGPT's response inside the conversation.1
Where it appears
Below the answer, inside the chat. "Ads may appear for users on the Free and Go plans. Plus, Pro, Business, Enterprise, and Edu accounts will not have ads."19
What it looks like
Concise native text — a short headline, a line of copy and a destination link, written to read as a useful recommendation rather than a banner. OpenAI's guidance is to keep titles focused and within its recommended character limits.13
How it gets chosen
A relevance-weighted auction: your bid is multiplied by a relevance score derived from your context hints, landing page, ad title and ad copy. Specificity buys position more cheaply than bidding does.14
Targeting & buying
The controls you
actually get.
Contextual matching is the engine — everything else is a constraint you place on top of it.
| Control | What is available | Notes |
|---|---|---|
| Contextual | Automatic. Ads are matched to the topic of the live conversation. Globally OpenAI also uses past chats and past ad interactions — but personalised ads are not initially available in the EEA or Switzerland, so in Hungary, Czechia and Slovakia the match is contextual only.119 | This is the basis of the whole channel. You are buying a moment of stated intent, not a demographic — and in CEE that is all you are buying, which makes the creative and the context hints do all the work. |
| Context hints | Free-text descriptions of the conversations, topics or keywords where your product may be relevant, set per ad group. You can add a large number of them — up to around a hundred.1424 | Directional, not literal. The system reads them semantically, so write intent, not keywords. The count is not the lever — coverage of how people actually phrase the need is. |
| Location | Country, county/region and city-level targeting are all available to us in Ads Manager for Hungary, Czechia and Slovakia.24 | Worth knowing: OpenAI's public help documentation still describes sub-country targeting as US-only.15 The platform is ahead of its own docs here — we plan against what the account actually shows. |
| Platform / device | Multi-select at campaign level: iOS app, Android app, web. Web covers both desktop and mobile web.715 | Reporting is coarser than targeting — insights group results into Mobile and Desktop only, with mobile web counted as Mobile.7 |
| Audiences | Custom audiences, with inclusion and exclusion.215 | Useful mainly for suppression — excluding existing customers — in the first weeks. |
| Objective & pricing | CPM for reach and awareness, CPC for traffic and engagement, oCPC — conversion-optimised cost per click — for tracked conversions.15 | We recommend starting on CPM. oCPC needs conversion history before it can optimise anything. |
| Budget | Campaign-total or daily. Daily spend will not exceed twice the daily budget on any one day, or seven times it across a seven-day period.15 | Plan weekly, not daily — the pacing logic is explicitly built around a seven-day window. |
| Measurement | OpenAI Pixel, Conversions API, and third-party measurement integrations.2 | Get the pixel live before the first impression. Retro-fitting attribution to this channel is painful. |
Pricing: campaigns are bought on CPM or CPC. Rates in Hungary, Czechia and Slovakia are quoted per campaign and depend on category competition and the relevance score of the creative — ask us for a current rate card.
Early benchmarks — handle with care
What the first advertisers
are reporting.
None of these are OpenAI figures. They come from a small, self-selected group of advertisers in mature markets, in their first weeks of spend, and they will not necessarily hold in CEE.9
≈ 0.9%, range 0.3–1.5%
A frequently cited Adthena figure puts ChatGPT ad CTR near 0.91%; industry ranges reported by early advertisers run 0.3–1.5%, with local and personal services at the top end (0.8–1.5%) and retail at the bottom (0.4–1.0%).916
$25–60 in mature markets
The format launched at a flat $60 CPM and category rates have since settled to roughly $25–60. These are US and UK rates and should not be read across to CEE.917
2–6% cross-industry
Early advertiser estimates: education 6–11%, local services 6–10%, SaaS and B2B 4–8%, ecommerce 3–6%. High relative to display, which is the point — the click arrives mid-decision.16
Why the CVR looks high and the CTR looks low: the card only surfaces when the conversation is already about the category, so impressions are few and qualified. Judged on impression volume this channel looks small. Judged on cost per qualified action it can look very different — which is exactly why the pixel matters more than the CTR column.
Who can run, and who already is
Check your category
before you plan a budget.
Cleared to run
- Retail & ecommerce
- Grocery
- Home & furniture
- Software & SaaS
- Automotive
- Travel
- Entertainment & streaming
- Education & training
- Telco
- B2B services
- Events & weddings
- Beauty & personal care
Allowed, but needs approval first
- Financial services
- Banking & lending
- Insurance
- Healthcare
These are the everyday categories most likely to trip an advertiser up. They are not banned — they need category approval before the first impression, so build two to three weeks of lead time into the plan.6
Not allowed at all
- Adult content & dating
- Alcohol
- Tobacco & vaping
- Gambling & casino
- Illegal drugs
- Counterfeit goods
- Weapons
- Political & advocacy
- Housing listings
- Employment listings
- Legal services
- Unsubstantiated wellness claims
Per OpenAI's advertising policies, last updated 10 August 2026.6 Note that housing, employment and legal services are prohibited outright — a restriction Google and Meta handle with special-category rules instead, so a plan ported from those platforms will fail here.
Who moved first
Retail and commerce dominated the opening wave in the US. Named participants include Target, Williams-Sonoma and Albertsons in retail and grocery, The Knot in wedding planning, Adobe in software, Audible in entertainment, and Ford and Mazda in automotive.21
The pattern is consistent: every confirmed early advertiser sits in a product-driven category where purchase intent is tightly bound to the kind of question people actually bring to ChatGPT.21 Pacvue puts roughly 20% of ChatGPT conversations as carrying shopping intent.22
Agencies reporting on the first months name five verticals as the strongest performers: SaaS, ecommerce, online education, health and wellness, and financial services — the last two on educational, trust-building creative rather than urgency.20
What this means for a CEE brand: if your category is on the cleared list and your customers ask comparison questions before they buy, you are the profile that has worked so far. If you sell on impulse or on price alone, wait.
One more thing worth putting in front of a legal team: personalised ads are not initially available in the EEA or Switzerland, so no behavioural profile is built on the way to your impression.19 Users can dismiss an ad, delete their ad data and view their ad history, and Free-tier users can turn ads off in exchange for fewer daily messages.1
How we'd start
A first CEE test
that actually teaches you something.
Expect a CTR above 2% — and expect it to fall
Early campaigns are clearing 2%+ click-through rates.24 We would not build a business case on that number holding. It almost certainly reflects novelty: the card is new, it is the only ad on the screen, and people click things they have not seen before. Mature-market benchmarks already sit at 0.3–1.5%.916 Bank the early efficiency, model the decay.
CPM, one market, one category
Start on CPM in a single market so the relevance score has a clean read. oCPC has nothing to optimise against until conversion history exists.
Context is the only targeting — and it is accurate
This is the single most important thing to understand about the channel. In this region there is no demographic, no interest and no behavioural layer: the conversation is the targeting.19 You can load a large number of context hints, up to around a hundred, but the count is not what makes it work.24 On the first campaigns the match has been strikingly precise — the ad lands in the conversation it was written for.24
Creative volume: untested, probably decisive
We have not yet been able to test running many creatives in parallel, so we will not pretend to know the answer. But given that relevance scoring reads the ad title and copy directly, the number and variety of creatives is almost certainly going to be a significant lever. Plan to produce more variants than a comparable search campaign would need, and treat the first flight as the experiment that tells us how many.
Let's talk.
We buy ChatGPT Ads in Hungary, Czechia and Slovakia. Tell us the category and the market and we'll come back with reach, a rate and a test plan — no pressure, just possibilities.
Sources
Every figure on this page, traced.
- OpenAI, Testing ads in ChatGPT — format, tiers, matching signals, user controls, rollout dates. Updated 11 Aug 2026.
- OpenAI, ChatGPT Ads expands across Europe — 31 European markets, CPM/CPC bidding, geo-targeting and custom audiences, OpenAI Pixel and Conversions API. Published 18 Aug 2026.
- NMHH, Az online médiatér közönsége — 2026. július, published 5 Aug 2026, based on Egységes Digitális Mérés (Gemius) data. Base: Hungarian internet users aged 16–75.
- Median SK / MML-TGI, reported by TECHBYTE.sk, 17 Jun 2026. Fieldwork 5 Jan – 29 Mar 2026, Slovak population aged 14–79.
- Deloitte, TMT Highlights Česká republika 2026, published 19 May 2026. Fieldwork Feb–Mar 2026, n=341, aged 15+.
- OpenAI, Advertising policies, last updated 10 Aug 2026 — prohibited and restricted categories, sensitive-context restrictions.
- PPC Land, ChatGPT Ads gains 3-platform targeting as reporting stays at 2 buckets, Aug 2026 — citing OpenAI help-centre documentation updated 20–21 Aug 2026.
- Search Engine Land, ChatGPT Ads are expanding to 31 European countries, 19 Aug 2026 — access via Ads Solutions team, agency and technology partners ahead of self-service.
- 360ROI, ChatGPT Ads CTR and benchmarks — citing an Adthena CTR figure of ~0.91% (9 Jun 2026) and noting OpenAI has published no aggregate cross-advertiser benchmarks.
- Český statistický úřad, ICT in households and ICT users, and Umělou inteligenci používá třetina populace, published 11 Nov 2025. Fieldwork Q2 2025, persons aged 16+.
- Lupa.cz, Netmonitor nově modeluje data globálních služeb — global services now modelled in Czech measurement; ChatGPT is not among the titles released.
- Eurostat, 32.7% of EU people used generative AI tools in 2025, published 16 Dec 2025. Base: individuals aged 16–74, three months preceding the survey.
- OpenAI Help Center, Create ads for ChatGPT — creative guidance. Exact character limits are not published by OpenAI; third-party specs circulating online are unofficial.
- Context Hints, Guide to context hints — quoting OpenAI's definition of context hints, plus the relevance-weighted second-price auction and hint-count guidance. Third-party source.
- OpenAI Help Center, Create campaigns for ChatGPT — CPM, CPC and oCPC; country targeting with state/DMA/ZIP in the US only; iOS app, Android app and web platform targeting; budget and pacing rules.
- Lapis, ChatGPT Ads conversion rate benchmarks by industry, published 25 Jun 2026, updated 1 Aug 2026 — early advertiser estimates, not OpenAI data.
- Index Lab, What ChatGPT ads cost in 2026, data dated 23 Jul 2026 — agency-observed CPM and CPC ranges in mature markets.
- OpenAI Help Center, Ads Manager availability — self-service is live in AU, BR, CA, JP, KR, MX, NZ, UK and US; Hungary, the Czech Republic and Slovakia are listed as "Coming Soon". OpenAI's European announcement named nine markets by name; the reported full composition of the 31 — all 27 EU states plus Iceland, Liechtenstein, Norway and Switzerland — comes from Index Lab, a third-party source. Verify eligibility before booking.
- OpenAI Help Center, Ads in ChatGPT — "Ads may appear for users on the Free and Go plans. Plus, Pro, Business, Enterprise, and Edu accounts will not have ads." and "Personalized ads are not initially available in the European Economic Area (EEA) or Switzerland."
- AdVenture Media, 5 industries winning with ChatGPT Ads in early 2026, published 23 Feb 2026 — qualitative agency observation; the article publishes no CTR, CVR or CPA figures.
- Floodlight, Which brands are advertising on ChatGPT, published 23 Feb 2026, last modified 6 Apr 2026 — named advertisers compiled from company announcements. No share-of-spend data is published.
- Pacvue, ChatGPT Ads in 2026: early results and best practices — 900M+ weekly active users, and roughly 20% of ChatGPT conversations carrying shopping intent.
- Paid-account figures compiled by Nerdynav, ChatGPT statistics, Aug 2026, from OpenAI disclosures — 12M+ Plus subscribers, 9M+ business seats, 50M+ total paying consumer subscriptions. OpenAI does not publish an official free-versus-paid ratio, and these categories may overlap.
- Madhouse, observed directly in ChatGPT Ads Manager and in live campaigns, August 2026 — available geo-targeting levels for HU/CZ/SK, context-hint capacity, early click-through rates and observed targeting accuracy. First-hand platform observation, not OpenAI-published data, and based on a small number of early campaigns.
Prepared by Madhouse, 30 August 2026. Market figures are reproduced from the sources listed above and are not Madhouse research. Performance benchmarks marked as early or advertiser-reported are directional only. ChatGPT and OpenAI are trademarks of OpenAI; Madhouse is not affiliated with or endorsed by OpenAI.