Season 7, Episode 35: OpenAI's advertising ambitions (with Asad Awan)

Principles and the trust rubric

Awan, who previously spent a decade on ads/monetization at Meta before joining OpenAI's pilot launched in February, opens by restating OpenAI's five ad principles: democratizing AI access, preserving answer independence (ads never influence the underlying model's responses), keeping conversations private, letting paid users opt out of ads, and avoiding optimization for empty calories like time spent. He says these principles are unusually public and function as an internal accountability mechanism, embedded into an ad score metric that blends user value and advertiser value at the engineering level.

Conversational intent vs. keyword search

Seufert's framing — that a search query gives little room to express intent while a conversation reveals budget, constraints, and preferences over multiple turns — anchors much of the discussion. Awan says users pursue multiple simultaneous user journeys across threads, bridged by ChatGPT's memory feature, which already personalizes organic answers (e.g., connecting a hiking-shoe question to a previously mentioned Alaska trip). He breaks the ad opportunity into three classes: personalized contextual ads one step removed from pure keyword matching, pure personalization moments like waiting for an image to generate, and letting contextual and personalized ad candidates compete for the same slot. He adds that not every conversation is inventory — a personal topic, like the photo of his kid's bruise he uses as an example, is not turned into a commercial opportunity.

Format innovation: sponsored agents and dynamic creative

Beyond ad selection, Awan highlights three native format ideas: dynamic creative (explaining why an ad was selected, e.g., emphasizing cushioning for a nurse who stands all day), variant selection (matching size/color from product feeds), and the sponsored agent, where a user enters a clearly labeled, separate conversation with the advertiser after clicking, keeping the original ChatGPT thread untouched. Awan frames this design as OpenAI's response to skepticism that chatbot ads inherently undermine trust, since the commercial interaction is explicitly demarcated from the assistant conversation.

The Meta comparison and value exchange

Seufert raises his prior writing on chatbot ads and a specific natural-experiment argument: unlike Meta, which introduced subscriptions only after users had a free experience (making it impossible to gauge true subscription demand), OpenAI introduced ads alongside a concrete value exchange from day one — accepting ads unlocks higher rate limits and more compute rather than removing something previously free. Awan agrees, tying it back to OpenAI's mission of making AI available without constraints, though he notes the inverse question — how paying users might still see ads — sometimes comes up.

Data, infrastructure, and roadmap

On advertiser-supplied data, Awan distinguishes product/brand information (landing pages, context hints on brand voice) from conversion data, which grounds and calibrates first-party intent signals against actual purchase behavior. Asked how building an ad stack from scratch differs from Meta's legacy build, he says OpenAI is skipping incremental constraint-driven stages that early Meta ad targeting required, in favor of maximum automation from day zero — simplified bidding (max bid plus a new auto-bidder), prompt-based campaign creation via a ChatGPT plugin, and integrations with Shopify and HubSpot. Roadmap priorities he cites: expanding to more countries, improving measurement (he flags a low-measurement problem affecting SMBs that lack the cross-channel tools larger advertisers already have), and continued format optimization for verticals like lead-gen and e-commerce via sponsored agents. He closes by reiterating that success is defined by deepening long-term user trust and habitual return, not maximizing impressions or time spent — a user deciding quickly and leaving still counts as success.