ChatGPT Ads runs ad selection through an LLM orchestrator separate from answer generation
OpenAI has detailed a separate LLM architecture for ChatGPT Ads that tracks conversational context through stateful sessions and uses tool calls to retrieve, filter, rank, and auction ads.

AT A GLANCE
Key points
- Stateful ad sessions follow intent across conversational turns
- A post-trained LLM orchestrates retrieval, filtering, ranking, and auctions
- Ad selection runs separately from organic answer generation
- Ads can be withheld when suitability or timing checks fail
What is changing?
OpenAI says it built a bespoke advertising system for ChatGPT that differs from traditional search-ad flows by tracking intent as it changes across multiple conversational turns. For eligible conversations, the system uses stateful ad sessions that update context rather than treating every message as a standalone search query.
The company’s post-trained LLM orchestrator prepares intent summaries, retrieves candidates through semantic search, filters them for budget and policy eligibility, evaluates quality and relevance, and then calls an auction tool. Ads that fail final suitability checks or arrive after the answer is ready are not shown.
Ad selection runs on separate models, code, and services from ChatGPT’s organic answer generation. OpenAI says a slow or unavailable ad service does not affect the answer path, while displayed placements appear below the response and are clearly labeled as sponsored. The company also says ChatGPT Ads reached a $1 billion annualized revenue run rate in less than 200 days.
Explore the original sources
Source date: 21 October 2026.
Published by AI Radar: 22 Oct 2026, 09:06 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.






Loading comments…