Agent Experience Optimization (AX/AEO) — The Complete Guide
Your website has two audiences now. One of them is an AI agent. This guide covers everything you need to make your site discoverable, parsable, and actionable by ChatGPT, Claude, Perplexity, and autonomous agents.
What is Agent Experience Optimization?
Agent Experience Optimization (AEO) — also called AX (Agent Experience) — is the practice of making websites and applications discoverable, parsable, and actionable by AI agents.
Just as SEO made websites findable by search engines in the 2000s, AEO makes them usable by AI agents today. The difference: search engines crawl and index. Agents crawl, understand, and act. They book appointments, start trials, compare pricing, and complete purchases — all without a human clicking anything.
The term was popularized by Mathias Biilmann (CEO, Netlify) who identified four pillars of AX: Access, Context, Tools, and Orchestration. We build on that framework with a quantitative scoring model: 18 independently-scored checks that sum to a 0–100 AX score.
AEO vs SEO — How They Differ
| Aspect | SEO | AEO |
|---|---|---|
| Audience | Humans browsing search results | AI agents acting autonomously |
| Goal | Rank high, get clicks | Be cited, be actionable |
| Discovery | robots.txt, sitemap.xml | llms.txt, AGENTS.md, agent.json, MCP cards |
| Content format | HTML optimized for rendering | Markdown, SSR HTML, .md routes |
| Success metric | Rankings, traffic, CTR | Citations, API calls, task completion |
| Interaction | Click links, read pages | Call APIs, parse content, complete goals |
You need both. Strong SEO fuels AEO — AI systems prioritize content from domains with strong organic rankings and authority. But AEO adds an entirely new layer of optimization that SEO alone doesn't address.
The 17 Checks Behind Your AX Score
Every AX Audit runs the same 18 named checks against your site, each worth a fixed number of points that sum to 100 — no hidden weighting, no category rollup. The checks range from 1 to 12 points depending on how directly they predict real agent behavior.
Agent Discoverability — 12 pts
llms.txt, AGENTS.md, agent.json, agents.txt, agent-manifest.txt, ai-plugin.json, agent-card.json
MCP Discovery — 10 pts
/.well-known/mcp.json server card and MCP discovery meta tags
API Documentation — 10 pts
/api and /docs resolve with real, non-placeholder documentation
Authentication Standards — 10 pts
OAuth 2.0/OIDC, Web Bot Auth, and /.well-known/content-signal.json
AR-PROV (Agent Provisioning Readiness) — 10 pts
/.well-known/provisioning — can an agent open an account on its own
Content Negotiation — 8 pts
Accept: text/markdown returns real Markdown, not an HTML fallback
x402 Agent Payment Discovery — 6 pts
/.well-known/x402 and a live HTTP 402 challenge on an unauthenticated request
SSR Detection — 5 pts
Content present in the initial HTML response, not only client-rendered
Security Transparency — 4 pts
/.well-known/security.txt with Contact and Expires fields
NLWeb /ask Endpoint — 4 pts
A conversational-search endpoint matching the nlweb.ai spec
?mode=agent Dedicated Agent View — 4 pts
A meaningfully stripped-down response when ?mode=agent is requested
API Catalog (RFC 9727) — 4 pts
/.well-known/api-catalog as a valid RFC 9264 linkset
Token Efficiency — 4 pts
Homepage token count, and whether llms-full.txt is available
WebMCP Tool Quality — 3 pts
document.modelContext exposes well-described, typed tools a browser-side agent can call
Structured Data — 2 pts
JSON-LD, Schema.org vocabulary, and OpenGraph tags
Machine-Readable Content — 2 pts
robots.txt, sitemap.xml, feed.xml
Error Handling — 1 pts
Content-type-appropriate error responses for unknown paths
Rate Limiting Headers — 1 pts
Standard rate-limit headers on API responses
Key Standards to Implement
These are the files and protocols that agents look for when evaluating your site.
llms.txt — /llms.txt
Markdown file at your site root providing a curated summary for LLMs. Adopted by Stripe, Cloudflare, Anthropic, Vercel.
AGENTS.md — /AGENTS.md
Open format for guiding coding agents. Contains build commands, conventions, and boundaries. Donated to AAIF by OpenAI and Anthropic.
agent.json — /.well-known/agent.json
Agent-to-website communication protocol. Defines actions agents can take, with JSON Schema parameters and an inbox endpoint.
MCP Server Card — /.well-known/mcp/server-card.json
Standardized MCP server discovery. Advertises tools, transport, and auth requirements before connection.
agent-manifest.txt — /agent-manifest.txt
Capability declaration for AI agents. Tells agents what they CAN do (vs robots.txt which says what they cannot).
The 4 Pillars of AX (Biilmann Framework)
Mathias Biilmann (Netlify CEO) defined four pillars for designing products that work with AI agents as primary users.
Access
Can agents access your product? Do they have permissions? Must a human be in the loop?
Context
Does the LLM know about your product? Do your responses build the right context for follow-up actions?
Tools
Are you building for agents? Do you offer MCP servers, APIs, and SDKs that reduce friction?
Orchestration
Can agents trigger workflows from your product? Do you offer sandboxes and environments for autonomous operation?
How to Measure Your AX Score
The fastest way to measure is our free AX Audit. Enter your URL and get scored against all 18 checks with specific, actionable remediation recommendations.
Frequently Asked Questions
Isn't this just SEO?▼
Do I need llms.txt if I have good SEO?▼
What's the difference between AGENTS.md and agents.txt?▼
Is JSON-LD structured data important for agents?▼
How often should I re-audit?▼
Ready to optimize for agents?
Start with a free audit. See exactly how agents experience your site today, and get a prioritized roadmap to fix it.
Run free AX audit→Free audit. No signup required.