How ChatGPT serves ads. Here's the full attribution loop.

Speaker 1: Up next, we have a deep dive into how ChatGPT is actually serving ads. It is a fascinating look at the technical architecture behind the scenes of those conversational recommendations we are starting to see.

Speaker 2: Oh, interesting. I have noticed the suggestions popping up, but I assumed it was just part of the model's output. How does it actually work under the hood?

Speaker 1: It is essentially a two-part system. On the backend, OpenAI injects a specific data object directly into the conversation stream while the model is generating text. It acts like a digital billboard that appears right in the middle of your chat.

Speaker 2: So, it is not just the AI 'thinking' of a brand, but a structured ad unit being pushed to the interface. What happens when you actually click on one of those?

Speaker 1: That is where the second half, the merchant side, kicks in. There is a tracking SDK called OAIQ that runs in your browser. It uses encrypted tokens to link your click back to the ad, almost like a digital breadcrumb trail that follows you to the merchant's website.

Speaker 2: That makes sense. It sounds like a sophisticated way to measure performance, but it also means they are tracking that interaction quite closely. Are these ads just random, or is there a pattern?

Speaker 1: The research shows they are highly contextual. If you are chatting about planning a trip to Beijing, you might get a food delivery ad or a tour booking link. It is tailored to the specific topic of your current conversation.

Speaker 2: It is a clever way to integrate advertising without feeling like a traditional banner ad. It feels more like a helpful suggestion, even if the underlying mechanics are quite complex and data-heavy.

Speaker 1: Exactly. It is a closed loop where OpenAI hosts the creative, tracks the click, and then monitors your activity on the merchant site through that SDK. It is a very integrated approach to monetization.