Why agent-friendly sites win

Answer first: if agents cannot fetch clean facts, they invent them — or skip you. Ship /llms.txt, markdown twins, and a for-agents entry.

The problem

HTML is for humans. Navigation, chrome, and scripts waste tokens and blur the product card. Agents need a short map, then deep markdown on demand — the pattern formalized at llmstxt.org (v2).

What “agent-friendly” means here

Curated overview at /llms.txt (H1 + blockquote + H2 link lists to markdown). Twins at page.md (or index.md). Link relations: rel="alternate" type="text/markdown" and rel="describedby" href="/llms.txt". Robots that allow GPTBot, ClaudeBot, PerplexityBot, Google-Extended — and never block llms.txt.

GEO is just accuracy under a new name

Generative-engine answers reward sites that state prices, SLA, and ownership without fluff. Fake testimonials and vague “we’re a team of experts” copy lose. Exact SKUs and agent cards win.

What Alloy Bots shipped

This parent site implements the stack end-to-end: llms.txt, AGENTS.md, for-agents, markdown twins, sitemap including .md, JSON-LD, and agents.json. Explorer #1 (Briefcase) keeps the transactional product card on its own domain.

Practical checklist

One sentence answer at the top of every page
Machine card that agents can copy verbatim
Markdown twin for every key HTML page
Articles dated, authored, and answer-first