Table of contents
Introduction
English version of an article first published in Polish on 10 September 2026. Read the Polish original.
Ads in ChatGPT have been running in Poland since 24 August 2026. On that day OpenAI launched them in 31 European countries at once: the whole EU plus Norway, Iceland, Liechtenstein and Switzerland. Since 31 August a Polish company can also open its own Ads Manager account and launch a campaign without an intermediary.
On 1 August I wrote that Poland was at the back of the queue and that a realistic date was the turn of 2026 and 2027. Three weeks later the ads were already running. I go through that forecast in a separate section, point by point. The August article stays online unchanged, so that anyone can check exactly what I wrote in it.
For an advertiser, though, two other things matter more. The first is the ad policy. For now OpenAI allows a narrow set of categories, and three large industries (finance, health and legal services) are in practice shut out everywhere outside the United States. The second is sign-up. You choose several settings once and for all when you open the account, and a mistake means opening a new account from scratch.
I take it in parts: the launch timeline, what an ad looks like in ChatGPT in Poland, the full list of allowed and prohibited categories mapped onto Polish industries, sign-up, the first campaign with a budget in zloty, and finally the DSA and the reckoning with my forecast. All facts about rules and process come from OpenAI’s documentation as it stood at the turn of August and September 2026.
What you will learn:
- What exactly happened on 18, 24 and 31 August and why Poland came in the first wave, not the second
- What an ad looks like in ChatGPT in Poland and why ads in Europe are not personalised at launch
- Which categories OpenAI allows, which it bans and why finance, health and legal services are in practice ruled out in Poland
- Where your industry stands under the ad policy: from a clothing shop to a dental practice
- How sign-up in Ads Manager works and which three settings you cannot change once the account exists
- What a campaign costs at minimum in Polish zloty and why the daily budget can double
- What the European Commission’s decision to designate ChatGPT as a very large search engine will change
- Which of my August predictions held up and where I was clearly wrong
Three dates that made up the launch
The launch was spread over two weeks, and each of the dates means something different. Press coverage often blurs them together, so it is worth separating them.
From announcement to self-serve buying
-
18 August 2026
Announcement on the OpenAI blog
31 European markets from the following week, buying first through the sales team and partners
-
24 August 2026
Ads visible to users
Free and Go plans across the EEA and in Switzerland, including Poland
-
31 August 2026
Self-serve buying in Ads Manager
Europe, India, the Middle East and North Africa on the same day
-
31 August 2026
ChatGPT designated a very large online search engine under the DSA
European Commission decision, announced the same day
-
January 2027
Deadline for the DSA obligations
four months from notification of the decision
In the first week ads could be bought only through the OpenAI sales team, partner agencies and technology partners. For a small Polish company that meant the ads were already being shown, but there was no way to buy them. That lasted seven days. On 31 August OpenAI added an update to its announcement: self-serve access through Ads Manager is available in all 31 markets.
Two other pieces of news came on the same day. OpenAI announced that, less than 200 days after launch, its advertising business had reached an annualised revenue run rate of one billion dollars, and that tens of thousands of advertisers were using ads. The European Commission, for its part, designated ChatGPT as a very large online search engine under the Digital Services Act. I come back to the DSA below, because it is the only date in this timeline that is still to come.
Why 31 countries exactly
That is the 27 EU member states, the three other countries of the European Economic Area (Norway, Iceland and Liechtenstein) and Switzerland. The United Kingdom had had ads since June. Before that the system ran in nine countries: the US, Canada, Australia, New Zealand, the United Kingdom, Japan, South Korea, Brazil and Mexico. On 10 September the list of countries from which a company can buy ads on a self-serve basis had 52 entries.
What an ad looks like in ChatGPT in Poland
The ad appears below the answer, as a block labelled as sponsored and visually separated from the model’s text. It consists of the advertiser’s name, its icon, a headline, a short description, an image and a link to the landing page. One block can contain several items, from one advertiser or from several.
Who will see an ad in Poland
Will see it
- an adult user on the Free plan
- a user on the paid Go plan
- someone who is not logged in, but only ads suitable for all audiences
Will not see it
- a Plus, Pro, Business, Enterprise or Edu subscriber
- an account believed to belong to someone under 18
- anyone in a temporary chat or in the Atlas browser
- anyone on the Free plan who chose the ad-free version with lower limits
The scale is considerable, though hard to pin down exactly. According to Mediapanel (the Polish internet audience measurement study), 10.2 million internet users in Poland used ChatGPT in October 2025, a reach of 34.42%. Only some of them will see ads, because subscribers on the higher plans are outside the system. OpenAI does not say how many Free and Go users there are in Poland.
No personalisation in Europe
This is the most important difference from the American market. OpenAI’s help centre states plainly that personalised ads are not initially available in the European Economic Area or Switzerland. A user in Poland gets ads matched to the topic of the current conversation and to general information such as approximate location and language. The system does not draw on earlier chats, memory or the history of interactions with ads.
For an advertiser this means that in Poland the match is decided by what happens in the conversation and by what you supply yourself: context hints, ad copy, the landing page, the country and, optionally, your own customer lists. In my view this is a set-up in which a small company loses less than it might seem. A large advertiser has no advantage here in the form of the user’s behavioural history, because at launch the system in Europe simply does not use that history. What counts is how precisely you describe the situations in which your offer helps.
The user has a good deal of control over the ad. They can hide it, report it, check why they are seeing it, or choose the “Ask ChatGPT” option and ask the model about the advertised product. OpenAI states that the advertiser has no influence over answers about an ad shared in the chat. If someone writes to the advertiser directly through the ad, the advertiser sees only those messages, not the rest of the conversation. From reports the advertiser gets aggregate data only: impressions, clicks and conversions.
What you can advertise
The rules are set out in the Ad policies document, updated on 31 August 2026 to version 1.5. It has three layers: where an ad may appear, what may be advertised and who may be an advertiser at all. I will start with the second, because it decides whether opening an account is worth it in the first place.
The most important sentence of the whole policy is easy to miss. OpenAI lists four groups of allowed categories and then, straight afterwards, writes that all other categories are not permitted at launch. It is an allowlist, not a blocklist. If your industry is not in any of the four groups, the default answer is “no”, even if you do not find it on the list of bans.
OpenAI’s ad policy in three buckets
Allowed
Four groups of consumer categories that OpenAI is focusing on in the early phase.
- household items and consumer goods
- local services
- travel and entertainment
- digital products and education
Restricted, US only
Approved advertisers, individual decision, manual review, sometimes proof of licence. Generally prohibited outside the US.
- financial services
- health and medicine
- legal services
Prohibited
Everything outside the allowlist, with these named explicitly, among others:
- adult content and dating
- alcohol and tobacco
- gambling
- politics and contested issues
- intoxicating substances
- counterfeits and scams
Restricted categories are for the US only
Financial services, health and legal services are restricted categories. In the United States OpenAI may admit approved advertisers: after an individual decision, with manual review and sometimes with a requirement to show a licence. The list is specific. In finance it covers, among others, car loans, credit cards, accounts, insurance, mortgages and payment services. In health it covers dentistry, hospitals, supplements, diagnostic tests and minimally invasive cosmetic procedures.
Outside the US the situation is simple. The policy says that ads for financial and health services are generally prohibited there, and ads for legal services currently prohibited. Legal services were added to the list of US exceptions only in version 1.5, so the catalogue is moving. For now, though, a Polish bank, dental practice or law firm cannot buy ads in ChatGPT.
Rules for every ad, whatever the industry
Even in an allowed category an ad can be rejected. OpenAI lists four baseline standards:
- Truthfulness. No unsubstantiated promises about what the product can do, its price, its results, affiliations with other companies or comparisons with competitors.
- Language. Profanity and shocking wording can block an ad, even when they sit in the name of a product or an event.
- No discrimination. No content that is offensive or exclusionary on the basis of protected characteristics.
- No imitating ChatGPT. The creative must not mimic the look, behaviour or voice of ChatGPT in a way that would make the user take it for part of the product.
On top of that comes a consistency rule. The standards cover the text, image, video and landing page together, and an approved ad may not lead to a page with content that is not permitted. An example from the policy itself: an ad for food delivery that leads to alcohol delivery will not pass. OpenAI also reserves the right to reject an ad or a link for any reason.
What you cannot advertise
Below is the full list of categories the policy mentions by name, together with their exceptions. The exceptions are more interesting than the bans, because they show where OpenAI draws the line.
Prohibited categories and their exceptions
Adult content
Banned: Dating apps, sexual services and products, sexual health products, adult events, nudity.
Exception: Underwear and swimwear shown in an ordinary fashion context.
Alcohol and tobacco
Banned: Drinks above 0.5% alcohol, cigarettes, e-cigarettes, nicotine products and events themed around alcohol.
Exception: An incidental mention, as long as alcohol is not the subject of the ad.
Gambling
Banned: Casinos, sports betting, lotteries, poker and games with cash prizes.
Exception: A stay at a casino hotel, if gambling is not the subject. Games without real stakes.
Financial services
Banned: Outside the US, generally all of them. Nowhere: credit repair, debt relief, alternative investments, for example in precious metals.
Exception: Budgeting apps, financial software and courses that do not offer financial services.
Health services
Banned: Outside the US, generally all of them. Nowhere: unapproved products, experimental therapies, invasive procedures.
Exception: Fitness equipment, wearables, menstrual products, general wellness without medical claims.
Health claims
Banned: Weight-loss pills, detox programmes, supplements with unproven health promises, health coaching.
Exception: Supplements and nutrition products widely sold by major retail chains, without medical claims.
Legal services
Banned: Outside the US, currently all of them: advice, representation, immigration and compensation cases, document preparation.
Exception: Legal education and media about law, if they do not offer services.
Housing and jobs
Banned: Specific listings of property for sale or rent and specific job offers.
Exception: A listings site, if the ad and the landing page do not point to a specific listing.
Politics and contested issues
Banned: Elections, parties, referendums, encouraging or discouraging voting, advocacy around government action, contested social issues.
Exception: None.
Intoxicating substances
Banned: Marijuana and THC products, psychedelics, paraphernalia.
Exception: Hemp products without THC, for example CBD creams and hemp clothing.
Counterfeits and scams
Banned: Replicas of branded products, impersonating companies and people, guaranteed returns, phishing, suspicious messaging apps.
Exception: None.
Violence and graphic content
Banned: Graphic violence, sexually explicit content, tools for generating undressed images.
Exception: Age-rated films and games, if the creative and the landing page are not graphic.
Conversations where no ad will appear
A separate matter is which conversations an ad can appear next to at all. The policy excludes two groups of contexts. The first is harmful or contested topics: child safety, self-harm, violence, hate speech, disinformation, politics, terrorism, weapons, regulated goods and a dozen or so others. The second is conversations in which the user is in a vulnerable position: conversations about mental and personal health, situations of emotional reliance on the model and other sensitive paths.
In practice this means that even a fully permitted ad will not reach every matching conversation. A mattress shop may want to appear to someone writing about trouble sleeping. But if the same person describes insomnia in the context of depression, the conversation falls under mental health and, under the policy, there should be no ad there.
One change from April works in the opposite direction. In version 1.1 OpenAI stopped automatically blocking ads next to conversations about medical, legal and financial advice. A conversation about a household budget can therefore get an ad for a spending planner app, even though the bank itself cannot advertise there.
Vetting looks at the whole company, not just the ad
The policy for advertisers says that a company whose core business model falls into a restricted or prohibited category may be found ineligible to advertise at all. The examples given in the document are healthcare and weapons. So a clinic should not expect to get round the ban by advertising its own blog or an e-book about healthy eating.
Your industry: yes, no or it depends
The policy is written from an American perspective, so I map it onto typical Polish industries. This is my reading of the policy text as it stood on 31 August 2026, not a decision by OpenAI. The last word always belongs to account vetting and ad review.
Can your industry advertise in ChatGPT in Poland
Shop selling clothing, footwear, home furnishings
consumer goods and household items
Local services: renovation, installation, cleaning, garage
local services are one of the four launch categories
Hotel, guesthouse, travel agency, amusement park
travel and entertainment
Online course, language school, SaaS, app
digital products and education
Gym, fitness equipment, sportswear
allowed as long as the ad does not promise medical effects
Budgeting app, personal finance course
a financial tool that does not offer a financial product
Dietary supplements
a restricted category in the US; outside the US, products from major chains without health claims stand a chance
CBD cosmetics
no THC and no promises of an intoxicating effect
Job or property listings portal
yes, if the ad does not promote a specific listing
Estate agency with a specific listing
specific sale and rental listings are prohibited
Dental practice, clinic, aesthetic medicine
health services are generally prohibited outside the US
Law firm, legal advice
legal services are currently prohibited outside the US
Bank, credit, loans, insurance, investments
financial services are generally prohibited outside the US
Off-licence, e-cigarettes
alcohol and tobacco are prohibited
Bookmaker, online casino
gambling is prohibited
Dating app
falls under adult content
Mixed businesses raise the most doubts. A grocery shop with an alcohol section, a chemist’s selling supplements, a hotel with a casino. The policy does not rule them out in advance, but the consistency rule is merciless here. An ad may lead only to a page with no content that is not permitted. A home page with a banner for a wine promotion is out, the dairy category will pass.
The second trap concerns promises. Fitness equipment is allowed as long as the ad does not claim it cures back pain. A supplement from a large chemist chain may pass as long as it does not promise immunity. In both cases the category is decided not by the product but by what you write about it in the ad and on the landing page.
Sign-up step by step
There is one precondition: the entity that will advertise and pay must be based in a country marked as available in the documentation. Poland has been on that list since 31 August. Access depends on the country chosen in the ad account, not on the country assigned to your ChatGPT account. OpenAI divides the process itself into five steps.
From an OpenAI account to the first campaign
- 01
OpenAI account and access to Ads Manager
The panel lives at ads.openai.com. You log in with an OpenAI account, ideally one registered on a company email. The company’s ad account is created by one person, who becomes its owner.
- 02
Company details and account settings
Name, website, logo and industry, plus country, currency and time zone. You cannot change those last three fields once the account has been created.
- 03
Verification and review
A few verification questions in the Persona service, then a review on OpenAI’s side, which also checks whether your offer fits the policy. The queue is long and there is no way to speed it up.
- 04
Completing the account and payments
After the access email you set the name and logo shown in the ad, create a billing profile and attach a card. Without the name and logo, ads will not be shown.
- 05
Inviting the team
Settings, then the users section and the invite option. This is where the agency or freelancer who will run the campaigns joins the account.
Step three can take a while. OpenAI writes that there are many applications, that the queue is processed on a rolling basis, and that neither account review nor ad review can be expedited at any stage. It also advises against submitting a second application while the first is pending, because a duplicate speeds nothing up and can cause confusion during verification. Completing identity verification does not yet mean the account can serve ads. Those are two separate stages.
There is also a second route: the ChatGPT Ads Manager app, available directly inside ChatGPT. It lets you start opening an account in a conversation and run campaigns there. Payments, tax details and identity verification still have to be completed afterwards in Ads Manager itself.
Three settings you cannot undo
Country, billing currency and time zone are fixed for good in an ad account. The documentation is explicit: if any of these values is wrong, you have to open a new account. They affect billing, reports and whether the account has access to the panel at all.
Poland
the country where the paying company is based
PLN
if you want to pay and report in zloty
Warsaw
time zone for reports and budgets
Payments: in arrears and by card
Billing works in arrears. First the ads are shown, then the card is charged. Every new account gets a payment threshold. When unpaid spend reaches it, the card is charged. The documentation gives the example of a 25-dollar threshold. If a balance below the threshold remains at the end of the month, the card is charged anyway. The threshold rises by itself over time as payments go through, and the support team does not change it.
When you attach a card, your bank may show a temporary hold of 50 or 100 dollars, converted into the card’s currency. This is not a charge and it disappears after verification, although it can remain visible for up to seven days. The card and billing address should match the country of the company the account was opened for. A failed payment stops delivery until the card is replaced. You enter the tax number in the billing settings and it applies only to invoices issued after it was added, so it is better to do this before the first campaign.
Agency, freelancer and the client’s account
Every advertiser needs a separate account, and one person can later be invited to many accounts and switch between them. For accounts opened from the app inside ChatGPT the documentation is clear: accounts for private individuals and accounts opened by an agency on behalf of a client are not supported, and the agency is invited after the client has opened the account.
Even where something else is technically possible, the arrangement I recommend is the same. The client opens the account, for their own company and on their own card, and then invites the agency or freelancer. The company details, currency and invoices match the paying entity from the start, and when the cooperation ends nobody has to transfer anything. Neither the country nor the currency can be changed later, so a mistake made by an intermediary stays in the account for good.
First campaign: structure and budget
The structure resembles what you know from Google Ads, with one important difference: an ad group has no keywords. There are three levels.
level 1
Campaign
- objective and billing model
- budget and dates
- countries and platforms
- custom audience lists
level 2
Ad group
- one topic or one need
- context hints
- maximum bid
- multipliers for audience lists
level 3
Ad
- headline and description
- square PNG or JPG image
- landing page with UTM
- several versions per group
Objective and billing model
There are three objectives to choose from. Reach, billed per thousand impressions (CPM); traffic, billed per click (CPC); and conversions, where the system looks for clicks that lead to a chosen event and you still pay per click (oCPC). In beta there is also a conversion variant billed per impression (oCPM). To optimise for conversions you need the OpenAI pixel or the Conversions API and at least one standard event. You cannot change the objective, the billing model or the conversion event once the campaign has been created. If you want something different, you create a new one.
You set the maximum bid at ad group level. For a click campaign OpenAI advises starting at 3–5 dollars per click, and the panel indicates whether the bid is competitive. The auction is a second-price model weighted by relevance, so a high bid alone is not enough. According to OpenAI, click and conversion campaigns already account for the majority of all campaigns in the system.
Budget: a 65 PLN minimum, but with multipliers
The minimum daily budget depends on the account currency. For zloty it is 65 PLN, for euro 15, for dollars 25. The trap is in the definition. The daily budget is a seven-day average, not a hard cap per day. On a single day a campaign can spend up to twice the daily budget, and over seven days up to seven times the daily budget. An amount left unspent on one day can be spent on the following days.
65 PLN
minimum daily budget for an account in zloty
130 PLN
what a campaign can spend on a single day
455 PLN
maximum spend over seven days
The other option is a total budget for the lifetime of the campaign, which the documentation describes as tighter control over spend. For new advertisers, however, OpenAI recommends a daily budget, because it makes delivery easier to watch. Choose the budget type on the assumption that it is permanent. The documentation is not consistent here: the billing article says the budget type cannot be changed, while the campaigns article says you can switch from total to daily but not back again. In both versions it is practically a one-off decision.
Pausing does not take effect immediately either. In some cases ads continue to be shown for up to 24 hours after a campaign is paused, and costs from that period are charged as normal. Data appears in reports with a delay of up to seven hours, and OpenAI asks advertisers to report delivery problems no sooner than a day after launch.
Targeting: context hints instead of keywords
The heart of an ad group is its context hints. In them you describe what the product offers, whom it helps and in what situation it is useful. The documentation gives a good example: rather than naming running shoes, it is better to describe cushioned everyday shoes for beginners training for their first 5 km run. A Polish equivalent would be, for example: instead of “kurs angielskiego” (an English course), “dorosły, który za pół roku wyjeżdża do pracy w Irlandii i chce szybko podciągnąć angielski konwersacyjny” (an adult who is moving to Ireland for work in six months and wants to improve their conversational English quickly).
OpenAI states that hints do not work like exact match or like targeting rules. They help the system understand where an ad makes sense, but they guarantee nothing. Alongside them you have the country, platforms (iOS app, Android app, web browser) and your own audience lists. Detailed targeting by state, city and postcode is described in the documentation for the US. For other countries the availability of locations varies, and you can check it in the location search in the panel or in a CSV file provided by OpenAI.
Audience lists: a threshold of 25,000
You upload your own lists as email addresses, phone numbers, hashed versions of either, or Google advertising IDs. To target a campaign only at such a list, or to raise the bid for it, the list must match at least 25,000 users, and OpenAI recommends 100,000. A smaller list can be used for exclusions only. With a typical Polish customer base of a few thousand addresses, that leaves in practice one use: excluding existing customers from a campaign that is meant to win new ones.
Ads and the product feed
OpenAI advises building several ad versions in each group, each with a different argument, rather than repeating the same message in other words. The headline should say straight away what the product delivers. The description adds new information rather than repeating the headline. The landing page should be a specific product or category page, not the home page. The image must be square, in PNG or JPG format, no larger than 1200 by 1200 pixels and publicly available at a direct URL. The panel can also suggest a draft ad based on your site’s metadata. According to the documentation this is not AI-generated text or imagery, only data pulled from the page.
Shops have an additional tool: campaigns from a product feed. You upload the feed in the tools tab as a CSV or TXT file, as a URL or over SFTP, and then build ad groups from product filters and a single ad template. OpenAI writes that feed-based ads are among the best performing in the whole programme. Two details matter in practice. Products expire after two weeks, so those from a manually uploaded file will drop out of the campaign after that time. That is why it is better to connect a URL or SFTP. And, at least in beta, products from the feed go only into ads, not into regular ChatGPT answers.
The landing page must let the bot in
OpenAI checks every ad together with its landing page. It does this automatically, with the OAI-AdsBot crawler, which visits the page, assesses its compliance with the policy and uses its content when matching the ad to conversations. OAI-AdsBot is required. OAI-SearchBot is recommended. A page the system cannot read cannot be advertised.
# Example from OpenAI’s documentation for advertisers
User-agent: OAI-SearchBot
Allow: /
User-agent: OAI-AdsBot
Allow: / The robots.txt file is only the first of three layers OpenAI tells you to check. The second is the firewall and bot protection, for example in Cloudflare or Akamai, which can answer crawlers with a 403 error. The third is human verification on the page itself: CAPTCHA, JavaScript challenges, behavioural analysis. Any of these layers can block OAI-AdsBot even if the previous one let it through. With bulk ad uploads there are also rate limits, which end in a 429 error. In that case OpenAI advises uploading ads in smaller batches.
Good news for sites behind Cloudflare: OAI-AdsBot is an officially verified and allowed bot there. But if at some point, in a reflex to protect your content, you blocked all OpenAI crawlers with a single rule, the ad will be rejected. OpenAI does not offer a manual workaround. Once access is fixed, the ad has to be submitted for review again. You will find the crawler’s IP addresses in the file openai.com/adsbot.json, but the documentation advises against relying on them alone.
I have described the four OpenAI crawlers, and which of them is responsible for what, in detail in my piece on AI bots in robots.txt. It also has four decision scenarios for which crawlers to let in and which to block. If your site sits behind Cloudflare, also check the settings described in the piece on the default blocking of AI crawlers.
DSA: what changes by January 2027
On 31 August 2026 the European Commission designated ChatGPT as a very large online search engine under the Digital Services Act. On the same day Reddit and Roblox were given the status of very large online platforms. According to Search Engine Journal, ChatGPT is the first chatbot with this status. The decision rests on web search, not on the fact that ChatGPT is an AI assistant.
The threshold is 45 million monthly users in the EU. According to OpenAI figures cited by Search Engine Journal, ChatGPT search had an average of 159.1 million monthly active recipients in the six months to 31 March 2026, more than three times the threshold. In my August article I gave a figure of around 120 million. That number came from the earlier report, for the six months ending in September 2025.
The Commission gives OpenAI four months from notification, which means until January 2027. By that deadline come obligations that ordinary services do not have: assessing and mitigating systemic risks (including the impact on minors, mental health, fundamental rights and elections), an annual independent audit, data access for the Commission and vetted researchers, and a public ad repository. Supervision is exercised by the Commission in cooperation with the Irish regulator, Coimisiún na Meán. Infringements carry fines of up to 6% of global turnover.
For an advertiser, the repository matters most
Today there is no official ChatGPT ad library like the one Meta runs. Third-party tools are trying to fill the gap, such as the British Adthena and the Polish PromptEye. The repository required by the DSA is meant to show the content of the ad, the advertiser, the period it ran, the main targeting parameters and the reach. Once it is up and running, every competitor will see your ChatGPT ads in one place. It is worth writing them with that in mind from the start.
One thing needs stressing, because it relates to my forecast. OpenAI did not wait for the Commission’s decision. Ads launched a week before it, and OpenAI will meet the obligations within a deadline that runs after the launch. I would not be surprised if ad personalisation in Europe arrived only after January 2027, but since August I phrase such guesses more carefully. This is a guess, not information.
My forecast of 1 August: the reckoning
On 1 August I published a long analysis of ads in ChatGPT, with a section on when they would come to Poland. Below I set ten claims from that article against what actually happened. The quoted claims are shortened versions of my own sentences, and the full wording is in the original.
What I wrote on 1 August and what happened
“First EU markets at the turn of 2026 and 2027”
Ads launched across the whole EU on 24 August 2026.
“Poland in a later wave”
One wave for 31 countries. Poland came in on the same day as France and Germany.
“September is a rumour, not a date”
The rumour was closer to the truth than my assessment. The launch came a week before September.
“The launch signal will not be a press release but the panel opening”
First came the announcement of 18 August, then the ads, and the panel opened last.
“A self-serve panel for Polish companies later than the ads themselves”
The order is right, but the gap was seven days, not months.
“Designation under the DSA is a formality, an announcement is possible in August”
The European Commission announced the decision on 31 August.
“Without consent the user will see contextual ads only”
OpenAI went further: at launch in the EEA, ads are contextual for everyone.
“Ads are shown only on the Free and Go plans”
No change, in Europe too.
“Prepare product data, do not move budget in advance”
OpenAI counts product feed ads among the best performing in the programme, and there are still no performance benchmarks.
“There is no budget minimum, a campaign starts with any amount”
Today the panel requires a minimum daily budget: 65 PLN or 25 dollars.
The tally looks like a draw, but it is not one. I was right on the details: who sees ads, how OpenAI would approach consent to personalisation, when the Commission would deal with the DSA. I was wrong on what the reader most wanted to know, namely when ads would come to Poland. I wrote about the turn of 2026 and 2027 and about a second wave. Ads arrived three weeks after publication, in the first wave.
Where I went wrong
I assumed that regulation was a precondition for launch. The whole section on Poland rested on three blockers: the GDPR, the DSA and the AI Act. OpenAI treated them differently: as obligations to meet along the way, not obstacles to clear before launch. Ads launched a week before the Commission’s decision, and OpenAI will meet the DSA obligations within a deadline that runs after the launch.
I carried over the pattern of big platforms that entered the EU in waves. Hence the queue of France, Germany, Ireland, with Poland somewhere further back. OpenAI did the opposite: one wave for the whole European Economic Area, but with the most cautious settings, without personalisation. Limiting features replaced waiting. I did not consider that option at all.
I underestimated the revenue pressure, even though I described it myself. In the same article I cited the target of 2.5 billion dollars in ad revenue for 2026 and forecasts of multi-billion losses. A company with a target like that does not put off until next year a market where its search engine has more than a hundred million monthly users.
I picked the wrong indicator and could have noticed it earlier. I wrote that the launch signal would be the panel opening for the first EU country, not a press release. Yet in the data I cited myself the order was the reverse: in the United Kingdom ads had been running since 6 June, and the panel opened on 22 July. In Europe the same thing happened again, only faster.
There is nothing shameful about a forecast that missed. Quietly correcting it would be shameful. The August article stays online in its original form, with its date, and the reckoning stands here, next to the new facts.
Next time I make a forecast I will separate two things: what a company has to do to operate lawfully, and what it has to have ready in order to launch at all. There is usually far less of the latter.
Where to start
If your industry fits the allowed categories, this channel is worth testing. Just have no illusions about predictability. OpenAI itself writes that there are no performance benchmarks yet for industries or campaign types, and that in beta delivery and spend may fluctuate. Below is the order in which I suggest going about it:
1. Check your industry before you open an account
The policy works as an allowlist. If your offer does not fit any of the four groups, the default answer is “no”. Vetting also looks at the company’s core business model, so a clinic should not expect to get round the ban by advertising its own blog about healthy eating.
2. Let OAI-AdsBot into your landing pages
Check robots.txt, the firewall, anti-bot rules and CAPTCHA. Do it before the first ad, because an ad rejected for this reason usually has to be submitted for review again.
3. Open the account for the company that pays
Country: Poland. Currency: zloty, if you want to pay and report in zloty. Time zone: Warsaw. The card and billing address must match that company. You invite the agency later.
4. Enter your tax ID before the first invoice
The tax number (in Poland, the NIP) is added in the billing settings and applies only to future invoices. Agree with your accountant how to settle VAT on this service, because OpenAI does not give tax advice.
5. Start with a daily budget and a clicks objective
A daily budget is OpenAI’s own recommendation for new advertisers. Remember the multipliers: 65 PLN a day means up to 130 PLN on a single day. Move to conversion optimisation once the pixel has collected its first data.
6. Describe situations, not keywords
A few ad groups, each built around one customer need. In each, a few ad versions with different arguments, not the same text in five disguises. Each version leads to a specific subpage rather than the home page.
7. Tag the traffic with UTM parameters
Without them, clicks from ads may end up in your analytics lumped together with traffic from ChatGPT citations, and those are two completely different channels. OpenAI explicitly encourages adding parameters to destination URLs.
8. In a shop, connect the feed by URL, not as a file
Manually uploaded products expire after two weeks. A feed fetched from a URL or over SFTP refreshes itself and will not drop out of the campaign halfway through a test.
Finally, something I already wrote about in August and which is just as true after the launch. An ad in ChatGPT does not buy presence in the answer. OpenAI repeats that ads run in separate systems and do not influence what the model answers, and that products from the ad feed do not go into regular answers. Whether ChatGPT mentions your company as a source or a recommendation depends on content, on the consistency of data about the company and on presence in the places models draw their knowledge from. That is what I work on as part of my AI SEO service, and you will find a comparison with Google’s ads in AI answers in my piece on ads in AI Mode and AI Max.
FAQ: common questions
Since when have ads in ChatGPT been running in Poland?
Since 24 August 2026. On that day OpenAI launched ads in 31 European countries at once: the whole European Union plus Norway, Iceland, Liechtenstein and Switzerland. The announcement came on 18 August, and self-serve buying in Ads Manager started on 31 August. For the first week ads could only be bought through the OpenAI sales team and partners.
Who sees ads in ChatGPT in Poland?
Adult users on the Free and Go plans. People who are not logged in may also see an ad, but only one that is suitable for all audiences. There are no ads on Plus, Pro, Business, Enterprise and Edu accounts. OpenAI also does not show them on accounts it believes belong to people under 18, or in temporary chats. In the EEA and Switzerland ads are contextual only at launch, meaning they are matched to the current conversation without drawing on chat history or memory.
Can a Polish company launch a ChatGPT campaign on its own?
Yes, since 31 August 2026. You open the account at ads.openai.com by logging in with an OpenAI account. There is one condition: the company that advertises and pays must be based in a country marked as available in the documentation, and Poland is on that list. Once the company details are entered, the account goes through verification by the Persona service and a review on OpenAI’s side. Campaigns start only after the account is approved and a card is attached.
What is the minimum cost of a ChatGPT campaign?
For an account billed in Polish zloty the minimum daily budget is 65 PLN. That is a seven-day average, though, not a hard cap. On a single day a campaign can spend up to twice the daily budget, that is 130 PLN, and over seven days up to seven times the daily budget, that is 455 PLN. For a click campaign OpenAI advises starting with a maximum bid of 3–5 dollars per click.
Can a medical practice, a law firm or a bank advertise in ChatGPT in Poland?
Not at present. Financial, health and legal services are restricted categories in OpenAI’s policy. Exceptions are granted individually, after vetting, and only in the United States. Outside the US, ads for financial and health services are generally prohibited, and ads for legal services are currently prohibited. Tools that do not offer the service itself remain allowed, for example a household budgeting app or educational material about law.
Can an agency open an ad account on behalf of a client?
Every advertiser needs a separate account, and its country, currency and time zone are fixed for good. The card and billing details should match the company the account was opened for. The most sensible arrangement is therefore an account opened by the client, for the client’s company and on the client’s card, with the agency invited as a user. For accounts opened from the app inside ChatGPT, the documentation explicitly excludes agencies acting on behalf of a client.
Are ads in ChatGPT personalised in Poland?
Not at launch. OpenAI’s help centre states plainly that personalised ads are not initially available in the European Economic Area or Switzerland. An ad is matched to the topic of the current conversation and to general information such as approximate location and language. OpenAI has not given a date for when personalisation might arrive in Europe.
Will a ChatGPT campaign make the model start recommending my company?
No. OpenAI states that ads run in systems separated from the model and do not influence answers. Even products from a feed uploaded to Ads Manager go only into ads, not into regular ChatGPT answers. Brand presence in the answer itself is a separate mechanism that depends on what the model finds in its sources.
Can I look up competitors’ ads in ChatGPT?
At present there is no official ad library like the one Meta runs, and third-party tools are trying to fill the gap. That may change in 2027. Following the European Commission’s decision of 31 August 2026, OpenAI has until January 2027 to meet the obligations of a very large online search engine, and one of them is a public ad repository.
Worth remembering
- Ads in ChatGPT have been running in Poland since 24 August 2026, and self-serve buying in Ads Manager since 31 August. Poland came in the first wave, together with the whole European Economic Area and Switzerland.
- Ads are shown to adult users on the Free and Go plans. In the EEA and Switzerland they are not personalised at launch: the system looks at the current conversation and general data such as approximate location and language.
- The ad policy works as an allowlist, not a blocklist: consumer goods, local services, travel and entertainment, digital products and education. Everything else is not permitted at launch.
- Finance, health and legal services are restricted categories, open only in the US and only after individual vetting. In Poland they are in practice ruled out.
- You cannot change the country, currency or time zone of an ad account once it has been created. The account should be opened by the company that pays, and the agency is invited afterwards.
- The minimum daily budget for an account billed in zloty is 65 PLN, but on a single day a campaign can spend twice that, and over seven days seven times that amount.
- The landing page must let the OAI-AdsBot crawler in. Blocking it in robots.txt, at the firewall or behind a CAPTCHA ends in a rejected ad.
- Since 31 August ChatGPT has been a very large online search engine under the DSA. By January 2027 OpenAI has to meet additional obligations, including a public ad repository.
- My August forecast of the timing was wrong. I assumed the turn of 2026 and 2027 for the first EU countries and a second wave for Poland.
Summary
ChatGPT Ads have been running in Poland since 24 August 2026, and since 31 August a Polish company in an allowed industry can open an account on its own and launch a campaign from 65 PLN a day. For now the ads are contextual only, they are shown to users on the Free and Go plans, and OpenAI allows four groups of consumer categories.
Before anyone starts working out bids, they should answer two questions. Is my industry allowed in at all? And am I opening the account for the right company, in the currency I want to be billed in? Only then does it make sense to talk about context hints, feeds and bids.
In Google Ads we learned to guess intent from three words. In ChatGPT the user describes it themselves, in full sentences. The winner will be whoever describes just as precisely whom their offer helps and in what situation.
As for the channel itself, after August I will keep more humility than I had then. I do not know how ads without personalisation will perform in Poland, and nobody knows yet, because there is simply no Polish data. What I do know is how to find out: one click campaign with a budget you can afford to lose, a few ad groups built around specific situations, UTM parameters and a comparison with Google Ads on the same conversions. At the minimum budget that comes to close to 2,000 PLN a month, so this is not a test that costs nothing. After a month, though, there will be something to draw conclusions from, and with a new channel that is more than any forecast will give you. Mine included.
Sources and materials
OpenAI announcements and documentation, the European Commission’s decision and trade press coverage on which I base the facts and dates cited:
- OpenAI – ChatGPT Ads expands across Europe (18 August 2026, updated 31 August 2026)
- OpenAI – A milestone in expanding access to AI (31 August 2026)
- OpenAI – Ad policies (version 1.5, updated 31 August 2026)
- OpenAI Help Center – Reklamy w ChatGPT (Ads in ChatGPT, user FAQ, Polish version)
- OpenAI Help Center – Ads Manager Availability (list of countries)
- OpenAI Help Center – Ads Manager Beta Account Setup (sign-up)
- OpenAI Help Center – Billing & Payment
- OpenAI Help Center – Create Campaigns for ChatGPT Ads (objectives and minimum budgets)
- OpenAI Help Center – Create Ad Groups for ChatGPT Ads (context hints)
- OpenAI Help Center – Set up Custom Audiences for your Campaign
- OpenAI Help Center – Create Campaigns from Product Feeds
- OpenAI Help Center – Advertiser Guidance for Allowing OpenAI Web Crawlers
- European Commission – Commission designates ChatGPT, Reddit, Roblox under Digital Services Act (31 August 2026)
- Search Engine Journal – ChatGPT Is Now A “Very Large Online Search Engine” In The EU
- Euronews – ChatGPT to start showing ads across Europe next week (21 August 2026)
- Marketing i Biznes – Reklamy w ChatGPT ruszyły w Polsce (Polish trade press on the launch in Poland: no ad library, monitoring tools)
- Press.pl – Mediapanel: z ChatGPT korzysta już ponad 10 mln internautów (Polish press: Mediapanel data, over 10 million internet users in Poland use ChatGPT, 22 November 2025)
Wondering whether ChatGPT Ads make sense in your industry?
We will check whether your offer fits the allowed categories, whether the landing page will pass review and how this channel relates to your visibility in AI answers.
Book a consultationHow 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