Generative AI as a Strategic ESG Consultant: A Comparative Analysis of Priorities in E‐Commerce Circular Logistics

Generative artificial intelligence (GenAI) is increasingly being explored as a tool for strategic decision‐making, but its reliability remains largely undefined. This study presents an innovative framework for empirically evaluating the consistency and practicality of large language models (LLMs) serving as strategic environmental, social, and governance (ESG) consultants. We focus on the field of circular logistics in e‐commerce, which is critical to sustainable development. Using the fuzzy analytic hierarchy process (FAHP) methodology, we assigned two distinct roles to four leading LLM models: first, to generate and prioritize an “ideal” set of circular logistics strategies, and second, to act as research analysts, extracting and classifying “real” corporate priorities from a corpus of public ESG reports. Our main results reveal significant variation in outcomes, representing a kind of “consultant‐analyst paradox”. When generating ideal strategies, LLM models showed significant inconsistency. However, when performing structured analyses of corporate reports, the same models showed exceptionally high internal consistency (Spearman's rank correlation coefficient, ). Furthermore, we estimated a significant “AI Strategy Gap” between ideal AI recommendations and identified corporate priorities ( ranging from 0.00 to 0.60). We concluded that the reliability of LLM models is highly task‐dependent.