StackAdapt, a demand-side platform, is pitching advertisers on placements inside ChatGPT with targeting based on "prompt relevance." A leaked pitch deck reviewed by Adweek reporter Trishla Ostwal shows CPMs ranging from $15 to $60, with a $50,000 minimum buy-in for the pilot. The deck, shared with select buyers on March 27, frames the offering as early access to a "discovery layer" that catches users while they research and compare products inside ChatGPT.
The technical implementation is where things get uncomfortable. "Prompt relevance" targeting almost certainly relies on vector embeddings, converting what users type into numerical representations that get matched against advertiser content in real time. That's a genuine step beyond keyword-based advertising. But it also means something derived from your prompts has to reach StackAdapt's servers for the targeting to work. This collides with prior assurances that ads wouldn't tap into prompt data. The distinction between "no prompt access" and "we pass a semantic fingerprint to a third-party DSP" is thin enough that it might not survive scrutiny.
Commenters on Hacker News questioned why OpenAI would outsource this to a middleman rather than building internally. Fair points. You trade margins for added complexity, and the security surface area only grows. The AI advertising gold rush is clearly accelerating regardless. Bluefish raised $43 million to help brands appear in ChatGPT and similar tools. Others are pushing LLM-driven media planning. StackAdapt simply got caught with its deck leaking.
Ads were always coming to conversational AI. OpenAI Tests Ads in ChatGPT. Users will decide whether to stay once they realize their questions drive the targeting. Regulators will have questions too, about what data flows where.