ChatGPT advertising with email and phone targeting

OpenAI published a technical overview of ChatGPT's advertising infrastructure, describing an LLM orchestrator that coordinates retrieval, ranking, and auction tools to interpret conversational intent for ad targeting. The system is designed to keep paid placements visually and contextually separate from AI-generated answers and to respond to an evolving conversation context. OpenAI described user trust as a core design constraint. Separately, 404 Media tested the platform and found two targeting modes: contextual hints based on prompt topic or keyword, and custom audience lists built from email addresses, phone numbers, or Android ad IDs. OpenAI cross-references uploaded identifiers against its ChatGPT user database to confirm active accounts before targeting. Lists of 25,000 or more entries can be used for targeting; smaller lists are restricted to exclusion, a floor OpenAI said is designed to prevent hyper-targeted harassment. OpenAI accepts raw email and phone number uploads and claims to delete the data after processing, though it is unclear whether the data is hashed before deletion, unlike platforms such as Meta which hash uploaded data. The ads personalization opt-out toggle does not stop ads from appearing or all tracking; OpenAI's interface states ads may still be based on a user's current chat. 404 Media estimated most of ChatGPT's roughly 1 billion monthly users are unaware these targeting mechanisms exist.