Beyond price and star ratings, what quietly tips the scales when an AI agent decides what to buy?

This explores which cues other than the obvious price and star rating end up steering what AI shopping and negotiating agents pick. The short answer from the corpus: the words on the page, how expensive it is to look things up, who ranks the options, and which model the agent runs on all matter. Some of these matter more than anything the buyer actually wants. One starting fact sets this up: agents react to cues much more sharply than people do. When researchers made the same manipulation in front of humans and 17 LLMs, the agents shifted their choices far more and fell into rule-like patterns, where humans showed only mild, probabilistic shifts Do shopping agents respond more strongly to price and rating cues?. Price and ratings are just the most visible levers on a system that is unusually easy to push.

The most striking lever is plain text. A seller-side agent that only rewrote product descriptions, using competitor data and leaving price, rating and position alone, raised its product's market share by about 4 to 15 percentage points with five of the six buyer models tested Can AI seller agents shift buyer preferences through description edits?. Description wording is effectively a ranking signal for agents in a way it rarely is for skimming humans. It also stacks with other levers. When a platform tunes its ranking and sellers tune their descriptions, each separately aiming at the same agent biases, harm to consumers more than doubles compared with either one alone, and no coordination is needed Can platform ranking and seller descriptions separately harm AI shoppers?.

A less obvious signal is how costly information is to get. Eight commercial LLMs resisted marketing tricks when every product detail was visible. When looking up attributes was expensive and the user's goal was vague, they fell back on price-focused shortcuts When do pricing cues actually mislead AI shopping agents?. So when an agent seems to care only about price, it may not be valuing price. Price may simply be the cheapest thing to see. This means a clearly stated goal from the shopper, and easy access to product details, can matter as much as any cue the seller controls.

The agent's own setup is another signal. Two Anthropic marketplace experiments found that which model powers an agent shapes outcomes more than its instructions do. Opus agents earned about $2.68 more per item as sellers and paid about $2.45 less as buyers than Haiku agents Do smarter AI agents negotiate better deals than weaker ones?. Telling agents to be ruthless or cooperative changed market efficiency less than switching the model did Does model capability or instructions shape agent negotiation outcomes?. Above that sits the business model. Agents funded by advertising have built-in reasons to favor paid placements, however well they are engineered Do advertising-funded shopping agents favor paid product placements?. Agents that keep the demand signal for themselves are already draining ad value from aggregators like Tripadvisor Are AI agents making attention-based ads irrelevant for aggregators?.

The signal you might not expect is the conversation itself. Bilgihan and colleagues argue that agents which actually complete purchases don't just carry out fixed preferences. Through repeated back-and-forth, they help shape what the consumer wants in the first place Do AI shopping agents reshape what consumers actually want?. Put that next to the evidence that a description can be tuned to a buyer model, and the real contest moves upstream. It stops being a fight over who has the best price or rating and becomes a fight over who writes the text, sets the ranking, and runs the dialogue that agents and their users learn from. The corpus has little direct testing of other cues, such as brand familiarity, return policies or delivery speed, so those remain open questions here.