A Markdown version for LLMs is a second, leaner version of a page in Markdown that AI agents and language models can request instead of the HTML. It carries the content without navigation, scripts and layout.
How is it served?
There are two common ways. A site can publish a separate URL per page, often ending in .md. Or the server uses content negotiation. The agent sends Accept: text/markdown and the server answers with Markdown instead of HTML. Cloudflare announced a beta of this on February 12, 2026 for Pro, Business and Enterprise plans. In its own post, the HTML version used 16,180 tokens and the Markdown version 3,150, which Cloudflare calls an 80 percent reduction.
Claude Code’s WebFetch tool is one client that uses content negotiation. Anthropic documents that it sends an Accept header preferring Markdown over HTML and a user agent beginning with Claude-User.
What does Google say?
Google writes that you don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search, and that Google Search ignores them. A Markdown version may still be useful for other services.
Example
GET /pricing/ HTTP/1.1
Accept: text/markdown
If the server supports it, it returns the page as Markdown. Otherwise it returns HTML.
More on the searchVIU blog
Sources
- Optimizing your website for generative AI features on Google Search – Google Search Centraldevelopers.google.com
- Introducing Markdown for Agents – Cloudflare Blog (12.02.2026)blog.cloudflare.com
- What the Claude Code leak reveals about AI grounding — and what it means for your GEO strategy – searchVIU blog (2026)searchviu.com
- Tools reference – Claude Code Docs (Anthropic)code.claude.com
Last updated: