ChatGPT ads: how they work and who can advertise

ChatGPT ads are live in Slovakia and Czechia

OpenAI began testing ads in ChatGPT on 9 February 2026, in the US only. By the end of the summer the test had become a real platform: on 24 August 2026 ChatGPT Ads expanded to 31 European markets, Slovakia and Czechia among them, and since 31 August companies in those markets can open an ad account themselves through the self-serve Ads Manager. Both countries appear on the official availability list.

For a business that means one thing: this is no longer something that might arrive one day. A campaign can start today, on a credit card, without an agency and without going through OpenAI's sales team. It is also still a beta — and that shows in several places.

The timeline is short and worth remembering, because it explains why most guides on the topic are older than the platform they describe:

• 9 Feb 2026 — test in the US for logged-in adult users,
• 26 Mar 2026 — expansion to Canada, Australia and New Zealand announced,
• 11 Aug 2026 — United Kingdom, Mexico, Brazil, Japan, South Korea,
• 24 Aug 2026 — 31 European markets, including Slovakia and Czechia,
• 31 Aug 2026 — self-serve access through Ads Manager in those markets,
• 16 Sep 2026 — AI tooling in Ads Manager, HubSpot and Shopify integrations, Sponsored Agents tested in the US.

Who actually sees the ads

Ads are shown only to users on the Free and Go plans. Plus, Pro and company plans (Business, Enterprise, Edu) stay ad-free. OpenAI also withholds ads from accounts it identifies as under 18, and a Free-plan user can opt out of ads in exchange for fewer daily messages.

This is the single most important planning constraint, and most write-ups skip it. If you sell B2B software or services to companies, part of your audience sits in exactly the plans where ads never appear — company licences are ad-free by definition. It does not kill a campaign, since decision-makers often use a personal account, but you cannot plan against a reach of "everyone on ChatGPT".

Ads also stay away from sensitive contexts. OpenAI excludes conversations about physical and mental health, politics, self-harm, weapons, gambling and other categories it marks as unsuitable for ad placement. The inventory is not the full volume of conversations, only the safe part of it.

What the ad unit looks like

There is one format so far, and it is restrained. The ad appears below the answer, visually separated, always labelled as sponsored. It carries the advertiser name, a favicon, a headline, a short description, an image and a landing page.

What matters is what the format does not do: it does not shape the answer. OpenAI states repeatedly that ads do not influence responses and that advertisers get no access to conversations, chat history or memory. They receive aggregate numbers only — impressions, clicks, conversions.

Since 16 September 2026 OpenAI has been testing Sponsored Agents with selected US advertisers: after clicking an ad, a user can start a clearly labelled conversation with a business-sponsored agent, separate from the original chat. It is not available in Europe yet, but it shows where the format is heading — from a link to a dialogue.

How the system decides which ad to show

This is where ChatGPT Ads diverge most from search advertising. There are no keywords in the sense Google Ads trained us to expect. The system works from the context and intent of the whole conversation, the landing page, the ad headline and copy, and — when ad personalisation is on — signals from the user's broader ChatGPT use.

The advertiser's lever is context hints at ad-group level: descriptive phrases stating the topics and situations where the product is relevant. OpenAI is explicit that these are not exact-match keywords and do not guarantee delivery in any particular conversation. They guide the model; they do not filter for it.

Selection runs through a relevance-weighted second-price auction. You set a maximum bid per ad group — CPM for reach campaigns, CPC for traffic campaigns. OpenAI suggests starting at USD 3–5 per click. Ads Manager adds bid-strength guidance on whether a bid is competitive enough for the ad to deliver at all.

The practical consequence: stop treating the landing page as where the campaign ends and start treating it as part of the targeting. It is one of the inputs the system uses to decide when your ad belongs. A thin page with three sentences gives the model nothing to work with.

Who may advertise — and who may not, yet

This is where the conversation ends for a large share of Central European businesses, and it is better to know before opening an account than after a rejection.

OpenAI's ad policies currently admit a limited set of consumer categories: household and consumer goods, local services, travel and entertainment, digital products and education. The list is expected to widen over time.

Disallowed categories include gambling, alcohol and tobacco, recreational drugs, adult content and dating, counterfeits, political content and contested social issues. Separately: ads for individual job listings and for property sales or rentals are prohibited — you may link to a platform, not to a specific listing.

And one rule that almost every overview omits, although it carries the biggest weight in this region: ads for financial services, health services and legal services are generally prohibited outside the US. Inside the US, OpenAI approves them case by case, often against proof of licensure. So if you do mortgages, insurance, investments, a clinic or a law firm, ChatGPT Ads is not a channel for you today — whatever the budget.

Tools that make no medical or financial claim are a different matter. A budgeting app, a course about personal finance, fitness equipment or a general wellness product without health claims are allowed.

The account belongs to a business, not an individual. OpenAI does not onboard individual advertisers, and an agency cannot create an account on a client's behalf — the client creates it and invites the agency afterwards. The legal entity that advertises and gets billed must be based in a country on the availability list.

What opening an account involves

There are five steps: create the account in Ads Manager, verify it, complete the account details, set up billing, invite your team. Identity verification runs through Persona and the account goes into a manual review that also checks whether your products and services fall into an eligible category at all. The queue is not instant.

Two things deserve a second look, because they cannot be changed later: the account country, the billing currency and the time zone. Get them wrong and the only remedy is a new account. Campaigns also will not serve until the account name and logo are filled in — exactly as they should appear in the ad unit.

What it costs and how it is measured

You pay per impression or per valid click. Campaign objectives come in three: Views (CPM), Clicks (CPC) or Conversions — conversion-optimised campaigns on click billing (oCPC) still charge per click while optimising toward the conversion.

A daily budget is an average, not a ceiling: with an unchanged budget the published limits are twice the daily amount in a single day and seven times it across a week. Anyone planning to the day will be surprised.

Measurement is what you would expect from a beta. Reporting covers impressions, clicks, spend, CTR, average CPC and CPM, and conversions. Impressions and clicks refresh roughly every 15 minutes, but spend lags by 7 to 8 hours — a spend of zero does not mean nothing has been charged. Conversions are measured through the OpenAI Pixel, the Conversions API or a measurement partner; UTM parameters on landing page URLs survive the click, so you can isolate the traffic in your own analytics too.

There are no benchmarks, and OpenAI says so outright: the platform has no cross-industry performance baselines yet. Numbers of the "average ChatGPT CPM is X" kind circulating online are trade-press estimates, not platform data. Treat them accordingly.

What this means in practice

The order of operations that works is not the one a new platform invites:

1. Check your category first. Five minutes in OpenAI's policies saves a week of waiting for a rejection. Finance, healthcare and legal are out outside the US; property and recruitment are restricted at the listing level.

2. Rewrite the landing page before the ad copy. It is an input to targeting. A page that names the problem, who the solution is for and what makes it different gives the system something to work from. This is, incidentally, the same work organic visibility in AI search demands — done once, it pays on both sides.

3. Set up measurement before launch. UTMs on every URL, the pixel or Conversions API, and a separate segment for AI-referred traffic. With spend lagging 7–8 hours and no benchmarks to lean on, your own measurement is the only firm ground.

4. Budget it as a test, not as a channel. Delivery fluctuates during the beta and OpenAI says as much. Smaller budget, longer window, decisions on conversions rather than clicks.

5. If you sell products, look at the feed. Ads Manager can build campaigns from product feeds, and the ChatGPT Ads app for Shopify has been available outside the US since 23 September 2026, in every market where ads run. For an e-shop that is the shortest way in.

What you should not do: cancel Google Ads and move the budget. ChatGPT Ads is currently a complementary channel with a smaller and differently composed inventory, not a replacement for a performance channel with ten years of data behind it.

What to watch

The platform is in beta, and OpenAI warns upfront that formats, inventory and the way campaigns are bought, measured and optimised will keep changing. That is not a footnote — a setup that works today may behave differently in a month.

The second thing is regulation. On 31 August 2026 the European Commission designated ChatGPT a Very Large Online Search Engine (VLOSE) under the Digital Services Act. A designated service has four months to meet the heavier obligations, which include a public ad repository. For an advertiser that has one very concrete consequence: what you run in ChatGPT will stop being private. As with the Meta Ad Library, competitors will be able to look at your campaigns — and you at theirs.

The third is attribution. A conversation with a model is long, the path to a decision is not linear, and part of the effect lands somewhere other than the last click. Judge the campaign on a single last-click view and you will most likely underrate it.

In short

ChatGPT Ads are available here, self-serve and cheap to test, but they come with three hard edges: only Free and Go users see them, the eligible categories are still narrow and exclude finance, healthcare and legal outside the US, and the platform is a beta without benchmarks. For an e-shop, a local service, travel or an education product it is worth testing now, while the auction is still thin. For a regulated industry it is a topic for next year.

If you need a read on whether your category will pass, and pages prepared so that you show up in the answers and not only in the ads, get in touch.