LLMs Misread Luxury Brand Cues, New Research Shows

As large language models increasingly influence consumer purchasing decisions, luxury brands face a critical challenge: AI systems fundamentally misunderstand the implicit cues that create desirability among human buyers.

New research from INSEAD and Jellyfish tested how three leading LLMs—ChatGPT 5.1, Claude Sonnet 4.5, and Gemini 3 Pro—respond to established luxury marketing signals. The findings reveal a significant perception gap between human and machine interpretation of brand value.

Why it matters

With AI agents mediating more purchase decisions, luxury brands risk losing visibility and having their premium positioning diluted in algorithmic recommendations. A Ferrari may rank alongside a BMW in AI-generated advice, eroding carefully cultivated brand hierarchies worth billions in market value.

The implicit cue problem

Researchers tested four research-backed luxury signals known to boost human desirability: higher physical positioning, association with art, spacious display, and slender design. Each model evaluated 150 samples across categories including jewelry, watches, and beauty products.

The results showed stark differences. While AI reliably processed explicit markers—brand names, stated prices, direct "luxury" claims—implicit signals failed to register. Products positioned higher in images did not gain prestige. Art associations carried less weight than for humans, with models preferring celebrity photographs over paintings. Minimalist white space, which humans associate with exclusivity, actually reduced perceived value in AI evaluations.

Brand hierarchy collapses

A second experiment examined six automotive brands (Alfa Romeo, BMW, Ferrari, Mercedes, Porsche, Tesla) displayed against simple versus luxurious backgrounds. Across 5,400 AI evaluations, the traditional luxury hierarchy disappeared. Premium and luxury brands received similar valuations, with Ferrari and BMW perceived as equally desirable.

More troubling: context effects varied wildly by model. When researchers showed a Ferrari beside a Van Gogh painting, Gemini remained indifferent, ChatGPT reduced its valuation, and Claude increased it. A Porsche in luxury settings saw reduced willingness-to-pay across all three models, while Mercedes benefited consistently.

A new optimization playbook

The research team, led by INSEAD associate professor David Dubois, recommends luxury brands adopt model-specific strategies across the traditional marketing mix:

Product positioning requires stress-testing asset inventories for AI readability. Brands should score craftsmanship descriptions, provenance claims, and design language for machine interpretation, then develop AI context briefs specifying use cases and moments that heighten relevance.

Pricing perception demands willingness-to-pay experiments across models. The same luxury item may be labeled "premium" by one system and "overpriced" by another, requiring targeted corrections where systematic undervaluation occurs.

Promotional content must anchor functional features explicitly. In a separate analysis, AI interpreted the rigidity of Atomic skis—valued by the skiing community—as a negative attribute, reducing recommendation likelihood.

Placement strategy becomes critical since AI infers meaning from surrounding information. In beauty category analysis, branded websites accounted for only 20 percent of LLM citations, with the remaining 80 percent drawn from e-commerce listings, news media, specialist blogs, and user-generated content.

The researchers emphasize that third-party content—reviews, marketplace listings, comparison articles—now forms the front line of luxury positioning in AI-mediated search. Brands must audit and optimize this broader ecosystem, not just owned channels.

The findings were first reported in Harvard Business Review by Dubois and co-authors Allison R. Hess, John Dawson, and Akansh Jaiswal.