When a shopping AI can scan everything for you, the real work shifts to explaining exactly what you want.
Why does preference articulation replace manual comparison in agentic search?
This explores why, when an AI agent does your shopping or searching for you, the main work moves from comparing options yourself to telling the agent clearly what you want, and what that change costs and makes possible.
This explores why, when an AI agent does your shopping or searching for you, the main work moves from comparing options yourself to telling the agent clearly what you want. The clearest answer in the corpus is a formal model of product search Why does manual search fail as product choices become complex?. When you search by hand, each extra thing you care about (price, fit, material, shipping, ethics) multiplies the number of side-by-side comparisons you have to make. Past a certain point manual search simply collapses: people give up, settle, or stick to one criterion. An agent can scan a huge catalog cheaply, so the size of the catalog stops being the limit. The new limit is how well you can describe what you want. Your effort moves from browsing widely to sharpening your request, and the cost that matters becomes articulation cost.
That raises the obvious follow-up: do people already know what they want well enough to say it? The corpus suggests often not. Bilgihan and colleagues argue that agents which actually complete purchases end the consumer's role as the only one forming their preferences Do AI shopping agents reshape what consumers actually want?. Preferences come out of a back-and-forth: the agent proposes, you react, the agent learns, and your trust shapes what you ask for next. So 'preference articulation' isn't one statement made at the start. It's a conversation, and the agent helps write your wants as well as carrying them out.
There are two ways to lower the cost of saying what you want. One is to have the agent ask better questions. Work drawing on conversation analysis describes when an agent should stop and check with you (to clarify intent, narrow scope, or confirm what appeals) instead of silently chaining searches and drifting away from what you meant When should AI agents ask users instead of just searching?. The other is to skip asking altogether: memory systems that watch a user over time and build a picture of the people and things they care about can infer preferences from behavior Can agents learn preferences by watching rather than asking?. Each lowers the articulation burden in a different way, and each carries its own risk. One risks annoying you with questions. The other risks quietly getting you wrong.
The shift also changes who benefits and when. A companion model shows that platforms come out ahead with agentic search before consumers do, because platforms count only conversions while consumers also pay the cost of describing what they want. The result is a rational delay in consumer adoption Why do consumers delay adopting agentic search?. And handing comparison to the agent doesn't make it neutral. Shopping agents show stubborn position bias, favoring items by where they appear in a list even in text-only interfaces and even when told to ignore position Do AI shopping agents show persistent position bias in text interfaces?.
The takeaway you may not have expected: once agents take over comparison, the scarce skill is knowing and saying what you want, not finding things. That skill is now shared with a system that has its own biases and its own business incentives. Whoever shapes the preference conversation shapes the purchase.
Sources 6 notes
A formal model shows manual search collapses as preference complexity rises, but agentic search avoids that collapse by shifting effort from browsing breadth to refining preference signals. The binding constraint becomes articulation cost, not catalog size.
Bilgihan and colleagues argue that transactional AI agents—which execute purchases rather than just recommend—end the consumer's sole cognitive role in preference formation. Preferences emerge through bidirectional feedback loops between agent learning and consumer trust, not independently beforehand.
Tool-enabled LLMs drift from user intent through silent tool chaining. Conversation analysis reveals insert-expansions—clarifying intent, scoping responses, enhancing appeal—as a formal framework for proactive user consultation that prevents misunderstanding instead of recovering from it.
M3-Agent demonstrates that separating episodic events from semantic knowledge in an entity-centric graph, combined with parallel memorization and control processes, allows agents to infer and act on user preferences without asking. This architecture mirrors human cognitive systems that bind disparate information about individuals across sensory modalities.
A formal model shows platforms rank agentic search by conversion revenue alone, while consumers also bear search costs. This divergence creates rational adoption lags where consumers keep manual search even after it would boost platform revenue.
Show all 6 sources
ACES found that AI agents exhibit strong, model-specific position biases even in headless text-only interfaces with explicit instructions to ignore position. The bias is a deep-seated model characteristic, not a superficial visual artifact, and recurs after each model update.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- From Product Search to Preference Articulation: The Economics of Agentic Commerce
- What Is Your AI Agent Buying? Evaluation, Biases, Model Dependence, & Emerging Implications for Agentic E-Commerce
- Shopping by algorithm: How agentic AI deploys human heuristics as a surrogate consumer
- Agentic markets
- Vertical tacit collusion in AI-mediated markets
- When Agents Shop for You: Role Coherence in AI-Mediated Markets
- A Framework for Studying AI Agent Behavior: Evidence from Consumer Choice Experiments
- The AI Agent in the Room: Rethinking Consumer Decision-Making in the Age of Autonomous AI