D. Żatuchin
This paper proposes a new set of metrics and an empirical framework to quantitatively analyze the impact of LLM recommendations on brand competition dynamics.
As large language models become key mediators in the consumer purchase journey, the competitive structure of AI-generated recommendations has become a strategic concern for brands. However, there has been a lack of large-scale empirical answers to the basic question of which brand a model recommends in a given category and how concentrated that ownership is.
The study collected a total of 3,750 responses by querying three models (GPT-5.2, Google Gemini 3 Flash, Perplexity sonar-pro) five times each on 250 brand-free category queries across five industries. To analyze the data, the authors proposed three exploratory metrics: the Category Ownership Index (COI) to measure a brand's share of mentions within a category, the Competitive Vacuum Index (CVI) to flag categories with no single leader, and the Displacement Score (DS) to quantify asymmetric substitution between brand pairs.
The analysis found that recommendation concentration was moderate (mean Gini coefficient of 0.28), placing the results in tension with a strong winner-takes-all narrative. Furthermore, the cross-model agreement on the top-recommended brand was only 41.6%, indicating that a top position on one model does not reliably hold on another. This research contributes a candidate, reproducible procedure and metrics for competitive-intelligence analysis of AI recommendations that future work can validate.