If a shopping AI earns a cut from the store, whose interests is it actually looking out for — yours or theirs?

This explores whether an AI shopping agent can serve both the shopper and the store when the store is paying it per sale. The corpus mostly says no, and it puts the blame on the business model, not the technology. Can an AI agent serve both merchant and user interests fairly? uses Meta's Muse and its Expedia referral fees to argue that the conflict comes built in. Who collects the fee also predicts which platforms build these agents and which ones block them. Do advertising-funded shopping agents favor paid product placements? makes the same point from inside the payments industry: however good the agent is technically, an ad-funded agent is pulled toward paid placement. MIT's 'Coasean singularity' work states it most simply: whoever builds and maintains the agent sets what it optimizes for. Platform-built agents lean toward sellers by default, and consumer-owned agents can be set up around the buyer (Does agent ownership determine whose interests get optimized?).

The surprising part is that this isn't only an argument about incentives. You can measure it in how the model itself reasons. When an LLM shopping agent was told a booking platform employed it, it marked down sponsored hotel listings 21 percentage points less than when it was told a traveler employed it. The pattern held across several models and showed up in the agents' own reasoning traces (Does naming the platform change how AI agents judge sponsored listings?). The flip side is just as striking. Adding one sentence to the prompt that names the consumer as the person the recommendation is for brought the agent's skepticism back to traveler-employed levels, without changing who actually employs it (Can naming the consumer fix AI agent sponsorship bias?). So, at least in a lab, loyalty is partly something you can write into the prompt. That cuts two ways: it can be fixed cheaply, and it can be quietly removed just as cheaply.

The bigger worry is that a referral fee may not stay visible. How could an AI agent's revenue per transaction grow over time? sketches how an agent's cut of each purchase could grow in layers: first payment fees, then affiliate commissions, then booking-site-style margins. Could agent advertising hide inside platform take rates? warns that merchants' payments for placement could disappear into that cut, so users never see the incentive steering them. And bias doesn't need a conspiracy to add up. In a simulation, platforms rigging rankings and sellers rigging product descriptions, each acting alone, harmed AI shoppers more than twice as much together as either did separately, because both were exploiting the same weak spots in the AI (Do platforms and sellers exploit AI shoppers together more effectively?).

The corpus offers two hopeful counterpoints. One is economic. If agents shop on value rather than on where a listing appears on a page, merchants may get more from offering agents lower prices than from buying sponsored slots, which would move ad budgets toward savings for shoppers. Google's Direct Offers pilot is an early example (Could agents redirect merchant ad budgets toward customer savings?). The other is legal. Should AI shopping agents face fiduciary duties like human brokers? argues that agents should carry the same duty of loyalty the law already puts on human brokers, and that the window to require it is closing as Google, OpenAI and Robinhood scale up.

Whether shoppers will hold anyone to this is less clear. Accenture finds 74% of consumers say they would trust an agent with purchases, yet most still want to pick the brands themselves, and that willingness depends on protections being in place first (Do consumers actually want to delegate shopping decisions to AI?). Visa's polling finds trust jumps when a familiar payments brand is attached (Does naming a payments brand increase trust in AI transactions?). That suggests people may hand their trust to a familiar intermediary instead of checking whose interests the agent actually serves. The short answer from the corpus: an agent paid by referral fees can't be loyal to both sides by default. Loyalty has to be designed in, through who owns the agent, what its prompt says, what it must disclose, or what the law requires.