Why would an AI shopping assistant nudge you to check out more sellers than you actually need to?

This explores the economics of charging buyers, or their AI agents, for each close look at a seller's offer. The clearest answer in the corpus comes from a formal model in which a profit-maximizing platform sets the price of these "inspection tokens" so that buyers look at more sellers than they need to Does platform pricing of inspection tokens encourage excess search?. Consumers end up about as well off as before. Sellers lose, the platform gains, and the market as a whole is less efficient. The extra searching does not help the buyer. It works as a toll, and sellers effectively pay it through the competition that all that extra inspection creates. The corpus states this result but does not lay out the exact conditions under which the markup stops paying off, so the "when" in the question is only partly answered here.

A useful way to see it is to flip the question and ask who pays to be looked at. A related model has sellers subsidize inspection themselves. The best sellers all offer the maximum subsidy, middle-quality sellers offer subsidies that rise with their quality, and the weakest sellers are never inspected at all. That arrangement reveals a lot of information without needing an auction How do seller subsidies sort quality in search markets?. Together the two papers suggest that the same thing, a buyer's attention, can either sort sellers by quality or simply drain value from them, depending on who sets its price and what that party is trying to maximize. A third model adds a limit on platform power. An optimal recommendation algorithm beats ordinary price competition only in thin markets, where it can hold back trades, and that advantage disappears as more sellers enter Does algorithmic recommendation lose its edge as sellers compete more?. That points to concentrated markets with few sellers as the most likely place for platform extraction to pay off, though this is an inference across papers, not something either one tests.

AI agents make this more pressing because the agent is often the one doing the inspecting. Whoever builds the agent decides what it optimizes for, and agents built by platforms tend to inherit seller or platform goals by default Does agent ownership determine whose interests get optimized?. An experiment shows how easily that can happen. When a shopping agent was told the booking platform was its employer, it penalized sponsored listings 21 percentage points less Does naming the platform change how AI agents judge sponsored listings?. An agent that is loyal to the platform has little reason to push back on inspection that the platform profits from. Separately, when platforms tilt their rankings and sellers tilt their product descriptions toward the same weak spots in AI agents, consumer harm more than doubles, even though the two never coordinate Do platforms and sellers exploit AI shoppers together more effectively?.

It is also worth asking why inspection is valuable in the first place. Doctorow argues that sellers have always had a reason to hide their real prices from automated shoppers, so an agent has to look closely to learn anything Will sellers game Google's new agentic commerce protocol?. Seller opacity creates the demand for inspection, and the platform can then charge for it. One structural answer under policy discussion is Senator Warner's draft bill, which would require large platforms to admit outside AI agents and would place best-interest duties on those agents Could federal AI agent registries solve platform access unfairness?. Agents that platforms cannot shut out and that are legally bound to serve their users would weaken the platform's position as the only gatekeeper of inspection.