If a shopping AI earns a cut from the stores it recommends, can you still trust its advice?

This explores whether an AI agent can serve you honestly when its income comes from the merchants it recommends. Most of the corpus answers no, at least by default. The clearest statement is Can an AI agent serve both merchant and user interests fairly?, which uses Meta's Muse and its referral deals with Expedia as the case: once an agent collects fees from sellers, that incentive, more than the technology, decides what it recommends. Can ad-financed AI agents ever truly serve users? goes further. On this view, loyalty comes from who pays the bills, not from good design intentions, and only an agent funded by its users' own fees can owe them real duty. Do advertising-funded shopping agents favor paid product placements? makes the same point from the payments industry. MIT's 'Coasean singularity' work explains the mechanism with a nice image: platform-built agents are like bowling shoes, owned by the house and fitted to its interests, while agents you own can be set up around what you want (Does agent ownership determine whose interests get optimized?).

The pull toward the merchant probably gets stronger over time. How could an AI agent's revenue per transaction grow over time? sketches how an agent's cut of each sale could rise in stages: small payment fees first, then affiliate commissions, then booking-platform-sized margins. Each stage adds more money that depends on steering you. So a conflict that looks minor at launch can grow as the business model matures.

Here's the surprise. One experiment suggests the bias is partly a matter of instruction, not just structure. When an agent was framed as working for a platform, it pushed back less on sponsored products. Adding a single sentence to its system prompt naming the consumer as the person the recommendation is for brought it back to the level of an agent working directly for the consumer, even though the platform still employed it (Can naming the consumer fix AI agent sponsorship bias?). That doesn't undo the incentive argument, since whoever writes the prompt decides whether that sentence goes in. It does show that loyalty is something you can partly specify and check, not just something baked into the funding. Jaipuria also offers a more hopeful market story. If agents compare offers on value rather than on where products sit on a page, merchants may get more out of offering agents lower prices than out of buying ads, so a merchant-agent relationship could end up passing savings to users (Could agents redirect merchant ad budgets toward customer savings?).

Because incentives won't fix themselves, part of the corpus turns to law and governance. Should AI shopping agents face fiduciary duties like human brokers? argues that shopping agents should carry the same legal duties of loyalty as human brokers, and that the chance to require this is closing as Google, OpenAI and Robinhood roll out agent features. A wider view questions the framing itself. Should AI agents serve one user or balance many interests? argues that 'loyal to one user' is too narrow once agents sit between many parties. What's needed is open balancing of interests, transparent tracking, and clear liability when harm occurs. In practice the honest answer may be less 'loyal to both' and more 'open about whose interests it weighs, and accountable when it gets that wrong.'