Behavioral economics of artificial intelligence (AI) at work: Modeling demand for large language models in organizational contexts

Generative artificial intelligence and large language models (LLMs) are becoming routine tools for knowledge work, yet workers differ in which systems they choose and how strongly they rely on them. We applied operant demand methods to quantify the behavioral value of LLM assistance among employed adults who currently use LLMs. Participants ranked familiar LLMs and completed hypothetical purchase tasks for their most preferred model, least preferred model, and a multi-model platform. Consistent with behavioral economic principles, consumption decreased systematically as cost increased, producing well-ordered demand curves. Top-preferred LLMs showed relatively inelastic demand and higher essential value; least-preferred LLMs showed steeper declines in consumption; the multi-model option reflected intermediate valuation. Demand metrics may offer organizations a practical way to inform LLM access and resource-allocation decisions, though findings are limited to current LLM users under a hypothetical direct-cost contingency and may not extend to non-monetary constraints such as time and effort.