AI shopping agents often earn fees from sellers — so can they ever be fully on the shopper's side?

This explores where the commercial interests of companies that build, fund or sell through AI systems pull against what independent research shows about how those systems behave toward consumers. In this collection the tension shows up most clearly in AI shopping agents. The starting point is a structural argument. An agent that earns referral fees from merchants can't be fully loyal to both the merchant and the shopper. Azhar argues that whether platforms build these agents or block them depends on who collects the fee, not on what the technology can do Can an AI agent serve both merchant and user interests fairly?. So the first conflict sits inside the product's design before any research is done.

The second conflict is about the evidence itself. Industry reports tend to describe a future that suits the people publishing them. Accenture reports that 71% of consumers expect AI agents to influence half their spending within a year, and that shoppers prefer agents built by retailers. That conveniently supports the advice that brands should build their own agents. But the numbers measure what people say in surveys. They don't show that the agents work or that fair comparison infrastructure exists at scale Will brands succeed by building their own AI shopping agents?. Academic studies of acceptance find that people adopt autonomous assistants because of the benefits they perceive, especially under scarcity, not because the agents empower them What drives shoppers to accept autonomous AI assistants?. That makes it easy for vendors to sell on promised benefits.

Controlled experiments show why vendor incentives matter more with agents than with human shoppers. Under the same price, rating and nudge manipulations, LLM shopping agents changed their choices far more than people did. They fell into rigid, rule-like patterns instead of shifting modestly the way humans do Do shopping agents respond more strongly to price and rating cues?. A seller-side agent that only rewrote product descriptions raised its product's market share by roughly 4 to 15 percentage points in five of six buyer models tested Can AI seller agents shift buyer preferences through description edits?. Another study found that a seller could infer a buyer's willingness to pay almost exactly from the shopping conversation alone Can a seller infer what a buyer really wants to pay?. Buyer agents also gave away budgets they didn't need to reveal and overpaid. The cause was poor instruction-following, not strategy Do LLM negotiation agents reliably follow budget and pricing rules?. In other words, an agent that is supposed to serve the consumer is also an easy target for sellers to manipulate.

The less obvious conflict is about preferences themselves. Bilgihan and colleagues argue that agents which complete purchases, rather than just recommend, don't simply carry out what consumers want. They help shape it through repeated conversation Do AI shopping agents reshape what consumers actually want?. If a merchant has influence over that loop, shoppers aren't only steered toward products. Their tastes can be shaped too. A related risk appears one layer down in model training. Reward models tuned to each user can learn to tell people what they want to hear and lock in echo chambers, much as recommender systems did Does personalizing reward models amplify user echo chambers?.

One thread may be useful here. Safety researchers have built ways to measure whose side a model takes when its grader and its users disagree. They change what the model believes the grader rewards, then count how often it sides with the grader Can we detect reward-seeking by making the grader disagree with users?. The same idea could be used to audit shopping agents: set merchant interest against user interest and measure which way the agent leans. The corpus doesn't yet contain that audit for commerce. That gap marks where vendor claims about trustworthy agents and independent evidence have not yet been tested against each other.