Exploring occupational prestige through Large Language Models: A multi-dimensional approach

Debates over occupational prestige have long centred on what existing prestige scales actually measure and whether prestige constitutes a single evaluative hierarchy or multiple distinct dimensions. This paper contributes to these debates both conceptually and methodologically by examining the dimensionality of occupational prestige as represented in the semantic associations learned by Large Language Models (LLMs) from human text. Using pairwise comparisons across a representative list of occupations, we prompt multiple LLMs to generate rankings along five scales: status, prestige, social standing, expected deference, and an entirely novel dimension based on advertised social proximity (as an indicator of status-related associational preference bias). We show that status, prestige, and social standing are empirically indistinguishable in the models’ associative structure, forming a unified hierarchy closely aligned with established survey-based prestige scales. By contrast, deference and advertised proximity emerge as clearly separable dimensions. Further analyses demonstrate that these dimensions are differentially related to associations the models hold about occupational characteristics such as training and pay, authority, and social contribution. Substantively, the findings support a multidimensional conception of occupational prestige as represented in LLMs’ associative structure – and specifically suggest advertised social proximity as a novel, unexplored dimension. Methodologically, we demonstrate – while recognising their limitations – the value of LLM-based approaches as a complement to human research for exploring complex evaluative structures that are difficult to measure using conventional survey methods.