AuditHow It Works

How the Audit Works

We evaluate your website across 6 AI agent readiness standards using real HTTP requests — no guesswork, no JavaScript rendering, just what AI systems actually see when they visit your site.

Scoring & Grades

Each category is scored 0-100 based on specific checks. The overall score is a weighted average, and the grade is curved to reflect the current state of AI readiness across the web — where even the best sites rarely exceed 65.

GradeScoreMeaning
A+75-100Exceptional — leading-edge AI readiness across multiple protocols
A65-74Excellent — strong implementation of key AI discoverability standards
A-58-64Very good — well above average with solid fundamentals
B+52-57Good — clearly ahead of most sites in AI readiness
B46-51Above average — meaningful effort across several categories
B-40-45Decent — basics covered with room for improvement
C+35-39Average — some signals present, gaps in key areas
C30-34Below average — minimal configuration beyond the basics
C-25-29Weak — limited AI readiness signals
D+20-24Poor — very few AI discoverability measures in place
D15-19Very poor — almost no intentional AI configuration
D-10-14Minimal — barely any AI readiness signals detected
F0-9No AI readiness — no meaningful configuration found

Category Weights

Established standards (robots.txt, Schema.org, llms.txt) carry more weight because they're actionable today. Emerging protocols (MCP, A2A, Actions API) are weighted lower but still reward early adopters.

MCP10%
Abilities10%
Google A2A Protocol10%
llms.txt20%
robots.txt & Sitemap25%
Schema.org25%

The 6 Protocols

MCP (Model Context Protocol)

10% weight

MCP is an open standard created by Anthropic that lets AI models connect to external tools and data sources. When a website exposes an MCP server, AI agents can directly interact with its services — reading data, performing actions, and using the site as a tool.

What We Check

  • •/.well-known/mcp.json endpoint exists
  • •Response is valid JSON
  • •Contains required fields (server name, endpoint URL)
  • •Correct Content-Type header
  • •Declares capabilities (tools, resources, prompts)

How to Improve

  1. 1.Create a /.well-known/mcp.json file describing your MCP server
  2. 2.Define the tools and resources your site can expose to AI agents
  3. 3.Implement the MCP transport layer (HTTP+SSE or Streamable HTTP)
  4. 4.Test with Claude Desktop or another MCP client
Learn more →

Abilities / Actions API

10% weight

This checks whether your site exposes a structured API specification that AI agents can discover and use. OpenAPI/Swagger specs and the ChatGPT ai-plugin.json manifest allow agents to understand your API endpoints and call them programmatically.

What We Check

  • •OpenAPI spec at /openapi.json, /.well-known/openapi.json, or /swagger.json
  • •Spec is valid and parseable
  • •Contains defined API paths and operations
  • •ai-plugin.json manifest exists for ChatGPT/agent plugins
  • •API documentation page accessible at /api/docs

How to Improve

  1. 1.Generate an OpenAPI 3.0+ specification for your public API
  2. 2.Serve it at /openapi.json or /.well-known/openapi.json
  3. 3.Add descriptions to all operations and parameters
  4. 4.Create an ai-plugin.json manifest at /.well-known/ai-plugin.json
Learn more →

Google A2A Protocol

10% weight

The Agent-to-Agent (A2A) protocol, developed by Google, enables AI agents to discover and communicate with each other. By publishing an Agent Card at /.well-known/agent.json, your site declares itself as an AI-capable agent that other agents can find and collaborate with.

What We Check

  • •/.well-known/agent.json endpoint exists
  • •Response is valid JSON
  • •Contains required fields (name, url, description)
  • •Declares capabilities
  • •Defines skills with at least one skill
  • •Includes security configuration

How to Improve

  1. 1.Create a /.well-known/agent.json Agent Card
  2. 2.Define your agent's name, description, and URL
  3. 3.Declare capabilities and skills your agent supports
  4. 4.Add security schemes for authenticated interactions
Learn more →

llms.txt

20% weight

llms.txt is a simple text file (similar to robots.txt) that provides LLMs with context about your website. It uses Markdown formatting to give AI systems a structured overview of your site, its purpose, key pages, and how to navigate your content — making your site more understandable to AI.

What We Check

  • •/llms.txt file exists and is accessible
  • •Starts with a Markdown H1 heading (# Title)
  • •Contains H2 section headings (## Section)
  • •Includes Markdown links to key pages
  • •Has a blockquote summary (> description)
  • •Content is substantive (200+ characters)
  • •Bonus: /llms-full.txt extended version exists

How to Improve

  1. 1.Create a /llms.txt file in your site's root directory
  2. 2.Start with # Your Site Name as the H1 heading
  3. 3.Add a > blockquote with a one-line description of your site
  4. 4.Organize content into ## sections (About, Products, Documentation, etc.)
  5. 5.Include [links](url) to your most important pages
  6. 6.Consider creating a /llms-full.txt with expanded content
Learn more →

robots.txt & Sitemap

25% weight

This evaluates how well your site is configured for AI crawler access. While robots.txt and sitemaps have existed for years, AI crawlers (GPTBot, ClaudeBot, Google-Extended, etc.) are a new class of user-agents. Having explicit rules for these crawlers — whether allowing or restricting access — shows intentional AI configuration.

What We Check

  • •robots.txt exists and is accessible
  • •References a sitemap via Sitemap: directive
  • •No blanket Disallow: / blocking all crawlers
  • •Explicit rules for AI user-agents (GPTBot, ClaudeBot, etc.)
  • •Sitemap is accessible and valid XML
  • •Sitemap contains sufficient URLs (5+)
  • •Sitemap includes lastmod dates
  • •Lastmod dates are recent (within 12 months)

How to Improve

  1. 1.Ensure your robots.txt is present and well-formatted
  2. 2.Add a Sitemap: directive pointing to your sitemap.xml
  3. 3.Add explicit rules for AI crawlers: GPTBot, ClaudeBot, Google-Extended
  4. 4.Keep your sitemap up to date with accurate lastmod dates
  5. 5.Decide your AI crawling policy intentionally — blocking is fine, silence is not
Learn more →

Schema.org / JSON-LD

25% weight

Schema.org structured data (typically embedded as JSON-LD) helps AI systems understand what your site is about, what it offers, and how its content is organized. Rich structured data enables AI agents to extract accurate information about your business, products, articles, and services without scraping.

What We Check

  • •JSON-LD script tags present on the homepage
  • •JSON-LD is valid, parseable JSON
  • •Contains recognized Schema.org @type declarations
  • •Includes foundational types (Organization, WebSite, LocalBusiness)
  • •3+ distinct Schema.org types for type diversity
  • •Key properties populated (name, url, description)
  • •Rich types present (Product, Article, FAQPage, BreadcrumbList, etc.)
  • •Microdata attributes (itemscope/itemtype) as supplementary signal

How to Improve

  1. 1.Add <script type="application/ld+json"> to your homepage
  2. 2.Start with Organization or WebSite schema as a foundation
  3. 3.Include name, url, description, and logo properties
  4. 4.Add rich types relevant to your content (Product, Article, FAQPage, etc.)
  5. 5.Use Google's Rich Results Test to validate your markup
  6. 6.Consider using @graph to bundle multiple types in one block
Learn more →

FAQ

Who built this?

This tool was built by Caffeine (drinkcaffeine.com), an AI solutions agency. We help companies ship AI-powered products and prepare for an AI-first web.

How does the audit work?

When you submit a URL, our server makes HTTP requests to check for specific files and endpoints (robots.txt, llms.txt, /.well-known/mcp.json, etc.) and parses your homepage HTML for structured data. Each check contributes points to a category score. An AI model then generates a narrative summary and actionable recommendations based on your results.

Why is my score low?

Most websites score low because AI agent readiness is a brand-new concept. The protocols we test for (MCP, A2A, llms.txt) are emerging standards that very few sites have implemented yet. Even major tech companies score in the C-D range. A low score isn't a failure — it's an opportunity to get ahead of the curve.

Does the audit work on JavaScript-rendered sites?

Currently, the audit uses static HTML fetching. This means content that is only rendered client-side (e.g., React SPAs that inject JSON-LD via JavaScript) may not be detected. Most well-known endpoints (robots.txt, llms.txt, mcp.json) are not affected by this limitation since they are standalone files. Puppeteer/Playwright support is planned for a future version.

Is my data private?

Audit results are stored so they can be shared via link and appear on the leaderboard. If you check "Do not show the results on the boards" before running your audit, your results will not appear in the Recently Scanned or Leaderboard sections — but anyone with the direct link can still view the report.

How often should I re-audit?

Re-audit whenever you make changes to your site's AI configuration. The landscape is evolving quickly — new protocols are being adopted, and your competitors may be implementing them. We recommend quarterly audits at a minimum.

Can I audit internal or private sites?

No. The audit only works on publicly accessible websites. Private/local URLs (localhost, internal IPs) are blocked for security.

Run Your Audit