What question did this study set out to answer?

The study aims to investigate how large language models affect brand differentiation during consumer searches.

April 11, 2026Open Access

Spectral Metamerism in AI-Mediated Brand Perception: How Large Language Models Collapse Multi-Dimensional Brand Differentiation in Consumer Search

Key Points

The study aims to investigate how large language models affect brand differentiation during consumer searches.
Introduced the LLM-as-observer model within Spectral Brand Theory's framework.
Conducted nine runs involving over 21,600 API calls across 24 models in 11 languages.
Used structured prompts with the PRISM-B instrument to elicit implicit spectral weights.
Computed a Dimensional Collapse Index (DCI) to measure brand collapse across different models.
Mean DCI significantly surpasses the uniform baseline with global (0.291) and local (0.353) measures.
Local brands collapse 25% more than global brands, highlighting an Economic default mechanism.
Western models exhibit lower collapse compared to non-Western models with significant p-values.
Providing brand function specifications reduced dimensional collapse by about 20%.

Abstract

Large language models (LLMs) are rapidly replacing traditional search engines as the primary intermediary in consumer brand evaluation. This paper introduces the LLM-as-observer model within Spectral Brand Theory's eight-dimensional framework and demonstrates that LLMs systematically collapse brand perception to Experiential and Economic dimensions, rendering brands differentiated on narrative, ideological, cultural, or temporal grounds metameric – structurally distinct yet functionally equivalent in AI recommendations. Across nine runs (21, 600+ API calls, 24 models from seven training traditions, 11 languages), we elicit implicit spectral weights via structured prompts using the PRISM-B instrument and compute a Dimensional Collapse Index (DCI). Mean DCI significantly exceeds the uniform baseline (global = 0. 291, p = 0. 017; local = 0. 353, p = 0. 0006; cross-cultural = 0. 357, d = 3. 449). Cross-model convergence is extreme (cosine similarity of spectral profiles = 0. 977). Local brands collapse 25% more severely than global brands (d = 0. 878), consistent with an Economic default mechanism. Western models exhibit lower collapse than non-Western models (p = 0. 0013). Native-language prompting across 11 languages produces no aggregate benefit (46/115 model-pair combinations positive; mean reduction = -0. 005). Geopolitical framing significantly modulates collapse (H12: mean absolute delta = 0. 040, p < 0. 0001). The single exception – Patagonia versus Columbia – shows that legally verifiable ideological commitments survive AI mediation. Providing structured Brand Function specifications reduces collapse by approximately 20%. These findings establish dimensional collapse as a structural feature of text-based AI observers, with implications for search advertising strategy, brand defensibility, and dual-track (human + AI) measurement. Includes paper. yaml (Paper Spec v0. 1. 0) – a machine-readable specification of the paper's claims, assumptions, and dependencies. See https: //github. com/spectralbranding/paper-spec for the standard.

Spectral Metamerism in AI-Mediated Brand Perception: How Large Language Models Collapse Multi-Dimensional Brand Differentiation in Consumer Search

Key Points

Abstract

Cite This Study