What question did this study set out to answer?

The research investigates whether corporate ownership influences brand perception as encoded in large language models.

April 16, 2026Open Access

Does Corporate Ownership Matter to AI? Portfolio Interference in Large Language Model Brand Perception

Key Points

The research investigates whether corporate ownership influences brand perception as encoded in large language models.
Conducted a preregistered experiment using 13 large language models from diverse training traditions.
Rated 20 brands across four prompt modalities including direct ratings and conversational contexts.
Collected 7,975 observations to analyze the impact of portfolio framing and brand ownership on brand perceptions.
Portfolio framing showed minimal change in brand perception as measured by the Dimensional Concentration Index.
Equivalence testing confirmed that 18 out of 20 brands displayed no significant perceptual changes.
Multi-turn revelations had limited effects only on certain brands, indicating context-dependent responses.

Abstract

Large language models (LLMs) increasingly shape how consumers discover and evaluate brands. Brand portfolio theory predicts that revealing corporate ownership should produce perceptual interference via an awareness gate (Keller 1993; Aaker and Keller 1990). Yet LLMs encode portfolio relationships permanently in their parameters, raising the possibility of either maximal interference or complete immunity. We test these hypotheses in a preregistered experiment with 13 LLMs spanning seven training traditions (Western, Chinese, Russian, Indian, Japanese, European, Korean). Twenty brands from seven portfolio archetypes (LVMH, Unilever, P&G, Toyota, L'Oreal, Geely, Yandex) were rated under four prompt modalities – direct rating, naturalistic recommendation, multi-turn conversation with mid-dialogue portfolio reveal, and native-language framing – yielding 7,975 parsed observations. Portfolio framing produces near-zero change in Dimensional Concentration Index (mean |delta DCI| = .26). Equivalence testing confirms the null for 18/20 brands within +/-1.0 DCI points. Multi-turn revelation unlocks modest flattening for reverse-aspiration structures (Geely Auto d = -1.11, FDR-significant), but effects remain portfolio- and modality-specific rather than systematic. Native-language prompts activate model-specific discourse layers without directional amplification. These results generalize spectral immunity across model traditions and portfolio types, implying that house-of-brands shielding is automatic in AI-mediated markets while constructive interference is impossible. Brand managers cannot rely on portfolio architecture to reshape LLM perceptions except in extended conversational contexts. Theoretical implications for awareness-gate mechanisms and practical implications for "Share of Model" strategy 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.

Read Full Paperexternally

Ask AI

Mark Helpful

Bookmark

Relay

View Full Paper