Polski
AI advertising · News

Ads in ChatGPT: how they work and when they will reach Poland

Ads in ChatGPT have been running since February 2026 and already appear in half of all answers in the US. In Poland there are none, and the blocker is not technology but EU law. I break down the formats, the rates, the rollout timeline and what is really missing before a launch in Poland.

Sebastian Baryś
Sebastian Baryś
1 August 2026 21 min read

Introduction

English version of an article first published in Polish on 1 August 2026. Read the Polish original.

Ads in ChatGPT are real, they have been running since 9 February 2026, and at the start of July they already appeared alongside 51% of answers in the United States. In Poland there are none at all, and nothing suggests that this will change in the coming weeks. The reason is not technical. It is legal.

The distinction matters, because in the Polish discussion the topic exists in two extreme versions. The first: "ads in ChatGPT are the future, get your budgets ready". The second: "it is an American experiment, it does not concern us". Both miss what the data shows. The system is already a mature advertising product with its own dashboard, auction and billing models, and at the same time it is entering the European Union more slowly than anywhere else, for reasons that are known precisely.

I take it apart piece by piece: what ads in ChatGPT look like and how they work, how much they cost, how the rollout went country by country, what exactly is blocking the launch in Poland and which things are worth doing before the system opens up.

You will learn:

  • What exactly ads in ChatGPT are, where they appear and who they apply to at all
  • Which formats are live today and how a conversational banner differs from a product carousel
  • How targeting works when it is based on the context of a conversation instead of keywords
  • What it costs to get in: from a $60 CPM at launch to a click-based model with no entry threshold
  • The full timeline: from the US test in February to the public Ads Manager of 22 July
  • Why Poland is still out of reach and which regulations really block the launch
  • How an ad in ChatGPT differs from an ad in Google AI Mode and where it performs worse
  • A concrete list of things to do before the system opens up to the Polish market

What ads in ChatGPT are

An ad in ChatGPT is a labelled block that appears below the model's answer, on a slightly highlighted background, with a label saying that it is sponsored. It is not woven into the answer and, according to OpenAI's declaration, it does not change what the model says. You buy a place next to the answer, not the answer itself.

The second thing to understand from the start: not everyone sees ads. They go only to logged-in adult users on the free plan and the Go plan at $8 a month. Plus, Pro, Business, Enterprise and Edu accounts stay ad-free. This has a consequence that is rarely mentioned: some of the users with the highest purchasing power are outside the ad system by definition.

The third difference from everything we know from search engines concerns the moment of contact. An ad on Google meets a query: short, stripped of context, typed into a text box. An ad in ChatGPT lands in the middle of a conversation in which the user has already described their situation, budget and constraints, and has sometimes said outright what they do not want. That is completely different material to match against, and a completely different discipline from bidding on keywords.

In a search engine the advertiser guesses intent from three words. In ChatGPT the user may have described that intent in three paragraphs, and that is everything this market is playing for.

Formats: what it looks like

The formats are simple for now, but one of them has no equivalent in any other advertising system.

Formats available in the system

01

Product carousel

below the answer

Product cards with a photo, a price and the shop name, scrolled horizontally. A format built for shopping queries and tied to the payment integrations inside ChatGPT.

02

Conversational banner

below the answer

A creative in which the user can ask ChatGPT follow-up questions about the advertised product without leaving the conversation. The biggest difference from everything we know from search.

03

Ad card

at the end of the answer

The basic block with a sponsorship label, on a slightly darkened background. Refreshed in July 2026 together with a batch of new manager features.

04

Multi-advertiser unit

below the answer

Several advertisers grouped in a single block. A way, still in testing, to increase the number of available slots without increasing the number of ads the user sees.

Every format sits below the answer. None of them enters its content.

The conversational banner: something that exists nowhere else

The most interesting format lets the user ask a question about the advertised product without leaving the conversation. Instead of a click that leads to a landing page there is a dialogue: a question about size, availability, the difference from a competitor's model. This turns the classic way of thinking about creatives upside down, because the headline and description stop being the whole message. They become the opening of a conversation whose content the advertiser does not fully control.

The practical conclusion for anyone who will set this up one day: the quality and completeness of product information becomes more important than a catchy headline. If the model has nothing to answer a question about a technical detail with, the conversation dies at exactly that point.

In July 2026 OpenAI refreshed the ad card format and added eight features to the dashboard at once: suggested ad drafts, an overview tab, first-party audience lists, geographic exclusions, conversion-optimised bidding, integrations with AppsFlyer and Adjust, and an interface for managing campaigns in bulk. The pace at which features are being added tells you more than any single one of them. This is what a product looks like once it has stopped being a pilot.

Targeting without keywords

This is where the biggest difference from Google Ads lies. In ChatGPT you do not bid on keywords. Ad selection is based on the context of the ongoing conversation, not on matching a string of characters to a list of phrases.

So the advertiser describes situations, not queries. This is done in a field called "context hints", where you write down the kinds of conversation in which the ad makes sense. Not "women's hiking boots", but "the user is planning their first multi-day hike and is asking how to put their kit together". On top of that come first-party audience lists and geographic exclusions, added in July.

Personalisation, if the user allows it, draws on previous conversations, memory, the history of interactions with ads and data supplied by advertisers. And exactly this element is subject to the strictest restrictions in the European Union, which I come to shortly.

We spent eighteen years learning to think in keywords. This system requires thinking in scenarios, and that is a change of habit, not a change of tool.

How much it costs

The history of prices in ChatGPT is a textbook example of what happens to a new channel once the first wave of curiosity subsides. Launch in February 2026: a default maximum bid of $60 per thousand impressions and an entry threshold in the region of $250,000. Ten weeks later rates had slid to around $25, the minimum had fallen to $50,000, and OpenAI had added cost-per-click billing.

What happened to rates in ten weeks

CPM, February 2026 (launch) 60 USD
CPM, April 2026 approx. 25 USD
CPC, model introduced in spring 3–5 USD

250k

entry threshold in dollars at launch

50k

threshold after the spring 2026 cut

0

minimum in the self-serve dashboard

From a channel for big brands to a channel with no entry threshold, in half a year.

The auction is a second-price model weighted by relevance: the winner is not the bid alone, but the bid multiplied by the fit with the conversation. There are two objectives to choose from: reach, optimised for impressions, and clicks. Conversion-optimised bidding was added in July.

The opening of the self-serve dashboard on 22 July removed the minimum budget. Today any approved advertiser can start a campaign, with any amount. Best Buy, Lowe's and VistaPrint were named among the first to use it, and Target, Ford, Adobe and Expedia in the February pilot wave.

Rollout timeline

Laid side by side, the dates show something that is not visible in individual news items: half a year passed between the first test and a public advertising dashboard. That is very fast for an ad system built from scratch.

Six months of building an ad system

  1. 16 January 2026

    Tests announced in the US

    logged-in adult users on the Free and Go plans

  2. 9 February 2026

    First live ads

    Target, Ford, Adobe, Expedia in the first wave

  3. 5 May 2026

    Campaign manager in beta

    first version of self-serve buying, US only

  4. May 2026

    Canada, Australia, New Zealand

    English-speaking markets outside the EU regime

  5. 6 June 2026

    United Kingdom, the first European market

    Europe, but outside the EU, and that is the heart of the matter

  6. July 2026

    Japan and South Korea

    in Japan, ads in 18.8% of answers

  7. 22 July 2026

    Public "Advertise in ChatGPT" dashboard

    end of the minimum budget; US and United Kingdom

  8. date unknown

    European Union, including Poland

    France, Germany and Ireland in preparation

Eight steps in half a year, and one that still has no date.

Where ads run and where they do not

United States

February 2026 full self-serve

Canada

May 2026 bought through a sales rep

Australia

May 2026 bought through a sales rep

New Zealand

May 2026 bought through a sales rep

United Kingdom

6 June 2026 full self-serve

Japan

July 2026 ads in 18.8% of answers

South Korea

July 2026 expansion phase

Brazil, Mexico

announced named as next

France, Germany, Ireland

in preparation first EU markets

Poland

– not on any list
live announced or in preparation no declaration of any kind
Seven markets live, two in the queue, the whole EU without a date.

Scale, money and user pushback

The numbers explain why OpenAI has no intention of backing out. The pilot passed $100 million in annualised revenue within ten weeks, with more than six hundred advertisers. The target for the whole of 2026 is $2.5 billion. For a company that recorded $13 billion in revenue in 2025 and at the same time forecasts multi-billion losses, advertising is not an add-on. It is the plan for profitability.

Saturation is growing just as fast. At the start of July 2026 ads appeared alongside 51% of answers in the United States. In Japan, fresh after launch, the figure was 18.8%. This shows the shape of the curve: a new market starts in the teens of per cent and steadily gets denser.

The other side of the coin can be counted too. The boycott campaign run under the name QuitGPT collected more than 200,000 sign-ups, and in surveys close to two thirds of American adults say that the presence of ads lowers their trust in the answers. This is a risk that interface design cannot solve: an assistant that advises and sells at the same time has a built-in conflict of interest.

A search engine shows a list and tells you to choose. An assistant chooses for you. An ad carries far more weight in the second model than in the first, and users sense it.

When ChatGPT ads will reach Poland

The short answer: nobody knows, and certainly not in the coming weeks. The longer answer is more interesting, because it lets you estimate the date yourself instead of waiting for an announcement.

Let us start with the fact that gets lost most often in Polish coverage. The United Kingdom was indeed the first European market, but the United Kingdom is not in the European Union. Neither the DSA nor the AI Act applies there, and the UK counterpart of the GDPR is more lenient on several important points. So launching ads in London says nothing about readiness to launch in Warsaw. Those were two different legal problems.

The second signal is the way ads are sold. The self-serve campaign manager, the tool that lets you launch an ad without contacting an OpenAI representative, works only in the US and the United Kingdom. The remaining markets are served through intermediaries such as Criteo or StackAdapt, on test budgets. The moment the dashboard opens to another country is the best available indicator of real readiness, better than any announcement.

The third signal is the queue. France, Germany and Ireland are named as being in preparation: three markets which, together with the United Kingdom, account for most of Europe's advertising revenue. Poland does not appear in any of these lists. Historically, large advertising platforms have entered Poland in the second or third EU wave, and nothing suggests that it will be different this time.

Three blockers to defuse before launch

Instead of guessing, it is worth looking at what OpenAI has to have in order before it shows the first ad to a Polish user.

What is blocking the launch in the European Union

GDPR

In the EU, ad personalisation requires explicit consent given in advance. OpenAI has declared consent as the legal basis, not legitimate interest, and that consent can be withdrawn at any time.

Effect for the advertiser A significant share of users will see contextual ads only, which lowers the effectiveness and predictability of campaigns.

DSA

ChatGPT search has around 120 million monthly users in the EU against a threshold of 45 million. Designation as a very large online search engine is a formality, and reports point to an announcement possibly as early as August 2026.

Effect for the advertiser A public ad repository, transparency reports, a ban on profiling based on sensitive data and fines of up to 6% of global turnover.

AI Act

Transparency obligations for generative systems overlap with advertising requirements. There is also an ongoing dispute about how far the labelling of AI-generated content should reach in commercial communication.

Effect for the advertiser An extra layer of labelling and documentation that has to be built before the launch, not after it.

Three independent legal regimes, each of which has to be handled separately.

The DSA thread is the most interesting. ChatGPT search reports around 120 million monthly users in the EU, against a threshold set at 45 million. Designation as a very large online search engine is therefore a formality, and reports point to an announcement possibly as early as August 2026. After such a decision come obligations that cannot be met in a week: a public ad repository, transparency reports, an assessment of systemic risks and a ban on ad profiling based on sensitive data, with fines of up to 6% of global turnover in the background.

Then there is the GDPR, in the version OpenAI itself has declared: the legal basis for personalisation is consent, not legitimate interest, and it can be withdrawn at any moment. Anyone who does not give consent will see contextual ads only, selected on the basis of the current conversation. The conversion pixel has been changed too: it gained a consent management mechanism and a country field. This is foundation work, under way since spring, and it shows clearly that entry into the EU is being prepared, only slowly.

And what about September?

A rumour is circulating in the industry that ads in ChatGPT could appear in Poland as early as September 2026. It is worth knowing, because it keeps coming up in conversations and on industry channels, and it is worth saying straight away how much it is worth. No OpenAI announcement stands behind it, no press report with a named source and no entry in the documentation. It is an extrapolation of the pace of expansion in the first half of the year, not an announcement.

Is it completely unrealistic? No. If the decision on DSA designation is made in August, and OpenAI decides that the consent layer built for the United Kingdom and the pixel with a country field are enough to launch, September is technically possible, though more likely for France, Germany and Ireland than for Poland. My cautious estimate, clearly marked as an estimate: the first EU markets at the turn of 2026 and 2027, Poland in the following wave, and the self-serve dashboard for Polish advertisers later still than the ads for users.

September is a rumour, not a date. The verifiable signal will be the moment the advertising dashboard opens to the first EU country. The real clock starts on that day.

ChatGPT versus Google AI Mode

The comparison is instructive, because both systems monetise an answer generated by a model, yet they do it in completely different ways, and on the Polish market they stand at opposite points.

Google chose continuity. Ads in AI Mode come from existing Search, Shopping and Performance Max campaigns, they enter the answers automatically and there is no way to opt out. A study by SE Ranking from July 2026 found an ad present for 29.45% of commercial queries. Advertisers in Poland have been there for a long time, often without being fully aware of it. I described this in detail in my analysis of ads in AI Mode and AI Max.

OpenAI built its system from scratch, with a separate dashboard, its own auction and a logic based on conversation context instead of keywords. That gives it full control over the rollout, and it is exactly why it can afford to hold back the launch in the EU. Google did not have that luxury, because its ads entered AI Mode through a channel that had been operating in Europe for years.

Click-through rate compared with other channels

Google Search ads 4–6%
ChatGPT, top brands 1.57%
ChatGPT, average 0.68%
Classic display 0.35%
Similarweb data. Peak results of individual campaigns reached 5.4%, but the market median is low.

That 0.68% looks weak next to search, and it should be said plainly instead of being explained away by "the different nature of the channel". At the same time it is worth remembering what this format really competes with: classic display hovers around 0.35%, and podcast advertising falls in the 0.5–1% range. An ad in ChatGPT is closer to that shelf than to ads in search results, and that is how its role in the mix, and the expectations placed on it, should be planned.

E-commerce: Instant Checkout and ACP

For online shops, advertising is only half of the story. The other half is called agentic commerce, and it is in far worse shape than the announcements suggested.

Instant Checkout, a purchase made without leaving the conversation, was launched by OpenAI together with Stripe. Along with it came an open standard, the Agentic Commerce Protocol, released under the Apache 2.0 licence. The launch covered sellers from Etsy, with a million Shopify shops announced to follow. It sounded like the beginning of a new sales channel.

Reality turned out to be more prosaic. In February 2026 Instant Checkout was actually being used by around thirty Shopify sellers. There was no solution for US sales tax, no support for multi-product baskets and no integration of loyalty programmes, and product data could be displayed incorrectly. On 4 March 2026 OpenAI changed course towards merchant-controlled checkout, a model in which the transaction is completed on the shop's side after all.

Two things follow from this for Polish e-commerce. First: there is no point rebuilding your checkout for a standard that its own author has just scaled back. Second, and more important: every problem that tripped this project up comes down to the quality of product data: complete attributes, correct prices, availability and unambiguous identifiers. That is work that pays off whatever happens to any protocol, because the same data feeds shopping results, ad feeds and model answers.

What this changes for brand visibility

The question I hear most often on this topic is: if there are going to be ads in ChatGPT, will buying them make the model start recommending my business?

No. These are two separate mechanisms. An ad buys a labelled place below the answer. A brand being named inside the answer depends on what the model finds in its sources, and no budget speeds that up. For Google AI Mode we have hard data on this: only 11.53% of advertisers' domains also appeared among the cited sources for the same queries. There is no reason to think ChatGPT is any different. The architecture of both systems separates these layers on purpose.

So the practical conclusion is that the arrival of ads takes nothing off the to-do list on the organic side. What still counts is content structured so that it can be cited, which I wrote about when covering the BLUF method, consistent data about the business, and a presence in the places models draw their knowledge from. How this works on Google's side is set out in my guide to citations in AI Overviews.

There is one more thread that is easy to forget: for a model to cite you, it has to have access to your content. The decision to let AI bots in has stopped being a technical detail and has become a business decision. I described the details and the pitfalls when writing about Cloudflare blocking AI crawlers by default.

What to do before the launch in Poland

Waiting with folded arms would be a mistake, but so would preparing campaigns for a system that has no launch date. Below are the things that make sense whether ads start in September or a year from now:

  1. 01 Start measuring traffic from ChatGPT separately. In Analytics, split out the source chatgpt.com and the other assistants. Without a baseline from the period before ads you will not be able to judge later what changed. It takes a quarter of an hour and is the most urgent item on this list.
  2. 02 Put your product and service data in order. Complete attributes, current prices, unambiguous identifiers, descriptions that answer specific questions. This data feeds ad feeds, shopping results and model answers at the same time. No other work has such a broad return today.
  3. 03 Prepare answers to questions, not advertising slogans. The conversational format rewards those who have something to answer a follow-up question about a detail with. Collect real customer questions from your support team and check whether the answers exist on the site at all.
  4. 04 Make a conscious decision about AI bot access. Blocking protects the content but excludes you from citations. It is a business decision that is better made on purpose than inherited from a provider's default settings.
  5. 05 Describe scenarios instead of keyword lists. A fifteen-minute exercise: write down five situations in which your customer talks to an assistant about the problem you solve. This will be the starting material for contextual targeting, and a good content brief along the way.
  6. 06 Decide who keeps an eye on the topic. The launch signal will not be a press release but the dashboard opening to the first EU market. It is worth having one person who checks this once a month, instead of finding out from LinkedIn three weeks after the fact.
  7. 07 Do not move budget in advance. With a click-through rate of around 0.68% and no data from the Polish market, any planning of a budget share is guesswork today. The channel will have to be tested with an amount you can lose without pain.
Every item on this list pays off even if ads in Poland do not start for another year. It is the only sensible way to prepare for something that has no date.

FAQ: common questions

Are there ads in ChatGPT in Poland yet?

No. As of 1 August 2026, ads in ChatGPT run in seven countries: the United States, Canada, Australia, New Zealand, the United Kingdom, Japan and South Korea. Poland is not on any of the announced lists, and OpenAI has not given a date for entering the Polish market or the EU market as a whole. A ChatGPT user in Poland simply will not see ads, whatever plan they are on.

When will ads in ChatGPT appear in Poland?

There is no official date. A rumour is circulating in the industry that they could start as early as September 2026, but no OpenAI announcement stands behind it. It is a prediction based on the pace of expansion so far. The hard evidence says something else: the self-serve campaign manager works in only two countries, France, Germany and Ireland are next in line for the EU, and over all of it hangs the expected designation of ChatGPT search as a very large online search engine under the DSA. My cautious estimate is the turn of 2026 and 2027 for the first EU markets, with Poland more likely in the second wave.

Who sees ads in ChatGPT?

Logged-in adult users on the free plan and on the Go plan at $8 a month. Plus, Pro, Business, Enterprise and Edu accounts do not show ads. This distinction matters when you estimate reach: those who pay the most are outside the ad system, so advertising reaches the broadest part of the user base, but not necessarily the wealthiest.

Do ads affect the content of ChatGPT answers?

OpenAI states that they do not. Ads appear as a clearly labelled block below the model's answer and do not change its content. The advertiser does not buy a recommendation or a mention in the answer, only a place next to it. It is worth checking this yourself once ads launch in Poland, because the line between a recommendation and a paid placement is thinner in a conversational interface than in a classic search engine.

How much does an ad in ChatGPT cost?

At launch in February 2026 the default maximum bid was $60 per thousand impressions, and the entry threshold was as high as $250,000. Within ten weeks rates slid to around $25, the threshold fell to $50,000, and OpenAI added cost-per-click billing in the $3–5 range. Since the self-serve campaign manager opened there is no minimum budget at all: you can start a campaign with any amount. The auction is second-price, weighted by relevance.

How do I turn off ads in ChatGPT?

There are two ways. The first is to move to a paid plan above Go: Plus, Pro, Business, Enterprise and Edu show no ads at all. The second concerns personalisation only: in the European Union the legal basis is consent, which can be withdrawn at any time in the settings. Withdrawing consent does not remove ads completely, but it reduces them to the contextual variant, selected only on the basis of the current conversation, without reaching into history and memory.

Will an ad in ChatGPT make the model start recommending my business?

No. These are two independent mechanisms. An ad buys a labelled place below the answer, while a brand being named inside the answer itself depends on what the model finds in its sources: the structure of the content, credibility, consistent data about the business and a presence in the places models draw their knowledge from. A comparable study for Google AI Mode showed that only 11.53% of advertisers' domains also appeared among the cited sources for the same queries.

Can a Polish company already advertise in ChatGPT abroad?

Technically yes, if it sells in a market where the system is live, but with limitations. The self-serve campaign manager serves only advertisers from the US and the UK, so entering from Poland requires an entity in that market or buying through an intermediary. For a company that sells only in Poland this is a theoretical exercise for now: with no Polish users in the system there is nobody to address the message to.

Worth remembering

  • Ads are shown only to users on the Free plan and the $8 Go plan. Plus, Pro, Business, Enterprise and Edu stay ad-free.
  • At the start of July 2026 ads appeared in 51% of answers in the US. The experiment has become a default part of the product.
  • Rates fell from a $60 CPM at launch to around $25 within ten weeks, and OpenAI moved to cost-per-click billing in the $3–5 range.
  • The average CTR of ChatGPT ads is 0.68% according to Similarweb data: many times lower than in Google Search, but twice as high as in classic display.
  • Poland is not on any announced list of markets. Self-serve works only in the US and the UK; the other markets buy through intermediaries.
  • France, Germany and Ireland are next in line for the EU. Unofficially people talk about September 2026, but OpenAI has confirmed nothing.
  • In the EU the legal basis for personalisation is consent, not legitimate interest. Without consent the user gets contextual ads only.
  • ChatGPT search has around 120 million monthly users in the EU against a threshold of 45 million, so it faces the strictest DSA regime, with fines of up to 6% of global turnover.
  • An ad does not buy a citation. Being present in answers as a source is a separate mechanism and a separate line of work.

Summary

Ads in ChatGPT stopped being an announcement in February 2026 and in half a year grew into a fully fledged advertising system: with a dashboard, an auction, two billing models and a presence in half of all answers on the American market. The direction is settled, because behind it stands the profitability plan of a company that needs billions in revenue.

Poland is last in the queue in this puzzle, and it is well known why. The United Kingdom turned out to be simple, because it had left the EU. France, Germany and Ireland are next, because the largest budgets stand behind them. Poland will come later: after the GDPR, the DSA and the AI Act have been defused, and after the dashboard has opened to smaller markets.

There is no opportunity to miss here. There is, however, a fair amount of work that can be done ahead of time, and that will be useful even if ads do not start in Poland for a long while yet.

The most sensible stance for today is calm observation with one verifiable reference point: the day the self-serve advertising dashboard opens to the first country in the European Union. Until then, tidy data, measurement of traffic from assistants and content the model has something to cite from do more for visibility than any budget, which there is nowhere to spend anyway.

Want to know how your brand looks in AI answers today?

I will check whether the models see you at all, where they take information about your business from and what to improve before ads take the place below the answer in your market too.

Book a consultation

How this text was written

I will not pretend it was written entirely by hand, by candlelight. For the research, the proofreading and the code of this page I used a language model. It is a good tool for digging through documentation and catching the typos you no longer see on the third read of your own text. The English version was also prepared with a model, from my Polish original.

What the model did not do: it did not choose the topic, form the arguments or decide what this data means for your site. That stays with me, as does the last word on what goes into the text and what comes out of it.

Where the model helped

  • digging through sources and organising data
  • proofreading and the English version of the text
  • code, page layout and infographics

What stays with me

  • the choice of topic and what is worth saying about it
  • the arguments, the reading of the data and the conclusions
  • fact-checking and the decision on what goes into the text
Sebastian Baryś – freelance SEO specialist

Written by

Sebastian Baryś

Freelance SEO specialist · Częstochowa, Poland, and remote

I started when nobody had dreamt of Core Web Vitals and rankings were checked by hand. Today I work on making companies visible not only in Google, but wherever people ask questions. I write only about things I check myself in client projects.

SEOGoogle AdsTechnical SEOSEO in the AI era

Let's check your visibility in AI

Before ads take the place below the answer in your market too, it is worth knowing what the models say about your business today and where they get it from.

Get in touch