Dimensional Collapse in AI-Mediated Search: Large Language Models as Metameric Observers of Brand Advertising
Year: 2026
Type: preprint
Abstract: AI-mediated search is replacing traditional consumer search, yet no formal model predicts which brand attributes survive AI mediation. This paper applies Spectral Brand Theory's eight-dimensional framework to model large language models as observer cohorts with systematically biased spectral profiles. An experimental study prompts six LLMs (Claude, GPT-4o, Gemini Flash, DeepSeek V3, Qwen3 30B, Gemma 4) with brand comparison queries across fifteen brand pairs: ten globally recognized pairs (Run 2, 3,240 API calls) and five locally embedded pairs from small non-English-speaking markets (Run 3, 1,620 API calls). A Brand Function resolution test (Run 4, 90 calls) demonstrates that providing behavioral specifications reduces local-brand collapse by 20%. Results provide evidence that LLMs systematically collapse brand perception to Experiential and Economic dimensions (H1 supported: global mean DCI = 0.291 vs. baseline 0.250, p = 0.0170; local mean DCI = 0.353, p = 0.0006). Cross-model convergence is extreme: cosine similarity of spectral profiles across six architectures is 0.975 (H2 supported), indicating that dimensional collapse is structural rather than model-specific. Cross-model heterogeneity is non-significant (H3 not supported), and collapse is approximately uniform across all soft dimensions rather than differential (H4 not supported). The conditional metamerism thesis is supported: local brands collapse significantly more than global brands (t = 6.483, p < 0.0001, d = 0.878), consistent with an Economic default mechanism in which models substitute price for all dimensions they lack data on. The single exception – the Patagonia/Columbia pair, the only pair below baseline DCI – demonstrates that legally and operationally grounded ideological commitments survive AI mediation, providing empirical support for the brand defensibility thesis. Implications for search advertising strategy, AI brand optimization, and dual-track brand measurement are discussed.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.
Author: Zharnikov, Dmitry
Language: English
Subfield: Marketing
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7149358937