ChatGPT Ads: how they work and whether they change answers

An example: someone asks ChatGPT which shoes suit trail running. ChatGPT writes its answer with the brands and models it considers suitable. Below it, if a store has paid for a campaign related to that conversation, a card can appear with its image, a title and a link to its website, labelled as an ad.

The difference from an organic recommendation is who decides. The model writes the answer from what it knows and what it finds on the web; the ad system places the ad because an advertiser paid for that space. A brand showing up in the card says nothing about whether ChatGPT would have named it in the answer.

OpenAI repeated the same commitment when it extended ads to Europe in August 2026: ads are labelled, kept separate from the answer and do not influence it. That is the company’s statement, not a third-party measurement, and from the outside nobody can check how each answer is put together.

What OpenAI’s documentation does make clear is that buying ads and appearing in the answer are two separate routes. For ChatGPT to name a brand in its answer, it has to be able to read the brand’s website and find sources that talk about it; the steps are in how to appear in ChatGPT without paying.

The practical consequence for a business is twofold. An ad does not improve its place in the answers, and stopping the ad does not hurt it either. And a competitor running ads can appear below an answer that recommends you, but cannot buy that answer’s silence about you.

The rules OpenAI has published on who sees ads and where:

• Plans: only Free and Go, the cheapest paid plan. Plus, Pro and Enterprise stay ad-free, according to OpenAI’s announcement of 21 October 2026.
• Age: no ads for accounts that say they are under 18, or that OpenAI predicts belong to minors, according to its January 2026 announcement.
• Topics: no ads near sensitive or regulated topics such as health, mental health or politics, according to the same announcement.
• Devices: the advertiser documentation lets campaigns choose between the ChatGPT iOS app, the Android app and the web.

According to the January announcement, someone who sees an ad can learn why it is shown, dismiss it and say why it is not relevant. OpenAI also says it keeps conversations private from advertisers and never sells the data of people who use ChatGPT.

The list of eight markets in the second row is the one Adweek published and The Next Web reported on 20 August 2026; OpenAI does not list them in a single document. In the October announcement, OpenAI also said that more than 100,000 advertisers had run campaigns on ChatGPT since the test began. It is a cumulative figure: it does not say how many advertise today or how much they spend.

What OpenAI’s advertiser documentation makes clear before you start:

• Campaigns are not targeted with keywords. Each ad group carries “context hints”, phrases that describe the product and its use, and OpenAI states that hints are not exact-match keywords.
• The advertiser chooses whether to pay for impressions or clicks, and can set the bid or let OpenAI adjust it to seek more clicks or more conversions.
• To measure sales or sign-ups there is a pixel on the website and a Conversions API that sends events from the server.
• When an ad is submitted, the OAI-AdsBot crawler visits the landing page to check it complies with the policies, and may use its content to decide when to show the ad.

OpenAI does not publish a rate card. According to PPC Land (21 October 2026), the cost-per-thousand-impressions model launched at 60 US dollars per thousand, and cost-per-click bidding was added in May 2026, when Ads Manager opened to all US businesses. The real price of each campaign depends on the bid and the competition.

GEO is the work of getting AI assistants to cite and recommend a brand; what GEO is and how it differs from SEO has its own article. Kulturo, a tech recruitment agency in Barcelona, gets about 7 qualified form submissions a month from people who say they found it by asking ChatGPT or Claude, according to Kulturo’s data (October 2026).

HeyTalent, a sourcing tool for recruiters, gets between 2 and 3 demos a week from people who arrived by asking ChatGPT, according to HeyTalent’s data. Kulturo’s full figures are in the Kulturo case study with DoGeo.

The second point is where they come from. According to what Kulturo’s leads say, they arrive mainly from ChatGPT, Gemini and Claude. A ChatGPT ad covers only the first of the three; organic visibility is built from the same sources for all of them.

Before paying to appear below an answer, it is worth knowing whether the website is ready to appear inside one. The starting point is to check whether AI crawlers can read your site, and the next is getting other sources to talk about the brand: why they weigh so much is covered in brand mentions in AI search.

Crawler logging is a firewall or CDN rule that records, without blocking, every request from the assistants’ crawlers, identified by their user-agent (the name they present) and verified against the IP addresses each provider publishes. Those logs show which pages each crawler visits and when.

According to OpenAI’s webmaster documentation, each crawler tells you something different. OAI-SearchBot surfaces websites in ChatGPT’s search results. ChatGPT-User opens a page when someone asks ChatGPT a question: it is the closest thing to knowing that ChatGPT read that page to answer a person. OAI-AdsBot only visits pages submitted as ads, and OpenAI says what it collects is not used to train its models.

Separating the three in the logs avoids mistaking an ad review for a recommendation. This is what DoGeo sets up for each client. What the logs do not show is what ChatGPT writes in its answer: that still takes a fixed set of questions asked every month.