Technical Summary: Branding Without Belonging

Problem Statement

The paper addresses the erosion of cultural context in heritage foods as they enter global markets, specifically focusing on the role of Artificial Intelligence (AI) in this process. While globalization has long been documented as a force that standardizes food products and strips them of their ritualistic and historical significance, this study identifies a new mechanism of cultural flattening: AI-generated content marketing.

The core problem is that when Large Language Models (LLMs) are tasked with generating commercial product descriptions for culturally significant baked goods, they systematically suppress the cultural, religious, and diasporic depth of these items. Instead of preserving the "living memory" and ritual significance that differentiate heritage foods, AI models default to generic marketing language focused on ingredients, price, and convenience. This creates a risk where AI becomes the primary source of information for consumers, inadvertently erasing the authentic narratives of communities and reducing sacred or traditional foods to commodified products.

Methodology

The study employs a mixed-method approach combining a literature review with a controlled prompt experiment across three leading AI models.

1. Sample Selection

• Subjects: Six culturally significant baked goods were selected based on their clear ritual connections and global circulation:
  • Challah (Jewish-Diaspora, Shabbat)
  • Matzah (Jewish, Passover)
  • Panettone (Italian, Christmas)
  • Mooncakes (Chinese, Mid-Autumn Festival)
  • Conchas (Mexican, community tradition)
  • Modak (Indian, Ganesh Chaturthi)
• Models Tested: Three distinct LLMs were utilized to ensure cross-model validity:
  • ChatGPT 5.2 (OpenAI)
  • Claude 4.5 (Anthropic)
  • Gemini 3 (Google)

2. Experimental Design

Each food item was queried under three distinct framing conditions (36 total prompts per model):

• Set A (Open-Ended): Neutral inquiries (e.g., "What is [food]?").
• Set B (Commercial Framing): Sales-oriented prompts (e.g., "Write a product description for an online grocery store").
• Set C (Preservation Framing): Context-focused prompts (e.g., "Explain the religious meaning...").

3. Scoring Framework

Responses were evaluated using a Cultural Depth Score (CDS) based on six thematic categories:

1. Passed Down Traditions: Inter-generational memory.
2. Ritual Roots: Religious and ritual anchoring.
3. Culture Away From Home: Diaspora preservation.
4. Product Mode: Shift to marketing language (commodification).
5. Modern Remixes: Fusion and innovation.
6. Meaning Gets Flattened: Homogenization risk.

Scores ranged from 0 (not mentioned) to 2 (specific contextual detail). The CDS was calculated by summing the scores of the first three categories and the "Meaning Gets Flattened" category (max score of 8).

Key Results

The study confirmed all three hypotheses, demonstrating a significant disparity between commercial and cultural framing.

• H1 Confirmed (Commercial Suppression): Commercial prompts resulted in a drastic reduction in cultural depth. The average CDS for preservation prompts was 4.8/8, whereas commercial prompts averaged only 0.8/8. When asked to sell, AI models stripped nearly all cultural, religious, and diasporic meaning.
• H2 Confirmed (Religious Erasure): "Ritual Roots" were the most vulnerable dimension. In commercial contexts, references to religious meaning dropped from 67% (preservation prompts) to just 8%. In many cases, religious foods were described solely as "sweet dumplings" or "braided breads" without mention of their sacred origins.
• H3 Confirmed (Ritual Resilience): Foods with formal religious laws (Challah, Matzah) retained slightly more cultural depth than those based on everyday practice (Conchas). However, even the most ritualized foods suffered significant score drops under commercial framing.
• The Modak Case Study: Modak exhibited the most extreme erasure. In cultural prompts, it was described as a sacred offering to Lord Ganesha. In commercial prompts, its Cultural Depth Score dropped to 0.0, with its devotional meaning completely absent.
• Cross-Model Consistency: While ChatGPT 5.2 retained slightly more cultural content than Claude 4.5 or Gemini 3, all three models followed the same directional pattern: commercial framing consistently reduced cultural depth.
• Response Length Correlation: Commercial responses were significantly shorter (avg. 139 words) compared to preservation responses (avg. 282 words), suggesting that the constraint of brevity in marketing contexts contributes to the loss of nuance.

Key Contributions

1. Identification of AI as a Mechanism of Cultural Flattening: The paper establishes that AI does not merely reflect existing market trends but actively accelerates the "McDonaldization" of cultural heritage by optimizing for efficiency and predictability over authenticity.
2. Quantification of Cultural Depth: The study introduces the Cultural Depth Score (CDS) and a specific rubric for measuring how AI models handle heritage content, providing a standardized metric for future research.
3. Evidence of Contextual Steering, Not Knowledge Gaps: The results demonstrate that AI models possess the necessary cultural knowledge (as evidenced by high scores in preservation prompts) but selectively suppress it when the context shifts to commercial intent. This challenges the notion that AI lacks cultural awareness; rather, it is "contextually silent."
4. Strategic Framework for Mitigation: The paper proposes six specific strategies for marketers and AI developers, including "Cultural Context Labeling," "Cultural Depth Auditing," and "Community-Authored Training Data," to mitigate the erasure of heritage in AI-generated content.

Significance and Claims

The paper claims that the findings have critical implications for brand authenticity and content marketing strategy. As businesses increasingly rely on AI to generate product descriptions and brand narratives, there is a risk of unintentionally homogenizing heritage products, turning unique cultural artifacts into generic commodities.

The study argues that:

• Cultural knowledge is available but suppressed: The issue is not a lack of data in the models, but a failure to deploy it in commercial contexts.
• AI accelerates globalization's negative effects: While physical globalization takes time to dilute cultural meaning, AI can reproduce this flattening instantly across thousands of digital touchpoints.
• Preservation requires intentional framing: To maintain the authenticity of heritage foods in the digital age, marketers must move beyond generic prompts and explicitly instruct AI to include origin stories, ritual significance, and community meaning.

The paper concludes that protecting cultural meaning in food requires attention not just to physical supply chains, but to the digital systems that are becoming the primary interface through which consumers learn about and understand these products.