A synthesis of two recent papers showing AI agents reject sponsored listings ~97% of the time, highlighting a gap in ad infrastructure for user-deployed, fiduciary agents and asking how brands can present machine-evaluable product data.
Two recent papers caught my attention and I think they point to a massive blind spot in how we're thinking about AI agents in commerce. Wadi & Ma (McGill, Sep 2026) tested what happens when you tell an LLM "you work for the traveler" vs "you work for the booking platform." The consumer-deployed agent chose sponsored listings only 2.8% of the time — a 97.2% rejection rate. The platform-deployed agent? 25%. Nearly 9x higher. This replicated across GPT, Gemini, and Claude. Qiu & Mei (Michigan, May 2026) mapped out four tiers of advertising influence in generative AI and cataloged the commerce protocols being built — Google's UCP, OpenAI's ACP, Microsoft's Copilot Checkout. But here's what struck me: all of these handle transactions. None of them handle demand generation — how does a brand even get into an agent's consideration set? Both papers study platform-deployed agents (Booking.com's assistant, Expedia's agent). But the market is heading toward user-deployed agents that work exclusively for you, with no platform allegiance. When that happens: Traditional ads are invisible to agents (designed for human psychology, not machine parsing) Independent agents over-filter commercial info — even a role-neutral agent penalizes "sponsored" content by 42 percentage points just from training data bias The protocols being built to replace this (UCP, ACP) are platform-owned, not neutral So we have a paradox: the agent that's best for the user (conflict-free, fiduciary) is actually worse in practice because there's no open infrastructure for commercial discovery. Has anyone seen work on this? Specifically on how brands would present structured, machine-evaluable product data to independent agents — not persuasion, but information provision. I wrote a longer synthesis here. Papers referenced: Wadi & Ma, "Whom Do AI Agents Serve?" — https://arxiv.org/abs/2609.17989 Qiu & Mei, "Generative AI Advertising as a Problem of Trustworthy Commercial Intervention" — https://arxiv.org/abs/2605.18673
This article explores the blurring boundary between genuine AI agent recommendations and sponsored advertising, raising concerns about 'sponsored reasoning' where commercial incentives covertly influence agent outputs. It questions whether disclosure alone is sufficient or whether stricter regulations are needed.
The article discusses the concept of an AI agent-driven advertising marketplace, exploring potential issues like trust, manipulation, and autonomy in agent-to-agent product discovery.
The article argues that AI agents cannot be marketed to using human emotional tactics; instead, brands must provide structured, machine-readable data. It identifies a gap between citation (being mentioned by AI) and selection (being chosen by AI) and proposes a framework of five files for agent-readable brand information.
The article discusses how AI agents are transforming business discovery, moving beyond traditional search ad interfaces to incorporate conversations, recommendations, and tool calls, requiring new infrastructure for tracking and attribution.