Measure how many thousands of context tokens your raw HTML wastes when AI coding and shopping agents fetch your pages โ and test Accept: text/markdown support.
| Delivery Format | Typical Homepage Payload | LLM Citation Reliability |
|---|---|---|
| Raw Modern HTML + Inline CSS/JS | 12,000 โ 45,000 tokens (85%+ noise) | Pricing & tables risk context-window truncation |
| Accept: text/markdown Negotiation | 1,200 โ 3,200 tokens (clean tables & headings) | 80% lower latency & exact table extraction |
Introduced into mainstream edge infrastructure in February 2026, it lets servers return clean Markdown when an AI agent sends `Accept: text/markdown` while still serving normal HTML to human browsers (`Vary: Accept`).
No โ provided the Markdown is derived 1:1 from the exact visible HTML text shown to human visitors without hidden keyword stuffing or prompt injection.