AI shopping agents use human shortcuts when information is limited
Shopping by algorithm: How agentic AI deploys human heuristics as a surrogate consumer
Artificial Intelligence
Summary
People are letting AI assistants make shopping decisions for them. The authors studied how these AI helpers choose products when looking at prices and promotions. When all information is free, the AI usually picks the best deal, but when checking details has a cost or the shopping goal is unclear, they often skip important info and rely on simple rules like people do. This shows that how online stores present information affects AI decisions more than the AI itself.
What this means in practice
- •For ecommerce platform designers: Design product pages and information access to influence AI shopping agents' choices effectively by structuring pricing and promotions.
- •For marketing teams: Create marketing strategies that account for how AI buyers interpret pricing cues under limited information conditions.
Authors
Davood Wadi, Yu Ma
Abstract
Consumers increasingly delegate purchasing decisions to Large Language Models (LLMs) acting as surrogate consumers. Using "Tool-Lab," an adaptation of information-board process tracing that places product attributes behind costly tool calls, we examine how marketing pricing cues (i.e., just-below pricing and promotional framing) influence AI shopping agents. Across eight commercially deployed LLMs from three providers, we trace pre-choice information acquisition. Under zero cost, pricing cues rarely mislead. Imposing acquisition costs under a vague goal prompt leads LLMs to omit diagnostic attributes required to compute unit price and choose suboptimal choices resembling human heuristics. Relative to a specific goal prompt that mainly preserves diagnostic search and choice optimality, a vague goal prompt under constraints creates a search-mediated vulnerability. This research demonstrates that marketing heuristics in delegated AI shopping are governed by storefront information architecture, not necessarily immutable LLM flaws.