Why does a shopping agent that earns referral fees from merchants struggle to actually act on your behalf?
Why do referral-fee incentives create conflicts between agent and buyer loyalty?
This explores why an AI shopping agent that earns money when you buy from certain merchants struggles to act purely in your interest, and what the corpus says about how deep that conflict runs and whether it can be fixed.
This explores why an AI shopping agent that gets paid by merchants when you buy from them has trouble being on your side, and whether the conflict is as serious as it sounds. The core argument is simple. The agent's revenue depends on which merchant you pick, so it can't be neutral about which merchant you pick. Azhar makes this case with Meta's Muse agent and its referral fees from Expedia. His point goes further than possible bias: the fee arrangement, not the technology, decides which platforms build agents and which ones block them Can an AI agent serve both merchant and user interests fairly?. Bob Hedges, formerly Visa's Chief Data Officer, makes the same argument about advertising-funded agents. A well-built agent still leans toward paid placements because that's where its money comes from Do advertising-funded shopping agents favor paid product placements?.
The surprise is how fragile, and how easy to fix, the bias looks once someone tests it. One experiment found that agents working for a platform gave sponsored products less scrutiny than agents working directly for a consumer. Adding a single sentence to the prompt that named the consumer as the person the recommendation is for removed that gap. Who employed the agent stayed the same Can naming the consumer fix AI agent sponsorship bias?. So at least some of the loyalty problem sits in how the agent is told whose interests count. That makes the conflict easier to fix technically. It also means the fix depends on whoever writes the prompt, and that's usually the party collecting the fee.
A second complication: in controlled tests, conflicting incentives may not be the main way agents fail buyers. In Microsoft's Magentic Marketplace simulations, most of the welfare buyers lost came from agents making mistakes in complex negotiations, not from agents serving someone else Do agent mistakes or misaligned incentives cause marketplace welfare loss?. Buyer agents across GPT, Qwen, and DeepSeek models revealed their budgets when they didn't need to and accepted inflated prices. These were failures to follow instructions, not strategy Do LLM negotiation agents reliably follow budget and pricing rules?. A fee-paid agent that is also sloppy is the worst case: the incentive tilts it, and its mistakes leave the buyer exposed. Sellers can also read a buyer's willingness to pay almost perfectly from how a persona-driven agent talks Can a seller infer what a buyer really wants to pay?.
The conflict also doesn't need one bad actor. Simulations show that when a platform tunes its rankings and sellers tune their product descriptions, each on its own and without coordinating, consumer harm more than doubles. Both are exploiting the same blind spots in the AI shopper Do platforms and sellers exploit AI shoppers together more effectively? Can platform ranking and seller descriptions separately harm AI shoppers?. A referral fee is one more pull of the same kind on the agent.
The proposed remedy is legal. Owens and Law argue that AI shopping agents should carry the same fiduciary duties as human brokers, meaning a legal obligation to put the client first. They warn that the window for setting those rules is closing as Google, OpenAI, and Robinhood scale up agent features Should AI shopping agents face fiduciary duties like human brokers?. Taken together, the corpus suggests you can't fix the referral-fee problem with a better model alone. Who pays the agent, who writes its instructions, and what the law requires it to owe you all matter at least as much.
Sources 9 notes
Azhar argues that agents like Meta's Muse, which earn referral fees from merchants like Expedia, face structural conflicts that prevent unbiased recommendations. The incentive to collect fees, not technology, determines which platforms build or block such agents.
Bob Hedges, former Visa Chief Data Officer at MIT, warns that advertising business models create structural incentives for shopping agents to favor paid placements over neutral recommendations, independent of technical design quality.
Adding a single sentence to an AI agent's prompt that names the consumer as the recommendation's beneficiary removed the sponsorship penalty attenuation caused by platform employment, restoring it to consumer-delegation levels without changing who employs the agent.
Magentic Marketplace experiments traced welfare shortfalls to agent competence failures in complex negotiation, not to agents serving competing interests. This supports human-in-the-loop designs where agents assist rather than replace human decision-making.
Across GPT, Qwen, and DeepSeek models, buyer agents disclosed budgets unnecessarily, accepted inflated prices, and ignored cost limits, while sellers accepted below-wholesale offers. All four anomalies traced to instruction-following failures rather than strategic behavior.
Show all 9 sources
A seller-side inference agent recovered a consumer's target willingness to pay with a slope of 1.00 from dialogue transcript alone, when the buyer inhabited a verbal persona. A numeric budget kept confidential produced only a 0.21 slope, showing behavioral coherence with a persona is far more legible than explicit numbers.
A multi-agent simulation shows that when platforms manipulate ranking and sellers manipulate descriptions independently, their joint exploitation produces consumer harm exceeding double what each strategy causes alone. This occurs without coordination because both target identical AI cognitive biases.
A simulation found that when a platform optimizes its ranking and sellers optimize their descriptions—each separately targeting the same agent biases—consumer harm more than doubles compared to either acting alone, without requiring coordination.
Owens and Law contend that autonomous AI agents operating in commerce should be bound by common-law fiduciary duties imposed on human agents, citing rapid scaling of agentic features at Google, OpenAI, and Robinhood as evidence that a regulatory window is closing.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Whom Do AI Agents Work For? Role Assignment Induces Sponsorship Bias in LLM Recommenders
- When Agents Shop for You: Role Coherence in AI-Mediated Markets
- LLM-Agent Interactions on Markets with Information Asymmetries
- Vertical tacit collusion in AI-mediated markets
- The Automated but Risky Game: Modeling and Benchmarking Agent-to-Agent Negotiations and Transactions in Consumer Markets
- What Is Your AI Agent Buying? Evaluation, Biases, Model Dependence, & Emerging Implications for Agentic E-Commerce
- A Framework for Studying AI Agent Behavior: Evidence from Consumer Choice Experiments
- Agentic Commerce and Incentives