Check reference

Explore every AI agent readiness check in the Sitebulb Labs Extension, grouped by category, and learn how results contribute to your site's score.

The Sitebulb Labs Extension has 50 checks in six categories. A normal scan leaves out checks for emerging standards. You can include those in the extension’s settings; they are marked emerging below. They can add bonus points, but missing an emerging signal does not lower a category score.

A check can return one of six statuses:

Status	Meaning	Effect on score
Pass	The check found the signal or a sound configuration	Full credit if scored
Warn	Something is present but incomplete, or needs attention	Half credit if scored
Fail	A problem the check can confirm	No credit if scored
N/A	The check does not apply or there is nothing relevant to assess	Excluded
Info	A finding without a clear pass or fail	Excluded
Unmeasured	The check applies, but the extension could not complete it	Excluded

Some checks are bonus checks. Passing one adds points to the overall score; its absence cannot reduce a category score. The extension calculates category scores from the other applicable checks, then combines those scores with extra weight on Content accessibility and Bot access control. Bonus points are capped at 10. Protocol discovery and Commerce contain only bonus checks, so neither has a category percentage unless scored checks are added in a future version.

The priority panel estimates how many points each fix could add. It draws those estimates from the current scan rather than a fixed list of “easy wins”.

Discoverability

Can an agent find the site’s pages, files, and published services?

robots.txt present & valid
checks that /robots.txt is reachable and contains usable crawl rules or sitemap references.
Sitemap available
looks for a sitemap that lists pages to crawl.
Link response headers
looks for HTTP Link headers such as alternate, canonical, and describedby.
Agent-friendly 404s
checks that a missing page returns HTTP 404 or 410 instead of a misleading 200 response.
Redirect hygiene
looks for JavaScript or meta-refresh redirects that an agent reading the initial HTML might miss.
Sitemap freshness (lastmod)
checks sitemap <lastmod> values that help an agent identify changed pages.
DNS for AI Discovery (DNS-AID)
looks for a DNS TXT record advertising agent endpoints.
ARD catalog (ai-catalog)
looks for a catalog of the domain’s agents, skills, APIs, or MCP servers.
ARD entry validity
checks catalog entries for a domain-anchored identifier, display name, media type, and a URL or inline data.
ARD trust manifest
checks whether the catalog points to a manifest that binds its entries to a publisher identity.
NLWeb Schema Feeds
looks for a schemamap: directive in robots.txt pointing to structured-data feeds.

Content accessibility

Can an agent read and understand the content it finds?

Content without JavaScript
compares the server’s HTML with the rendered page to see whether meaningful content depends on JavaScript.
Markdown content negotiation
checks whether the page serves Markdown when requested with Accept: text/markdown.
Markdown negotiation caching (Vary)
checks for Vary: Accept so caches do not mix up HTML and Markdown responses.
llms.txt links resolve
checks that links in /llms.txt do not point to missing pages.
JSON-LD entity linking (sameAs)
checks whether structured data links an entity to its other known profiles or identifiers.
Organization schema completeness
checks the details in the page’s Organization structured-data node.
Schema type breadth
looks for specific schema types, such as articles or products, rather than only generic types.
Structured data (JSON-LD)
looks for Schema.org JSON-LD on the page.
Markdown URL fallback (.md)
looks for a Markdown version of the page at a .md URL.
Markdown alternate link
looks for a rel="alternate" link to a text/markdown version of the page.
llms.txt present
looks for a Markdown map of the site’s important content at /llms.txt.
Modular llms.txt
looks for section-specific llms.txt files.
Speakable content markup
looks for speakable markup identifying content suitable for voice output.

Bot access control

What do the site’s crawl rules say about AI bots and user-triggered agents?

AI bot rules in robots.txt
checks the rules for named AI crawlers, including whether they are allowed or blocked.
robots.txt agent-user policy
checks whether rules for user-triggered agents, such as ChatGPT-User or Claude-User, block access.
Content Signals
looks for Cloudflare’s Content-Signal: directive in robots.txt.
Web Bot Auth
looks for the signatures directory used to identify agents with HTTP Message Signatures.

Protocol discovery

Can an agent find an API, tool, or service it can use? All checks in this category are bonus checks.

MCP Server Card
looks for a well-known descriptor of an MCP server.
API Catalog
looks for the RFC 9727 /.well-known/api-catalog endpoint.
Rate limit response headers
checks whether an API advertises limits so an agent can pace its requests.
OAuth Authorization Server discovery
looks for RFC 8414 metadata about obtaining authorization.
OAuth Protected Resource metadata
looks for RFC 9728 metadata identifying a resource’s authorization servers.
Agent Skills
looks for a manifest of callable skills.
WebMCP
checks the live page for tools exposed through document.modelContext.
A2A Agent Card
looks for a card describing an agent’s capabilities and endpoint.

Agent experience

Can an agent make sense of the page and use its controls?

Content efficiency
measures readable text against the amount of HTML.
Page token budget
flags pages whose size could cause an agent to miss content when working within a context limit.
Metadata completeness
checks the page title, description, and canonical URL.
Trust anchor pages
looks for about, contact, and privacy pages.
Document structure
checks landmarks and heading order.
Native interactive controls
checks whether interactive elements use controls that keyboard users and agents can operate.
Accessible names on controls
checks whether buttons and other controls have names an agent can identify.
Form control labelling
checks whether form fields have labels.
Accessibility-tree injection safety
looks for instruction-shaped text in labels and alternative text that could steer an agent.
Pagination signals
looks for rel="next" and rel="prev" links on paginated content.

Commerce

Can an agent transact with the site? All four checks are emerging bonus checks, so an ordinary site is not penalised for lacking them.

x402 payments
looks for HTTP 402 payment support.
Merchant Payment Protocol (MPP)
looks for merchant payment capabilities.
Universal Commerce Protocol (UCP)
looks for agent-driven commerce support.
Agentic Commerce Protocol (ACP)
looks for a way for agents to complete purchases.

Check sets

For most addresses, the extension runs all checks, including content and API checks. Individual checks can still report N/A. For hosts whose first label is api, apis, graphql, or rest, or begins with one of those followed by a hyphen (for example, api-eu.example.com), it skips content-only checks. It does not currently offer a manual check-set selector. You can turn emerging checks on or off in settings; they are off by default.