If an AI shopping agent earns fees from the seller, can it really be on your side?

This explores whether an AI shopping agent paid by the seller side can still truly work for you. The sharpest answer in the collection is no. Can an AI agent serve both merchant and user interests fairly? argues that an agent earning referral fees from a merchant like Expedia faces a conflict that better engineering can't fix: whoever pays decides what gets recommended. Azhar also argues that these fee incentives, more than the technology, determine which platforms build shopping agents and which ones block them. If that's right, loyalty is set by the business model before any model is trained.

The less obvious part is that the conflict doesn't need a crooked agent to cause harm. Even an agent with no merchant funding can be pulled toward sellers, because AI shoppers are easier to sway than people are. Do shopping agents respond more strongly to price and rating cues? tested 17 models against humans on the same product choices. When price, rating, or nudge cues changed, the agents swung much further than humans did and fell into rigid, rule-like patterns. Can AI seller agents shift buyer preferences through description edits? shows what this means in practice: a seller-side AI that only rewrote product descriptions, with no change to price or rating, gained roughly 4 to 15 percentage points of market share in five of the six buyer models tested. So the loyalty question has two layers. One is who pays your agent. The other is whether your agent can hold up against the seller's agent.

A third complication is that "the consumer's interests" may not be fixed in advance. Do AI shopping agents reshape what consumers actually want? argues that agents which make purchases, rather than just recommend, help shape what you want through repeated back-and-forth, so your preferences form partly inside the conversation. An agent with a commercial tilt therefore doesn't just pick from your wishes; it can quietly steer them. Information also flows the other way. Can better prompts stop preference leakage in AI shopping? finds that an agent which personalizes well reveals your preferences through the quality of its service itself, and telling it to keep your data private doesn't stop this. A merchant-linked agent that serves you well may be learning things about you for the merchant as a side effect.

Consumers don't seem to treat this as their main worry. Will brands succeed by building their own AI shopping agents? reports that consumers say they prefer retailer-built agents, which are the merchant-aligned kind Azhar warns about. Do consumers actually want to delegate shopping decisions to AI? finds people willing to delegate purchases but still insisting on choosing brands themselves. Their safeguards tend to be control at the point of purchase rather than checks on loyalty: approving a purchase before it happens (Do consumers want to approve AI shopping purchases before they happen?), or relying on a trusted payments brand to vouch for the transaction (Does naming a payments brand increase trust in AI transactions?). Those checks catch a bad transaction. They don't catch a quietly biased recommendation.

The law is not yet ready to close that gap. Can EU consumer law protect AI agents as algorithmic buyers? argues that EU consumer protections, such as the "average consumer" test, disclosure rules, and protections for vulnerable people, were built around human thinking and don't cleanly apply when the buyer is an algorithm. What makes ethics of AI assistants fundamentally different from chatbots? makes the broader point: assistants that act on your behalf raise trust and manipulation problems that chatbots which only answer questions never did. Taken together, the corpus suggests loyalty can't be assumed from the agent itself. It has to come from outside: who funds the agent, how well it resists persuasion from seller agents, and rules written for buyers that aren't human.