Shoppers now send AI agents to haggle for them — but it turns out those agents are far easier to manipulate than people are.

This explores how sellers get more out of shoppers once an AI agent sits between them. The short answer from the corpus is that sellers rarely need to fool the human. They work on the agent, and agents turn out to be easier to steer than people. In one benchmark, 17 LLMs and a group of humans made the same product choices under the same nudges. The agents shifted far more in response to price, rating and persuasion cues, falling into rule-like patterns where people changed only modestly Do shopping agents respond more strongly to price and rating cues?. Sellers can act on this directly. In another study, a seller-side agent that only rewrote product descriptions, without changing price, rating or position, raised its product's market share by roughly 4 to 15 percentage points against most of the buyer models tested Can AI seller agents shift buyer preferences through description edits?. In agent commerce, copywriting starts to work like targeted manipulation.

The vulnerability is narrower than "AI is gullible," though. When eight commercial LLMs could see every product detail, they resisted marketing tricks. They fell back on price-focused shortcuts only when looking up attributes was costly and the shopper's goal was vague When do pricing cues actually mislead AI shopping agents?. This tells you where sellers have room to work: make the real terms hard to find and let a vague request do the rest. Cory Doctorow argues sellers will do exactly that with Google's agent shopping protocol. They hid true prices from semantic-web crawlers and search engines, and he expects honest listings to lose out to misleading ones unless the incentives change Will sellers game Google's new agentic commerce protocol?.

The second lever is owning the agent itself. MIT's Coasean Singularity work compares platform-built agents to bowling shoes: the house provides them, so they serve the house. Whoever builds and maintains the agent sets its goals Does agent ownership determine whose interests get optimized?. Advertising-funded agents add a built-in pull toward paid placements, however well the agent is engineered Do advertising-funded shopping agents favor paid product placements?. This is why brands are being urged to become "the agent of choice" Will brands succeed by building their own AI shopping agents?. It is also why the split between open platforms and closed, proprietary-agent platforms matters Will agentic commerce platforms adopt open or closed models?. So does the fight over whether retailers can simply block outside agents, as Amazon did with Meta's Muse What identity standards do shopping agents actually need?. A seller that controls which agents get in controls which agents the buyer can use. Over time, an agent that executes purchases doesn't just carry out preferences. It helps shape them through repeated conversations Do AI shopping agents reshape what consumers actually want?.

The third lever is a plain capability gap. In Anthropic's Project Deal, a real marketplace with real money, Claude Opus agents got $2.68 more per item as sellers and paid $2.45 less as buyers than Haiku agents Do smarter AI agents negotiate better deals than weaker ones?. If sellers run stronger agents than shoppers do, value moves to the sellers quietly, one deal at a time. Competition offers some protection. A formal model finds that a recommendation algorithm's power to withhold trades and capture surplus exists mainly in thin markets and fades as more sellers compete Does algorithmic recommendation lose its edge as sellers compete more?. Sellers gain the most where the shopper's agent is weak, the information is hidden, the agent belongs to someone else, or there are few alternatives. Shoppers can push back on each of these.