A recent research paper from Stanford outlines a problem that has received limited attention as AI increasingly mediates internet search and e-commerce: it is becoming harder to tell whether recommendations from agentic AI—chatbots or virtual agents that can act on users’ behalf—stem from advertising payments or from impartial information.
The study gives a simple consumer example: in the near future you might ask an AI agent for a new toaster while driving home, and the toaster could arrive the next day. Behind the scenes, five toaster manufacturers might have bid to influence which product the agent sees or orders. In such a scenario, how can a user be confident they received the best toaster or the best price rather than the product favored by commercial agreements?
The authors warn the stakes rise when this pattern spreads to more consequential domains such as healthcare or financial services. If every person has a sufficiently capable personalized agent that resists deceptive marketing, the risk would be reduced. But the paper argues a more probable outcome is widespread use of free chatbots subsidized by ads and partner payments, which creates incentives for paid placement and potential conflicts of interest.
The Stanford researchers note that government regulation could help—for example, laws requiring chatbots to disclose financial incentives before offering advice. However, they caution that disclosure alone may be difficult to enforce and may not provide sufficient transparency. If an adtech-like industry develops around agentic commerce, the paper argues, it could become very hard to untangle who paid whom and how recommendations were shaped.
Authored by Reed Albergotti, the paper calls for early attention to these issues so that policy and technical safeguards can be designed before entrenched business practices make remediation difficult.
