Advertising learns to converse! OpenAI reveals how it selects ads
OpenAI explained how it built its ad infrastructure for ChatGPT conversations. Unlike systems based on isolated searches, this mechanism interprets goals that can evolve as the user chats. The platform uses an orchestrator model, ad retrieval, eligibility filters, relevance evaluation and an auction kept separate from response generation.
Traditional advertising usually starts from specific signals such as a search or a visit. Conversations present a different challenge: a person may change their intent, ask for recommendations or explore hypothetical scenarios without each message signaling an immediate commercial interest.
OpenAI describes a system that maintains context to determine when an advertising opportunity exists. A specialized model coordinates retrieval, filtering, evaluation and auction tools. Ads must meet eligibility and usefulness criteria; if no appropriate match is found, the system may choose not to show any. Ad selection runs on a service separate from the one that generates the assistant's responses.
For the marketing sector, these capabilities open a channel focused on conversational intent rather than isolated keywords. They also raise questions about measurement, relevance and trust. The company maintains that private conversations are not shared with advertisers and that ads do not alter the model's responses.


