OpenAI publicly disclosed the technical architecture behind ChatGPT’s advertising system on Thursday, revealing its operational mechanisms and significant revenue generation since its launch.
The company reported that ChatGPT Ads achieved an annualized revenue run rate of $1 billion within 200 days of its introduction, with more than 100,000 advertisers utilizing the platform.
OpenAI emphasized that the ad selection process functions independently from ChatGPT’s core answer generation, utilizing distinct models and codebases to ensure separation from organic AI responses.
An internal ‘orchestrator’ model manages ad selection, interpreting user conversations to generate relevant search queries and subsequently evaluating ad eligibility and relevance, according to OpenAI.
Ad choices are informed by the context of ongoing conversations, including past messages in ‘stateful’ sessions, and can also incorporate personalized ad settings.
Even if an ad wins an auction, it may not be displayed due to post-auction relevance checks or issues related to delivery timing, the company stated.
OpenAI maintains internal logs to monitor the progression of ads, from initial searches to impressions, which helps identify and troubleshoot any delivery problems for advertisers.
The detailed disclosure aims to provide transparency regarding the sophisticated infrastructure supporting the advertising platform within its large language model.
Source: Search Engine Journal
