通过 Accept 请求头向 AI 代理提供 Markdown 内容

一篇文章介绍了通过 HTTP Accept 请求头进行内容协商，从而向人工智能（AI）代理提供 Markdown 格式内容的技术方案 1。该方案允许 AI 客户端通过发送 Accept: text/markdown 请求头来获取 Markdown 变体 1。由于 Markdown 格式去除了导航、样式、脚本和布局包装，AI 客户端得以跳过导航、脚本和布局标记，直接读取正文内容 1。

该方案能够减少 token 消耗，并降低抓取和解析量，从而加快首个 token 生成速度并降低延迟 1。同时，它有助于提高检索增强生成（RAG）管线的信噪比，避免广告、相关内容栏或模态覆盖干扰文本嵌入 1。符合 AI 就绪（AI-ready）标准的 URL 需满足特定条件，包括支持 Accept: text/markdown、设置 Vary: Accept、对不支持类型返回 406 状态码，并遵守 q-values 规则 1。

A new technical approach has been introduced for serving Markdown content to AI agents by leveraging HTTP Accept Headers for content negotiation 1. By utilizing the Accept: text/markdown request header, AI clients can skip navigation, styles, scripts, and layout wrappers to directly read the body text 1. This method reduces token consumption, decreases the volume of crawling and parsing, and accelerates the speed of the first token generation 1.

Furthermore, this Markdown formatting improves the signal-to-noise ratio in Retrieval-Augmented Generation (RAG) pipelines by preventing advertisements, related content columns, or modal overlays from interfering with text embeddings 1. To be AI-ready, a URL must support the Accept: text/markdown header, set the Vary: Accept header, return a 406 status code for unsupported types, and respect q-values 1.