For weeks the question we hear most often from clients has been the same: "Should we be advertising on ChatGPT?". It is a fair question, because in the space of eight months OpenAI’s ads have gone from a statement of intent to a product you can buy in Europe. It is also a question worth answering calmly, because the topic is surrounded by second-hand figures and very few certainties. In this article we set out what OpenAI has confirmed, what the press has reported and what is still unknown, with dates. Everything here reflects the position as of October 2026.
What ChatGPT ads are
Ads appear below ChatGPT’s answer when there is a relevant sponsored product or service for the conversation. They are labelled as sponsored and visually separated from the organic answer: OpenAI has said from the outset that advertising does not influence what ChatGPT replies.
Who sees them and who does not, according to OpenAI:
- Ads are shown to logged-in adult users on the Free and Go plans (the budget plan, around €8 a month in Europe).
- Ads are not shown to Plus, Pro, Business, Enterprise and Education subscribers, nor to accounts identified as belonging to under-18s, including through age prediction.
On format, this is what we know as of October 2026: a single ad at the foot of the answer; a product carousel for feed-based campaigns, showing several items from the same advertiser in one slot; and, according to several industry sources, a more experimental format that lets users ask ChatGPT questions about an ad. On 5 October 2026 OpenAI announced a new visual format that will sit alongside images users generate, in testing in the US only with an initial group of advertisers.
How we got here: the timeline
- 16 January 2026 OpenAI announces plans to test ads in the US on the Free and Go plans and publishes the principles that will govern them.
- 9 February 2026 The first ads go live in the US with a small group of advertisers. The press reports prices of around $60 per thousand impressions and high minimum spend commitments.
- 5 May 2026 Self-serve Ads Manager in beta for US advertisers, CPC bidding alongside CPM, minimum spend removed, Conversions API and pixel for conversion measurement.
- Spring-summer 2026 Gradual expansion to markets including the UK, Canada, Australia, Japan, Korea, Brazil and Mexico.
- 24 August 2026 Ads go live in 31 European markets: the 27 EU countries plus Iceland, Liechtenstein, Norway and Switzerland. Initially, buying runs through OpenAI’s sales team and partner agencies.
- 31 August 2026 Self-serve Ads Manager opens to European advertisers. The same day, according to reports, the European Commission designates ChatGPT a very large online search engine under the Digital Services Act.
- Late September 2026 Bulk creation of product feed campaigns from a spreadsheet, product review status in Ads Manager, more conversion data in the Insights API, expansion to South-East Asia and Taiwan.
- 5 October 2026 Announcement of the visual format alongside generated images (US only), around twenty measurement partners and brand suitability pilots with DoubleVerify and IAS.
The rules OpenAI has set itself
In its January announcement OpenAI listed a set of principles, which also form the frame for judging the channel:
- Answer independence. Ads are separate and labelled, and do not influence the answer.
- Conversation privacy. Chats are not shared with advertisers and data is not sold.
- Choice and control. Users can turn off personalisation and clear the data used for ads, and there is always a way not to see ads, starting with the paid plans.
- No optimising for time spent in the app.
OpenAI’s help centre adds the operational exclusions: no ads for minors, no ads next to sensitive conversations such as personal health, and for now no political advertising. The October announcement describes controls that assess whether a conversation is suitable for advertising, to keep ads out of emotionally vulnerable contexts.
In Europe the rules are tighter. According to reports at the European launch, in the EEA and Switzerland ads are initially contextual only: they draw on the topic of the current conversation, approximate location, device, time of day and language, with legitimate interest as the legal basis. Personalised ads, which would also use chat history and memory, will only come after users are explicitly asked for consent. For an advertiser that means something simple: in Europe today you are buying context, not profiles.
How you buy and how you measure
The point that most surprises people coming from Google Ads: there are no keywords. According to guides from people using the platform, at ad group level you write "context hints", plain-language descriptions of the conversations and situations where the product may be useful. The system treats them as guidance alongside the ad copy and the landing page, but they do not guarantee delivery in any particular conversation. It is closer to an intent-driven auction, like the one Google is building with ads in AI Mode, than to traditional keyword search.
| Topic | Status as of October 2026 | Reliability |
|---|---|---|
| Availability to users in Europe | Ads live since 24 August 2026 on Free and Go | Confirmed by OpenAI |
| Self-serve buying from Europe | Ads Manager self-serve since 31 August 2026, alongside partners | Confirmed, reported by several outlets |
| Pricing model | Auction with CPM or CPC bidding | Confirmed |
| Minimum spend | None in self-serve (before May 2026 high commitments were required) | Reported by the press |
| Real cost per click and per conversion in European markets | No reliable public benchmark | Unknown |
| Measurement | Pixel, Conversions API, Insights API, third-party attribution and incrementality partners | Confirmed |
| Personalised ads in the EU | Not at launch; consent only | Reported by the press |
| Visual format in image generation | US-only test from October 2026 | Confirmed |
| Arrival date of new formats in Europe | No public date | Unknown |
On measurement, the technical picture emerging from the documentation and industry guides is familiar: when a user clicks an ad, a click identifier is appended to the URL; the OpenAI pixel reads it and stores it in a first-party cookie, and the Conversions API lets you send conversions server-side, as you would with Meta. Partners such as Hightouch, Tealium and LiveRamp can forward existing conversion data. Static UTM parameters are preserved, so campaigns can be read in GA4 as well.
What about shopping inside ChatGPT?
In September 2025 OpenAI launched Instant Checkout in the US, purchases completed inside the chat, built on the Agentic Commerce Protocol. According to reports in March 2026, OpenAI then scaled back native checkout, sending users to the retailer’s app or site to complete the purchase. The pattern taking hold is "discover in ChatGPT, buy on the site". For a European retailer the practical consequence is that product data and the landing page matter today, not an in-chat checkout integration. We cover the wider agentic commerce picture in our piece on the Universal Commerce Protocol.
ChatGPT, Google and Microsoft compared
| ChatGPT | Google (AI Overviews and AI Mode) | Microsoft Copilot | |
|---|---|---|---|
| Where the ad appears | Below the answer, in a sponsored block | Above, below and within generated answers; new conversational formats in AI Mode | Alongside conversational answers |
| What triggers it | Conversation context and context hints, no keywords | Existing Search, Shopping and Performance Max campaigns, AI Max | Existing Microsoft Advertising campaigns |
| Buying tool | Separate Ads Manager | Google Ads | Microsoft Advertising |
| Maturity | Young platform, little historical data | Mature, but the new AI Mode formats are in US testing | Mature on search, smaller volumes |
The substantive difference is that with Google and Microsoft you reach AI surfaces through the campaigns you already run, whereas ChatGPT is a new channel with its own account, its own tag and its own learning curve. And the effect of generated answers on clicks remains an issue on the Google side too, as we discuss in our analysis of AI Overviews and CTR.
What we recommend doing now
1. Organic first: getting found by ChatGPT without paying
For almost every business this comes before any paid test, because it costs little and lasts:
- Accessibility. Check robots.txt and your firewall or CDN to make sure
OAI-SearchBot, the crawler behind ChatGPT search, is not blocked. It is separate fromGPTBot, which is used for training: you can block the latter and let the former through. - Product data. Clear titles, complete attributes, prices and availability consistent across feed, page and structured data. It is the same work Google asks for in Merchant Center.
- Content that answers. Pages that answer concrete questions (indicative prices, lead times, differences between options, who the service suits) are cited more readily than pages written around a keyword.
2. Measure what is already arriving
Links ChatGPT shows in its answers carry the utm_source=chatgpt.com parameter. In GA4 you can therefore already see how many sessions and conversions come from ChatGPT. The way we set it up: a custom channel group, "AI assistants", that collects chatgpt.com, perplexity, gemini and copilot sources, separate from generic referral. Bear in mind that some traffic carries no parameters, for instance when a user copies an address, and lands in direct. The figure is therefore a floor, not the total.
3. Test paid, but only if it makes sense
A test makes sense if at least two of these are true: you sell a product or service people ask for advice on before deciding; you have a solid feed or landing pages; conversion tracking is already reliable; you have budget headroom after covering demand on Google. It is better to wait if you are a local business whose searches are almost all "near me", if your Google budget is already stretched, or if you work in sectors where ads are excluded or heavily restricted, such as health and politics.
How to set up the test. An example, not a rule: 6-8 weeks, a budget you can afford to lose without touching the campaigns that work, one or two campaigns on your clearest products or services. Before you start, decide the yardstick (cost per qualified lead or return on ad spend, compared with Google and Meta) and reconcile OpenAI pixel data with GA4 and your CRM. Without a public benchmark, your own test is the only data point that counts.
Two technical notes. Use dedicated UTMs for ads (for example utm_source=chatgpt and utm_medium=cpc) so paid traffic stays separate from organic chatgpt.com traffic. And treat the OpenAI pixel like any marketing tag: it belongs under your consent management platform, exactly as you handle Consent Mode v2.
The risks to keep in mind
- Young measurement. Attribution is done in-house by OpenAI and third-party partners have only just arrived. Judge incrementality, not just attributed conversions.
- Limited control. With no keywords and no search terms report, you know less about where your ad appears than in a Search campaign. The DoubleVerify and IAS brand suitability work is still at pilot stage.
- Shifting rules. In Europe the framework is still evolving: personalisation only with consent, and new Digital Services Act obligations that, according to reports, include a public ad repository. Formats, targeting and prices may change from month to month.
- A partial audience. Paying subscribers, often the most engaged users with the most spending power, do not see ads. Worth remembering when estimating volume.
In short: ChatGPT ads in Europe are real and buyable, but they are a channel to explore methodically, not one to chase for fear of being left behind. If you want to know whether it makes sense for your business, our free audit starts from what you already have: tracking, product data, AI assistant traffic in GA4. You can see how we work in our process and which services cover new channels.