We built ChatGPT's novel ad stack for user trust and relevance (openai.com)

OpenAI has introduced a novel advertising stack for ChatGPT that prioritizes user trust and relevance by fundamentally redefining how ads are delivered within conversational contexts. Unlike traditional ad systems, which respond to discrete user actions, this bespoke platform is designed to navigate the evolving intent of users across multiple conversational turns. It employs a custom, post-trained large language model (LLM) that orchestrates ad retrieval, ranking, relevance evaluation, and auction processes while maintaining a seamless interaction with the core ChatGPT service. This architecture ensures that the ad system operates independently, preserving the integrity of user responses and preventing advertisement interference. Significantly, this new approach has enabled ChatGPT to achieve a remarkable annualized revenue run rate of $1 billion within just 200 days post-launch, attracting over 100,000 advertisers. By establishing stateful ad sessions and prioritizing user engagement and interaction history, the ads stack adapts dynamically to ongoing conversations. The orchestrator LLM simplifies the retrieval of suitable ads by summarizing users' evolving intents, thus ensuring that ads are presented only when they genuinely align with user needs. This innovative framework not only enhances the relevance of advertisements but also safeguards user experience, making ad placements more effective for both users and advertisers.