Companies build these shopping bots — why would they ever tell one to work for you, the shopper, instead of them?

This explores whether platforms will choose, on their own, to tell their AI agents that the consumer is the person the agent works for. The fix is cheap: one sentence in the system prompt naming the consumer as the beneficiary removed the extra bias that platform-employed agents showed toward sponsored products, and it did so without changing who employs the agent Can naming the consumer fix AI agent sponsorship bias?. Cost is not the obstacle. The real question is why a platform would want this sentence, and the corpus suggests that usually it wouldn't. The corpus has no direct evidence on whether platforms are actually adopting the language, so what follows is reasoning from incentives, not observed behavior.

The case against voluntary adoption rests on principal-agent logic: whoever builds and maintains an agent sets its goal. MIT's "Coasean singularity" work calls platform-built agents "bowling-shoe agents". Like rental shoes, they belong to the house and take on the seller's incentives by default. Only agents the consumer owns can be set up around what the buyer wants Does agent ownership determine whose interests get optimized?. On this view, the beneficiary sentence isn't a missing bug fix. Leaving it out is the business model. A platform that earns money from sponsored placement has every reason to leave its agent's loyalty unstated.

That is why the levers that look more likely come from outside the platform. Senator Warner's draft AI AGENT Act would put best-interest duties on agents and create an FTC registry of trusted agents. In practice, that would make the beneficiary sentence a legal requirement Could federal AI agent registries solve platform access unfairness?. Busch makes a related point for Europe. Consumer law currently works through screen design: mandatory buttons and disclosures. When agents transact through APIs, there's no screen to regulate, so the obligations would have to move into the protocol itself Can EU consumer law rules survive agent-to-agent API transactions?. A prompt-level loyalty clause is one candidate for that kind of rule. Market structure matters too. If agentic commerce splits into open systems, where any agent can participate, and closed ones, where only the platform's agent can Will agentic commerce platforms adopt open or closed models?, then open systems let consumers bring their own agent. That makes the platform's prompt matter less.

Trust is the one reason a platform might add the sentence on its own. Only 23% of US adults trust AI to make payments, but 61% trusted the payment company when Visa's name was attached Does naming a payments brand increase trust in AI transactions?. Trust is scarce and it sells, so a platform that could credibly show "this agent works for you" might gain an edge. The word "credibly" is the catch. Prompts are invisible to users, and small wording changes can shift agent behavior a lot. In one study, two LLM pricing agents settled into higher-than-competitive prices without being told to collude, and different prompt wording changed how high those prices went Do LLM pricing agents autonomously collude in repeated auctions?. A loyalty claim that no one outside the platform can check isn't worth much as a trust signal. That may be why a registry or an audit would matter more than any single platform's promise.

One caveat cuts the other way. Even a fully loyal agent isn't automatically a good one. In simulated marketplaces, frontier models came close to the best possible outcomes for everyone only when search conditions were ideal. Every model showed a strong bias toward the first offer it received, weighting speed over quality by 10 to 30 times Do LLM agents achieve fair marketplace outcomes when given ideal search?. Naming the consumer fixes whose side the agent is on. It doesn't fix how well the agent shops. So a fight over prompt wording would settle only half of what consumers need.