AI Capability Signaling

Beyond being reachable (ai-crawler-access), readable (ai-retrievability), and parseable (structured-data), you can proactively publish agent-friendly representations of your content. As AI assistants and autonomous agents increasingly fetch, summarize, and act on the web, these signals help them consume your content correctly and represent your brand accurately. This skill is the forward edge of AEO — sometimes called "agentic engine optimization." It works for any business, B2B or B2C.

Honest framing up front: most of these are emerging and low-adoption today. Engines still consume ordinary HTML for the most part, and few broadly honor signals like llms.txt. Do this work because it is cheap, future-facing, and differentiating — not because it guarantees citations. Land the proven levers first (answer-engineering, evidence-and-citations, authority, structured-data), then add these.

When to use this skill

• Making a site "AI-readable" / "agent-ready" beyond standard SEO/AEO.
• Setting up Markdown versions of pages, llms.txt, or copy-for-AI affordances.
• Helping agents and users feed your content to assistants accurately.
• Future-proofing for an agent-consumed web.

Before you start

Read aeo-foundation.md for the priority pages and prompts — signal the content that matters, not everything. The dedicated llms-txt skill handles that file in depth; this skill is the broader umbrella. Details on every format are in references/ai-readable-formats.md.

The signals

1. llms.txt and llms-full.txt. A curated, Markdown map of your key pages (and an optional full-content version) at the site root. Cheap to publish; see llms-txt for the file itself and its current, limited reception.
2. Markdown twins. A clean Markdown rendering of each important page — stripped of nav, ads, and boilerplate — served alongside the HTML (e.g. a .md companion URL or via content negotiation). Gives agents pure, unambiguous content. Used effectively by documentation platforms.
3. Copy-for-AI / raw view. UX affordances — "copy page as Markdown," a raw/plain-text view — so a human can hand your page to an assistant accurately and completely. Highest value on docs and long-form pages people paste into chat.
4. Machine-readable metadata. Clean per-page metadata (title, author, published and updated dates, summary) and structured content feeds (RSS/JSON). Optionally surface per-page token counts so agents can budget context.
5. Capability descriptions. Plainly state what the product, service, or API does in terms an agent can use to recommend or operate it correctly. For API/tooling brands this extends to agent/tool manifests (e.g. an MCP server) — so agents can actually use you.
6. Parity. Whatever agent-readable version you publish must match the human page. Drift between the two is worse than not having it.

Workflow

1. Pick the pages. From the foundation, list the priority pages worth exposing (key answers, docs, product/comparison pages). Don't twin the whole site.
2. Choose formats per page. llms.txt for the whole site (always, it's cheap); Markdown twins for top content; copy-for-AI on docs/long-form; metadata everywhere.
3. Publish llms.txt / llms-full.txt. Hand off to llms-txt for the file.
4. Generate Markdown twins for the priority pages and wire up a stable convention (companion URL or content negotiation). Automate so they stay in sync.
5. Add copy-for-AI / raw affordances where users paste content into assistants.
6. Expose clean metadata and ensure it agrees with the visible page.
7. Set expectations and monitor. Treat as forward-looking; revisit as engine and agent adoption grows.

Output format

• A capability-signaling plan: which formats for which pages, and why.
• The llms.txt (via llms-txt) and a sample Markdown twin for a priority page.
• Copy-for-AI implementation guidance for the relevant templates.
• A parity/sync note describing how the agent-readable versions stay current.
• (assumption) tags where page priorities or platform capabilities were unknown.

Common pitfalls

• Treating any of this as a citation guarantee — it is forward-looking, not proven ROI.
• Markdown twins drifting from the live page (stale or contradictory content).
• Over-investing here while the proven AEO levers are still weak.
• Exposing machine content that is incomplete or out of date.
• Twinning everything instead of the pages that matter.

Related Skills

• llms-txt — the llms.txt / llms-full.txt file in depth, with honest expectations.
• ai-retrievability — being readable at all (server-rendered, clean HTML) comes first.
• structured-data — machine-readable entities and facts via schema.
• answer-engineering — the actual content quality these formats expose.