Prompt Impact on Brand Recommendation
Design multiple sets of controlled prompts to test how different questioning approaches, context settings, and constraints affect AI brand recommendation results. The study found: contextual information in prompts such as scenario descriptions, budget ranges, and target user groups significantly changes the brand distribution of recommendations; open-ended questions produce more diverse brand recommendations than constrained questions; different platforms show明显 differences in prompt sensitivity. The experimental results provide empirical evidence for brand prompt strategy optimization.
Research Objective
This experiment aims to study how different prompt designs affect AI brand recommendation results. Specific objectives include: testing the impact of different questioning approaches (open-ended vs. constrained) on recommended brand distribution; analyzing the degree to which contextual information in prompts (scenario descriptions, budget ranges, target users) changes recommendation results; comparing sensitivity differences across platforms to prompt changes. The experimental results provide empirical evidence for brands' prompt strategy optimization.
Methodology
The experiment designed 30 sets of controlled prompts, each containing an open-ended version and a constrained version. These were executed on four platforms, totaling 240 tests. Data recorded included brand lists, brand rankings, and recommendation descriptions for each recommendation. Analysis used metrics such as brand appearance frequency, ranking distribution, and diversity index.
Key Findings
The study found: (1) Open-ended questions produce 45% more diverse brand recommendations than constrained questions; (2) Including budget ranges in prompts causes recommendation results to concentrate on brands in the corresponding price segment, reducing brand coverage by 60%; (3) Scenario descriptions (like "business hotels suitable for business trips") cause recommendations to favor professional brands in that scenario; (4) Different platforms show明显 prompt sensitivity differences: ChatGPT is most sensitive to prompt wording, DeepSeek reacts more strongly to Chinese-context prompt changes, Doubao and Kimi are more sensitive to prompt structure than wording changes.
Implications for Brands
Brands should focus on prompt patterns commonly used by target users and optimize content to match these patterns. Embed natural language expressions that users might use in brand content to help AI build associations between the brand and specific scenarios and needs. Meanwhile, brands need to monitor prompt sensitivity differences across platforms and develop targeted content strategies.