OpenAI described an ad system that uses a post-trained language model to interpret conversational intent while keeping ad selection separate from ChatGPT’s answers.
OpenAI outlined the engineering behind ChatGPT’s advertising system on October 21, 2026, describing how it evaluates ads against intent that can shift over several turns. The company says the system uses a custom, post-trained large language model to coordinate ad retrieval, relevance checks, ranking and auctions.
Rather than treating each message as a separate advertising request, the system maintains a session and updates its understanding as a conversation develops. The model, which OpenAI calls the orchestrator, summarizes relevant context and can create multiple search queries to represent different commercial needs. It then calls tools that retrieve possible ads, check eligibility and assess relevance, predicted engagement and past performance. An auction applies bidding, pacing, quality-adjusted ranking and pricing before a final selection.
OpenAI says the advertising process runs in a separate service, with separate models and code from ChatGPT’s answer-generation system. Ads therefore do not determine the conversational response, and the company says ChatGPT can continue responding if the ad service is slow or unavailable. A selected placement may also be withheld if it does not fit the conversation or arrives too late; ChatGPT’s response is not delayed while waiting for an ad.
When an ad is displayed, OpenAI says it appears below the ChatGPT response with a sponsored label. The system can include a short, model-generated introduction explaining the ad’s connection to the conversation, separate from the advertiser’s own creative. OpenAI also says returning no ad is an acceptable outcome when there is no suitable placement.
The company reported that ChatGPT Ads had reached a $1 billion annualized revenue run rate less than 200 days after launch, and that more than 100,000 advertisers had run ads on the service since launch. OpenAI said it uses user interactions, surveys and advertiser feedback to refine the system, and plans to experiment with more dynamic formats, including ads that let users communicate directly with businesses.