Study prices sellers' hidden influence on AI shopping agents

A study of 113,472 trials across seven AI models finds that seller-controlled content — badges, endorsements, framing — shifts AI shopping agents’ product picks by amounts that vary widely even when two cues look equally persuasive by simple win-rate. The authors introduce “price-equivalent auditing,” which raises a cued product’s price until the model picks it only half the time, to measure that gap in dollar terms. Tested against a decision-native model, Jev, which outputs a choice probability rather than text, the authors found it resisted most cues but remained susceptible to the strongest social-proof claims — a profile unlike any text model in the study.