OpenAI has announced that it will show ads inside ChatGPT, and it is updating its privacy policy to match. The change affects the free version and the low-cost Go tier; anyone paying for Plus, Pro, or a business plan stays ad-free. The company promises to keep ads strictly separate from the conversation: advertisers will not see chats, and OpenAI says it will not use stored chat history to choose which ad to show. Whether that dividing line holds is the real question.
Key takeaways
- Ads will appear in the free version of ChatGPT and in the Go tier. Paid plans (Plus, Pro, Enterprise, Business, Education) remain ad-free.
- Ads are selected from the context of the current conversation, plus coarse signals such as approximate location, time of day, and device type.
- Past chats and saved memories will not feed into ad selection, according to OpenAI. Advertisers receive only aggregated metrics such as views and clicks.
- In the EU, personalized ads are allowed only with explicit consent. In the US, ads are on by default and users can opt out.
- OpenAI bases the data processing on a legitimate interest and argues that free access to capable AI depends on ad revenue.
What changes in the privacy policy
The revised policy spells out for the first time how ads in ChatGPT are meant to work. The basis is contextual targeting: ads follow the topic of the current chat rather than a profile built up over weeks. Ask about hiking boots and a relevant ad may appear, without earlier conversations about health or finances being pulled in. On top of that, OpenAI says it uses only a few environmental signals: approximate location, time of day, and device type.
Anyone who later allows genuine personalization has to opt in actively. OpenAI may then also draw on ad interactions and details from chats to place ads more precisely. In the EU, the company grounds that consent in Article 6 of the General Data Protection Regulation, which sets out the legal bases for processing personal data. For users in the European Economic Area and Switzerland, the Irish subsidiary OpenAI Ireland is the data controller; the company bases the transfer of data to the US on the EU adequacy decision.
What advertisers see and what they do not
OpenAI stresses that companies buying ads get no access to conversations, chat history, personal details, or saved memories. They receive only summarized performance data: how often an ad was seen and how often it was clicked. The targeting itself runs internally at OpenAI. Ads are meant to be clearly labeled as sponsored and visually separated from the model’s answers.
This architecture resembles what search engines have done for years, but it carries a different weight in a chat. A conversation with an assistant is often more personal than a search query, and the line between a neutral answer and a paid recommendation is harder for non-experts to spot. That is exactly where user trust in the separation will be decided. How much an AI service learns about its users anyway was laid out in the analysis of Proton’s AI Paper Trail, which we covered last week.
Why OpenAI is doing this
OpenAI justifies the ads with cost. Free access to capable AI, the argument goes, can only be sustained through additional revenue streams. The company already earns money from subscriptions and programming interfaces, but the number of paying customers is growing more slowly than the number of free users. Ads reach the majority who will never buy a subscription. The trial began in the spring of 2026 in the US, and in mid-August it expanded to the United Kingdom, Mexico, Brazil, Japan, and South Korea.
The move has been foreseeable since OpenAI hired experienced ad executives from Meta and Google. For the industry, an AI assistant with hundreds of millions of weekly users is an attractive place to advertise. From the user’s point of view, the business model shifts: the product does not get more expensive, but attention inside the conversation now has a price. That people have long used ChatGPT for far more than work, including very private questions, is clear from the independent usage data collected by the AI Observatory.
Outlook
Whether the promised separation between ad and answer holds will only become clear in daily use. Three points will matter: whether personalized ads in the EU really run only on genuine consent, whether ad labeling is prominent enough, and whether contextual targeting is not gradually widened after all. European users should check their settings: the default here should not allow personalized ads without consent. Anyone who wants to verify that and document their objection should do so soon, before ads roll out broadly.