# AI-Ready Website Guidance

Websites that want to be “AI-ready” must expose explicit, static protocols and metadata for AI agents, rather than relying on hidden hooks or unstructured scraping. In practice this means dedicating a section (or well-known endpoints) on your site to machine-readable content. For example, UAIX’s own guidelines stress publishing discoverable files (sitemaps, schemas, etc.) and keeping them in sync with site structure: *“Keep sitemap, llms.txt, and public navigation aligned with current routes”*. In concrete terms, an AI-ready site should include:

- **Crawl and Permission Files:** A standard **robots.txt** at the domain root to govern crawler access (even AI bots often honor this).  Pair this with an up-to-date XML **sitemap** so AI tools can find all pages.  In addition, adopt an emerging **llms.txt / llms-full.txt** file: a human‑ and AI‑readable text file at `/llms.txt` (or `/llms-full.txt`) that lists key content and links in concise form.  This follows Jeremy Howard’s proposal: *“add a /llms.txt markdown file to websites to provide LLM-friendly content”*. In practice, companies like Anthropic, Hugging Face, Perplexity and Zapier already host `llms-full.txt` at their root, containing the flattened text of their sites. The llms.txt file (analogous to robots.txt) explicitly tells AI agents which content to read or ignore, without needing active crawling or server-side code.

- **Structured Data & Site Semantics:** Use **Schema.org** markup and semantic HTML so AI agents can parse your content. Embedding JSON‑LD in `<script>` tags clearly labels content (e.g. Articles, Products, Videos). As Geoff Mosher notes, schema markup “speaks the language of modern search engines and AI,” telling them *what kind of information* is on each page.  For example, tagging videos with a `VideoObject` schema (name, description, uploadDate, contentUrl, etc.) makes their metadata machine‑readable. Likewise, all images and media should have descriptive alt text and transcripts (so AI “sees” their meaning). In short, expose rich metadata (JSON-LD) and clear HTML structure (headings, lists, ARIA labels, etc.) so both AI and humans navigate easily.

- **UAIX Memory & Communication Files:** Provide UAIX‑style “memory packages” and metadata as static files. In the Teleodynamic/UAIX model, an AI agent should be able to fetch its context and instructions from your site in self‑contained bundles (rather than learning on the fly). Practically, you might host JSON or `.uai` files with page- or session‑level context.  UAIX’s own Memory Package Wizard page demonstrates one pattern: it embeds an AI‑digest JSON blob in the HTML (`<script data-ai-digest>…</script>`) that agents can read, while human visitors see a normal wizard UI.  Similarly, you can provide endpoints (e.g. `/ai/agent-profile.json` or `.uai` files) containing the agent’s allowed actions, boundaries, and recent history.  The Teleodynamic onboarding guidelines explicitly recommend **“UAIX-style memory packages”** alongside JSON and llms.txt as agent interaction channels. (And importantly, any such packages must contain *only public or contextual info* – UAIX forbids including secrets or tokens in memory: *“Do not promote…private keys, or API credentials”*.) 

- **Strict Boundaries (No‑Op Rule):** Clearly declare what agents may **not** do, and program them to “no-op” (do nothing) if a request is out of scope. Both UAIX and Teleodynamic emphasize that agents must respect strict boundaries. In the static onboarding wizard, agents are instructed to “reject *certification* language” and explicitly *“use no-op when unsure”*.  Thus your AI section should contain machine-readable policy (e.g. in JSON or text form) stating forbidden actions or data scopes. If an agent encounters ambiguity, it should fall back to no-action and request human review. This safe fallback is a core UAIX rule: ambiguous or unsafe queries must trigger no-op and involve a human, not hidden code execution.

- **Portable Evidence & Static Content:** Avoid hidden backend logic. Instead provide **portable, static evidence** of your site’s content and rules. For example, rather than letting an agent trigger a live action, hand it a “fact” by letting it read a static file. Examples include CSV/JSON downloads for data tables, flattened text transcripts for charts, or Markdown page exports. In practice, sites are encouraged to produce human‑ and machine‑readable exports. UAIX memory packages are one form of this, and the llms.txt can list links to Markdown versions of each page.  The goal is that an AI can fully consume everything it needs via HTTP GET (scripts, JSON, text), without privileged POST calls.  As Howard notes, the llms.txt model *“strips complete sites down to their bare… text, making it easier fodder to parse”*.  So include plain-text versions or structured summaries of your content (for instance, you could host Markdown or JSON versions of key pages at predictable URLs). 

**Implementation Checklist:** In summary, an AI‑ready section on your site should include at least the following:

- A root‐level **robots.txt** and **sitemap.xml** (kept current).
- A `/llms.txt` or `/llms-full.txt` file linking or containing your most important content in LLM-friendly form.
- Embedded **JSON-LD schema.org** markup (Article, Product, VideoObject, etc.) on your pages.
- Accessible markup (semantic HTML, ARIA, alt text) so content is machine‑interpretable.
- A designated “AI section” (e.g. `/uaix/` or `/.well-known/ai/`) hosting UAIX memory/profile files (JSON or `.uai`) describing the agent’s context, role, and limits.
- Explicit machine-readable policy or metadata on allowed actions (with a default no-op for anything outside scope).
- **No secrets or live commands** in any agent-facing file (as UAIX notes: *“Do not promote…API credentials”*). All agent guidelines and data must be public and auditable.

By following these practices, your site gives AI agents a clear, safe communication layer. Agents can prefetch context (via llms.txt and memory bundles), read structured data (schema/JSON-LD), and respect hard boundaries (via no-op rules) before they ever try an action. This aligns with UAIX’s vision of auditable, interoperable AI interfaces and complements related standards like schema.org markup and robots.txt. 

**Sources:** Authoritative AI-web integration guides and standards (UAIX schema docs and Teleodynamic onboarding) stress using static JSON/Markdown files (llms.txt, memory packages, JSON-LD) to define AI behavior. SEO and webmaster resources likewise recommend schema markup and accessibility for AI discovery. Industry examples (Anthropic, Hugging Face, Zapier) show llms.txt usage in practice. All citations above detail these best practices.