OpenAI explains how ChatGPT picks ads, separately from its answers
OpenAI has published a technical account of how advertising works inside ChatGPT, describing a dedicated model that reads a conversation, turns relevant parts of it into search queries, and submits candidate ads to an auction. The company says the ad system and the model that writes answers run on separate models and code, so ad selection does not feed back into what ChatGPT writes.
The ad model, which OpenAI calls the orchestrator, tracks commercial intent as it develops across messages rather than judging only the latest question. It checks candidates for location targeting, policy restrictions, budget and pacing, then hands them to a separate auction tool that sets ranking and price. A winning ad can still be withheld if a later check finds it unsuited to the discussion, or if it misses the delivery window while the answer is being written.
OpenAI says ChatGPT Ads reached a $1 billion annualised revenue run rate less than 200 days after launch, with more than 100,000 advertisers running ads since then. It published no performance benchmarks, no conversion or return figures, and no controlled studies showing whether the ads produced purchases that would not otherwise have happened.
For brands, the account suggests that matching an ad to a specific customer requirement raised mid-conversation matters more than mirroring a single search term, and that low impressions can stem from eligibility checks, relevance checks or timing rather than bidding. OpenAI has not said how much of that pipeline advertisers can see in their own reporting.
Reported by Search Engine Journal ↗
Topics: ChatGPT, AI advertising