dervig/NorGEO-Bench

NorGEO-Bench v1.0是一个挪威生成引擎优化（GEO）基准测试，用于测量领先的具备网络搜索功能的LLM如何回答关于挪威商业市场的挪威语问题。数据集包含253个提示，分布在11个业务垂直领域，每个提示发送给ChatGPT、Claude和Gemini各三次，每个响应由独立的LLM（Claude Opus 4.6）验证，产生准确性、幻觉严重性、品牌引用、引用域和提及实体信号。数据集的主要目的是测量哪些挪威网站和企业在生成式AI模型中被突出显示——这一信号随着用户从谷歌转向基于聊天的界面而变得越来越重要。

NorGEO-Bench v1.0 is a Norwegian benchmark for Generative Engine Optimization (GEO) that measures how leading web-enabled LLMs answer Norwegian-language questions about the Norwegian business market. The dataset includes 253 prompts distributed across 11 business verticals, each sent three times to ChatGPT, Claude, and Gemini, with each response verified by an independent LLM (Claude Opus 4.6) to produce accuracy, hallucination severity, brand-cited, cited-domain, and mentioned-entity signals. The primary purpose of the dataset is to measure which Norwegian websites and businesses are highlighted by generative AI models—a signal that is becoming increasingly important as users shift from Google to chat-based interfaces.