How OpenAI Built ChatGPT's Ad Stack Around One Orchestrator Model
OpenAI published an engineering post on October 21, 2026, written by two Members of Technical Staff, describing how it built the ads system inside ChatGPT. Everything below comes from that first-party post.
Transcript
OpenAI detailed today how ChatGPT ads actually work. Returning no ad counts as success.
OpenAI says chat intent develops over many turns, so it moved beyond the request and response architecture search ads use.
OpenAI's orchestrator model runs the flow in a loop of tool calls. Ads sessions remember context across turns and multiple conversations.
Ad selection runs in a separate service. OpenAI says ChatGPT has no dependency on it and never delays answers for ads.
This is OpenAI describing its own system on its own website. It publishes no latency numbers, no relevance measurements and no outside review.
Inboxsmith builds AI receptionists for small business. Please like and subscribe for more news.
Sources
Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:
- Traditional advertising systems are built to respond to discrete units of user intent. Building an ads stack for ChatGPT was different: user intent develops across conversations and shifts over multiple turns of a conversation, sometimes without a fitting advertising moment. To support this new paradigm, we built a bespoke ads delivery system. Building our ads system required us to move beyond the request-response advertising architecture and build an ads stack from scratch.(OpenAI (openai.com), first party engineering post titled "How we built ChatGPT's novel ad stack for user trust and relevance", by Ritendra Datta and Karthik Gopalratnam, October 21, 2026)
- The other key architectural component in our ads stack is the post-trained LLM that interprets conversations and runs the entire ads flow in a loop, making tool calls to the core functions of the ads system, from the retrieval stage to ranking to auction and, ultimately, delivery. We call this LLM the orchestrator.(OpenAI (openai.com), first party engineering post titled "How we built ChatGPT's novel ad stack for user trust and relevance", by Ritendra Datta and Karthik Gopalratnam, October 21, 2026)
- Our ads system supports the nuance of evolving user goals by making three key design decisions: 1) implementing stateful ads sessions, meaning information is remembered across turns of a conversation and across multiple conversations. The ads system follows and adapts to the developing exchange rather than relying on a single flattened query.(OpenAI (openai.com), first party engineering post titled "How we built ChatGPT's novel ad stack for user trust and relevance", by Ritendra Datta and Karthik Gopalratnam, October 21, 2026)
- Running the ads selection process separately from ChatGPT's answer generation, in its own service, to keep ads from influencing answers and maintain ChatGPT's neutral voice. This important architectural boundary, different services that use separate models and separate code, lets core ChatGPT servers handle generating the organic response. ChatGPT has no dependency on the ads system, the services are in no way intertwined and don't share a content generation model. So if the ads serving system is unavailable or slow, ChatGPT's organic response continues unaffected. This means the ads system can't take down or delay the ChatGPT product.(OpenAI (openai.com), first party engineering post titled "How we built ChatGPT's novel ad stack for user trust and relevance", by Ritendra Datta and Karthik Gopalratnam, October 21, 2026)
- The other benefit of this separation is that it keeps the bar very high for valuable ads. Returning no ad is an ordinary, successful outcome. The system only delivers an ad if it's clearly beneficial to users and advertisers.(OpenAI (openai.com), first party engineering post titled "How we built ChatGPT's novel ad stack for user trust and relevance", by Ritendra Datta and Karthik Gopalratnam, October 21, 2026)
- The ad request might not arrive in time: If the ChatGPT model is ready to respond, but the ad is not yet ready to place, the ad request times out and the ad doesn't get delivered. The ChatGPT model response time is dynamic, and we don't delay chat responses to wait for ads to load. Every claim is OpenAI's own, published on OpenAI's website by two Members of Technical Staff: the post publishes no latency figures, no relevance or accuracy measurements, no third-party audit of how often ads influence answers, and it names no outside reviewer.(OpenAI (openai.com), first party engineering post titled "How we built ChatGPT's novel ad stack for user trust and relevance", by Ritendra Datta and Karthik Gopalratnam, October 21, 2026)