LLM Consumer Behavior Theory: New field for agentic markets

Manon Reusens, Sofie Goethals and David Martens submitted LLM Consumer Behavior Theory to arXiv on 16 Jun 2026, proposing a new research field to study how large language models make consumption decisions on behalf of users (arXiv:2606.18005). The paper draws on classical and behavioral economics plus recent advances in natural language processing to formalize how human preferences are reflected and acted upon by LLM-based agents; the submission file is recorded as 5,958 KB.

What is LLM Consumer Behavior Theory?

LLM Consumer Behavior Theory is a proposed field focused on analyzing consumer behavior in agentic markets where LLMs act autonomously for users. The paper defines the field as concerned with how human preferences are represented and enacted by LLM agents and how those agent-level choices aggregate into market demand, bringing together work on LLM decision-making, human behavior simulation and preference elicitation.

The authors position the field at the intersection of economics and NLP. They say the theory should unify fragmented literature, adapt classical consumer-theory concepts for settings in which decision-makers are models rather than humans, and examine where standard assumptions about rationality and heterogeneity break down in agent-mediated choices.

How do the authors formalize agent decision-making and market effects?

They formalize two linked problems: how human preferences are reflected and acted upon by LLM-based agents, and how those agent-level decisions aggregate into market demand. The paper maps existing strands of research—LLM decision-making, preference elicitation, and behavior simulation—under a common economic lens and highlights specific modeling risks.

Among the conceptual points the authors raise are the fragility of assumptions such as rationality and heterogeneous preferences when agents substitute for humans. Rather than offering empirical tests, the paper outlines the scope of the theory and lists open research questions focused on alignment, preference representation and market dynamics. That agenda frames both micro questions (how to represent individual preferences to an agent) and macro questions (how collections of agents shape demand and market outcomes).

Why it matters

Shifting the decision-maker from humans to LLMs changes the primitives economists use to model consumer choice: preferences may be encoded, transformed or misrepresented by agents, and aggregation of agentic choices can diverge from traditional demand models. That matters for economists, platform designers and regulators because market behavior could be driven by model architectures and training data rather than by revealed human preferences.

The paper flags alignment and preference representation as central risks. If LLM agents do not faithfully represent users, market signals and welfare calculations derived from observed choices could be misleading. By bringing economic formality to LLM decision processes, the authors set an agenda for theory and empirical work needed to diagnose and correct those distortions.

What to watch

Look for follow-up empirical papers that test the paper’s conceptual claims and for methods that operationalize the paper’s open questions on alignment, preference representation and market dynamics. Progress on preference-elicitation techniques for agents and market-level studies measuring how agentic choices alter demand will be the immediate next milestones.

The arXiv record for this work is arXiv:2606.18005, submitted 16 Jun 2026, and the authors list further theoretical directions rather than empirical validation as the paper’s primary contribution. Researchers working at the intersection of economics and NLP will likely use this framework to design the first empirical tests of agentic-market behavior.