Study goals
To assess the reliability, sensitivity, and effectiveness of ChatGPT-4o and DeepSeek-V2 in prioritizing neighborhoods for new business openings, comparing recommendations with and without structured data, using Porto Alegre as a geomarketing case study.
Relevance / originality
The study addresses a gap on generative AI use in geomarketing, offering an empirical comparative analysis of LLM models. It highlights these tools’ potential for territorial decisions by integrating structured data and qualitative inferences in urban market contexts.
Methodology / approach
Standardized prompts were applied to two AI models in scenarios with and without neighborhood-level structured data. Comparative analysis used criteria of reliability, information sensitivity, and efficacy, employing official indicators and geomarketing methods.
Main results
Both models generated recommendations aligned with geomarketing practices, even thought changing outputs when given access to structured data. ChatGPT-4o adopted a weighted quantitative approach, while DeepSeek-V2 combined quantitative analysis with qualitative filters and contextual consistency.
Theoretical / methodological contributions
Expands evidence on integrating generative AI into Spatial Decision Support Systems. Demonstrates distinct LLM reasoning approaches, proposes evaluation criteria for territorial decisions, and highlights AI sensitivity to the quality and organization of data.
Social / management contributions
Provides managers and consultants with insights on using generative AI to support location decisions, enhancing market analysis, identifying opportunities, and complementing traditional geomarketing methods with qualitative and contextual interpretations of territories.