Checks

Checks

The 100+ checks a scan runs. Open one to read why it matters, or run it alone against any page.

100+ checks.

IdCheckSeverityLast runRun
Crawler access & permissions 20
A1

File present at root, valid syntax, no malformed or orphaned rules.

High
A2

Explicit directives for GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot and peers — not just User-agent: *.

High
A3

Wildcard rule doesn't lock the root away from every crawler.

Critical
A4

robots.txt Content-Signal (or IETF AI-Preferences) statements of what AI may do with the content.

High
A5

noindex / noarchive / noai / noimageai in tags or headers that quietly override robots.txt.

Critical
A6

Whether a security layer blocks well-behaved agents even when robots.txt permits them.

Critical
A7

Actually requests the page as the agents of ChatGPT, Claude, Perplexity, Gemini, Meta AI, Apple Intelligence and Alexa instead of trusting the stated policy.

Critical
A8

Googlebot and Bingbot still allowed — a proxy for accidental over-blocking.

High
A9

tdm-reservation header, meta element or /.well-known/tdmrep.json declaring text-and-data-mining rights.

High
A10

Cross-checks that the endpoints published manifests advertise actually answer.

High
A11

Whether robots.txt distinguishes the bots that train models from the bots that fetch a page when a user asks about you.

High
A12

Whether the page still serves its substance to a client that accepts no cookies.

Critical
A13

GET endpoints that refuse an agent without a CSRF or session token it has no way to obtain.

High
A14

Whether a 429 carries Retry-After and a machine-readable body rather than an HTML error page.

High
A15

Human-verification widgets attached to contact, newsletter or signup forms.

High
A16

A Really Simple Licensing (RSL) document, found from robots.txt, a Link header or the page, stating what AI systems may do with the content and on what terms.

High
A17

Whether AI crawlers are answered 402 Payment Required with a crawler-price, which prices a page rather than blocking it.

High
A18

Requests robots.txt as GPTBot, ClaudeBot and PerplexityBot and compares each answer with the copy everyone else gets.

High
A19

Whether robots.txt disallows the scripts and stylesheets the homepage loads from its own host, for Googlebot and Applebot.

High
A20

Requests the sign-in, cart and checkout pages the homepage links to as a browser and as ChatGPT-User and Claude-Agent, and looks for a CAPTCHA on them.

High
Discovery surfaces 12
B1

Sitemap reachable at the standard path and referenced from robots.txt.

High
B2

A markdown index for language models at /llms.txt.

High
B3

H1 name, blockquote summary, H2 sections and correct link syntax.

High
B4

Extended single-file concatenation of the full markdown corpus.

High
B5

rel=describedby, api-catalog, service-desc, alternate advertised in response headers.

High
B6

Root-level instruction files written for agents rather than crawlers.

High
B7

Presence of the well-known namespace agents probe first (RFC 8615).

High
B8

SVCB/HTTPS records under _agents.<domain> advertising agent endpoints.

High
B9

Human-readable URLs that keep working — treated as a public contract.

High
B10

JSON, RSS/Atom or markdown endpoints served alongside HTML.

High
B11

Navigation built from real anchors rather than click handlers on divs.

High
B12

Whether the pages behind a listing can be reached by following links.

High
Content accessibility & extractability 16
C1

The substance is in the first response, not injected by client-side JS.

Critical
C2

Measures what share of visible text survives with JavaScript switched off.

HighNeeds a browser
C3

Returns text/markdown when an agent sends Accept: text/markdown.

High
C4

A .md sibling or /index.md fallback for every HTML page.

High
C5

One h1, logical order, real landmarks instead of div soup.

High
C6

Docs, pricing and key pages reachable without authentication.

High
C7

Compares the response served to an AI user-agent against the baseline browser response.

Critical
C8

Main content small enough to be read whole — under 8,000 tokens, failing above 25,000.

High
C10

Buyer questions answered directly in a quotable Q&A shape.

High
C11

Scannable structures an engine can lift verbatim.

High
C14

Whether an AI user-agent is shown different claims, not merely different markup.

High
C15

Whether comments, reviews and other user submissions are structurally separable from what the site itself says.

High
C16

Prices, specifications and tables that exist only as pixels.

HighNeeds a browser
C17

How far into the page the substantive answer begins.

High
C18

Whether asking for a language actually returns it.

High
C19

Where the main content starts and ends in the page's text, against the character caps of Cline (50,000), Claude Code (100,000) and Gemini CLI (250,000).

High
Structured data & metadata 10
D1

Structured data on the page and syntactically correct.

High
D2

Organization, Product, Article, FAQ, LocalBusiness rather than generic markup.

High
D4

Unique, descriptive, present in the raw HTML.

High
D5

Correct rel=canonical, no conflicting signals.

High
D6

Social/preview metadata completeness.

High
D7

Links out to Wikipedia, Crunchbase, LinkedIn that tie the site to a known entity.

High
D8

Declarative actions an agent could invoke.

High
D9

schema.org speakable for voice agents.

High
D10

Valid lang attribute; hreflang where relevant.

High
D11

Conflicting or duplicated Organization identities in the structured data.

High
Agent capability & protocol exposure 15
E1

A live Model Context Protocol server (streamable HTTP or SSE).

Medium
E2

/.well-known/mcp/server-card.json describing the server (SEP-1649).

Medium
E3

WebMCP tools on the page: forms declared with toolname and tooldescription, and tools registered on document.modelContext.

Medium
E4

/.well-known/agent-card.json for agent-to-agent delegation.

Medium
E5

Machine-readable API description, optionally with agent extensions.

Medium
E6

/.well-known/api-catalog indexing the site's APIs.

Medium
E7

An agent-skills index at /.well-known/skills.json.

Medium
E9

Conversational endpoint returning schema.org-typed JSON.

Medium
E10

UI resources an agent can render on the user's behalf.

Medium
E11

agent-permissions.json, UCP capability profile, or an equivalent declaration of what agents may do.

Medium
E12

Whether search runs as a GET with a named parameter or only as a JavaScript POST.

Medium
E13

Whether the advertised MCP server answers tools/list with usable tool definitions.

Medium
E14

Whether the paths in the OpenAPI spec actually answer.

Medium
E15

Whether an endpoint that is going away says so in a header an agent can read.

Medium
E16

Whether the advertised MCP endpoint is public HTTPS on a host with an IPv4 address and no redirect to another host.

Medium
Identity, trust & auth 15
F1

/.well-known/http-message-signatures-directory so agents can prove who they are (RFC 9421).

Medium
F2

/.well-known/openid-configuration or RFC 8414 authorization-server metadata.

Medium
F3

RFC 9728 metadata for the resource an agent wants to reach.

Medium
F4

RFC 7636 S256 required of agent-grade OAuth clients.

Medium
F5

Secure transport and a clean certificate chain.

Medium
F6

The trust furniture an engine looks for before citing you.

Medium
F7

rel=author, JSON-LD author/publisher, named bylines.

Medium
F10

security.txt, and no sensitive paths (admin, .env, .git) leaked through robots.txt.

Medium
F11

A did:web identity document at /.well-known/did.json.

Medium
F12

Whether a request signed under RFC 9421 is treated differently from an unsigned one.

Medium
F13

Whether the published key directory is a valid JWK Set a verifier can actually load.

Medium
F14

Text hidden from readers but visible to parsers, carrying instructions rather than content.

Medium
F15

Forms that send, book or charge, with nothing in front of them.

Medium
F16

Whether an agent can authenticate on its own, or needs a human at a login screen.

Medium
F17

Walks the MCP authorization chain from an unauthenticated call: 401, protected resource metadata, the first authorization server, issuer, S256, client registration, and a token endpoint that takes form posts.

Medium
Agentic commerce & payments 13
G1

A paid endpoint answering an unauthenticated request with a decodable 402.

Low
G2

Payment discovery document with network, facilitator and services array.

Low
G3

/.well-known/acp.json, including delegated-payment tokens.

Low
G4

/.well-known/ucp merchant profile.

Low
G5

WWW-Authenticate: Payment challenge and required parameters.

Low
G6

Stripe SPT, L402, Coinbase Commerce, AP2 and similar.

Low
G7

A machine-readable catalogue an agent can find without crawling every page.

Low
G8

Prices declared in structured data, and MCP tools or API operations an agent can order through.

Low
G9

The same price in the visible text, the structured data, the preview tags and the catalog feed.

Low
G10

Stock state consistent between the page and its machine-readable claims.

Low
G11

Whether numbers carry their own meaning or depend on the reader's locale.

Low
G12

The Verifiable-Credentials authorisation layer beneath an agentic checkout.

Low
G13

Whether the card fields on the checkout page carry cc-number, cc-exp, cc-csc and cc-name autocomplete tokens.

Low
Freshness 6
H1

Human-readable dates on the page.

Low
H2

Freshness expressed in machine-readable form.

Low
H3

Sitemap actually reports meaningful modification dates.

Low
H4

Server headers not signalling stale content to crawlers.

Low
H5

Whether the machine-readable catalog is as current as the pages it describes.

Low
H6

Whether a repeat visit carrying If-None-Match or If-Modified-Since gets a 304.

Low
Performance & reliability 11
I1

Latency measured from the crawler's side, not a browser's.

Low
I2

Real-user field metrics folded into the readiness score.

Low
I3

Scrunch fails pages slower than five seconds outright.

LowNeeds a browser
I4

200s, no long or cross-host redirect chains, correct handling of 403/404/5xx.

Low
I5

Infrastructure failures that only show up in AI-crawler traffic.

Low
I6

How much output an agent burns to get an answer.

Low
I7

How much of the delivered payload is the content an agent came for.

LowNeeds a browser
I8

Whether the substance arrives in the first response or several requests later.

LowNeeds a browser
I9

What a client that does not wait for the page to settle actually receives.

LowNeeds a browser
I10

Whether the page is served gzipped or brotli-compressed to a client that asks for it.

Low
I11

Whether the homepage's links are under 250 characters and its linked documents are served inline in a type fetch tools read.

Low
Accessibility (WCAG) 14
J1

Text alternatives on every meaningful image.

Medium
J2

Every input and button named, every control reachable.

Medium
J3

Valid roles and attributes, named dialogs and tooltips, nothing focusable hidden from the accessibility tree, no positive tabindex, and no dangling aria-labelledby / aria-describedby targets.

Medium
J4

A real title, a main landmark, named iframes, and landmarks that can be told apart.

Medium
J5

What a screen reader announces matches what the eye reads.

MediumNeeds a browser
J6

Structural markup an assistive client can navigate.

Medium
J7

No zoom suppression; explicit media dimensions.

Medium
J8

autocomplete attributes that let assistive tooling fill forms.

Medium
J9

Every input, select and textarea carrying the name its form needs to submit it.

Medium
J10

Selects, dates and file inputs built from real elements rather than reconstructed in JavaScript; no field that can only be picked, no item that can only be dragged, no page drawn on a canvas.

Medium
J11

Sticky headers, footers and floating widgets covering the controls beneath them.

MediumNeeds a browser
J12

Every link says where it goes, including the icon-only ones.

Medium
J13

Counts the buttons and form fields that sit inside iframes from another origin.

MediumNeeds a browser
J14

Whether the page virtualises lists or removes rows as it scrolls, which leaves agents' references stale.

MediumNeeds a browser