针对产业政策文本冗长、术语密集导致企业获取困难，以及通用大模型生成摘要存在事实失真与要素遗漏等问题，本文构建了涵盖2020~2024年国家及省市10类共200份政策的多源文本数据集，并设计了一套包含要素覆盖模板、事实性约束规则与格式规范的提示框架。实验表明了这套提示框架的有效性，为政务智能服务与企业政策精准获取提供了可落地的解决方案。

In response to the challenges faced by enterprises in accessing industrial policy information due to lengthy and terminology-dense texts, as well as issues of factual distortion and missing elements in summaries generated by general-purpose large language models, this paper constructs a multi-source text dataset comprising 200 policies from 2020 to 2024, covering 10 categories at national, provincial, and municipal levels. A standardized prompting framework incorporating element coverage templates, factual constraint rules, and output format specifications is designed. Experimental results demonstrate the effectiveness of this prompting framework, providing a deployable solution for government intelligent services and enterprises’ precise policy access.