Can cheap agent search actually harm shopping markets?

Swiss Pricing Lab, reviewing research on AI shopping agents, notes that "economists have long assumed that when searching costs almost nothing, competition gets sharper and prices fall." It then cites a 2026 paper in a market design journal that argues "this is not automatic here" once an AI agent, rather than a person, does the shopping. The newsletter's summary gives three linked reasons the paper offers for why cheap search does not translate into sharper competition: "agents do not search perfectly, and that behaves like a hidden search cost"; "when everyone can generate polished descriptions, offers start to look alike, so real differences get harder to see"; and "when opening a shop is easy, a lot of low quality ones show up." Put together, the paper's claim is that these three effects "can leave a market stuck somewhere worse than where it started."

The reasoning is explicitly economic rather than behavioral: it is not that any single agent is gullible, but that the market-level equilibrium changes once agentic shopping is cheap and descriptions are easy to generate at scale. An imperfect searcher — even a tireless one — still fails to fully discriminate between offers, which functions like a cost even though no fee changes hands. When sellers can cheaply produce agent-optimized descriptions, the pool of offers converges toward similarity, erasing the differentiation that would otherwise let search resolve into better prices. And because entry barriers fall along with search costs, low-quality sellers enter the market too, diluting the average quality an agent has to sort through. None of the three mechanisms requires bad faith; they compound as structural side effects of agentic commerce becoming cheap and frictionless.

This note sits beside the batch's behavioral experiments rather than restating them. Can AI seller agents shift buyer preferences through description edits? and Do shopping agents respond more strongly to price and rating cues? show that individual agents are exploitable at the level of a single purchase decision. The market design paper's argument, as relayed here, is the aggregate consequence of that exploitability becoming common practice across many sellers at once: if everyone learns to write agent-flattering descriptions, the market-wide effect is not better-informed matching but convergence and adverse selection. It also complicates the advice in Can platform ranking and seller descriptions separately harm AI shoppers?, a companion note from the same source, by supplying the standard economic theory (cheap search sharpens competition) that the field is finding does not hold once agents are the searchers.

The excerpt gives no description of the paper's model, data, or the size of the welfare loss it derives — only the three mechanisms and the summary conclusion that markets can end up "worse than where it started." Swiss Pricing Lab does not name the paper, its authors, or the journal, so the claim cannot yet be checked against the underlying analysis or weighed against competing market-design accounts of agentic search. The excerpt's own closing caveat — that this body of work is largely preprints and simulated stores, "early signals, not settled facts" — applies here too: the argument is a theoretical claim about equilibrium behavior, not a measurement of any real marketplace.