Methodology · version 4.3
Methodology 4.3 published ; public criterion catalog updated .
How to interpret technical AI Readiness
Methodology 4.3 preserves the Methodology 4.2 AI Readiness model and bounded internal-page sampling. Browser WebMCP and Remote MCP are separate optional agent capabilities and do not change the primary score. Remote MCP diagnostics are limited to server/discover and tools/list and never execute advertised tools.
This page explains what is evaluated, why each criterion matters, and which part of the assessment it belongs to.
AI Readiness · 0–100
The primary score contains four dimensions with a defensible connection to technical AI readiness. Delivery Reliability is shown separately and does not change AI Readiness.
- Machine-readable semantics. Structured entities and data that can be interpreted without guessing from prose alone.
- Discovery and crawler rules. Public discovery surfaces, robots policy and supported search-crawler access. Other content-use policies are shown separately and do not change this dimension.
- Metadata and indexing controls. Canonical URLs, metadata and explicit indexing directives.
- Technical accessibility. Successful HTTP delivery and usable content without requiring unsupported client-side execution.
Delivery Reliability · 0–100
HTTPS, redirect safety, HSTS, Content-Security-Policy, framing protection, X-Content-Type-Options, Referrer-Policy, mixed-content and form-action safety are useful web-security signals, but they are reported as a separate reliability index and do not reduce AI Readiness.
Experimental coverage
Browser WebMCP, Remote MCP, agent interaction semantics, llms.txt and llms-full.txt are shown separately as optional signals. Their absence does not reduce AI Readiness, and their presence does not guarantee indexing, citation or use by an AI product.
AI-commerce 1.3
AI-commerce 1.3: published ; public catalog updated .
AI-commerce is a separate context-gated index. It is evaluated only when a commerce context is confirmed and uses a bounded product sample plus optional agentic discovery signals such as UCP, A2A and OpenAPI.
UCP profile compatibility is verified against the current public 2026-08-25 profile shape without changing the AI-commerce scoring contract.
The checker never performs checkout, payment, order creation, A2A tasks, MCP tools or other state-changing agent operations.
Statuses and completeness
- Pass: the expected technical signal was observed.
- Warning: a usable signal exists but has a relevant limitation.
- Fail: a confirmed contradiction or technical problem was observed.
- Not evaluated: the available response or context is insufficient for a defensible conclusion.
- Optional: an experimental or non-required capability is absent or not applicable.
A separate completeness indicator shows how much of the evaluable model was actually assessed. A high score with low completeness should not be interpreted as equivalent to a fully evaluated result.
Criterion catalog
The criteria below use the same definitions as the Russian methodology. Public links point to stable criterion anchors so guides and reports can reference the relevant explanation directly.
Machine-readable semantics
Structured data
AI ReadinessWhat is evaluated: Readable Schema.org markup in at least one supported format.
Why it matters: Structured data exposes entity types, properties and relationships explicitly instead of leaving them to inference from prose.
Schema.org JSON-LD
DiagnosticWhat is evaluated: Syntactically readable JSON-LD blocks using Schema.org vocabulary.
Why it matters: JSON-LD is a common way to publish structured entities separately from the surrounding HTML markup.
Schema.org Microdata
DiagnosticWhat is evaluated: Schema.org markup embedded in HTML through Microdata attributes.
Why it matters: Microdata is a valid structured-data format, but it does not need to duplicate an already correct JSON-LD or RDFa representation.
Schema.org RDFa
DiagnosticWhat is evaluated: Schema.org markup embedded in HTML through RDFa attributes.
Why it matters: RDFa is another valid structured-data format and should be evaluated as an alternative, not as a mandatory companion to JSON-LD.
Discovery and crawler rules
Search/discovery crawler access
AI ReadinessWhat is evaluated: robots.txt access for supported search/discovery crawler controls; answer-use, user retrieval and model-use policies are reported separately.
Why it matters: Search/discovery controls affect whether automated search systems can crawl public content, while training or model-use policy is a separate publisher choice and should not reduce this score.
Sitemap discovery in robots.txt
AI ReadinessWhat is evaluated: A valid Sitemap directive published in robots.txt.
Why it matters: A Sitemap directive gives automated systems a standard discovery path to the site’s published URL inventory.
Metadata and indexing controls
Canonical URL
AI ReadinessWhat is evaluated: An absolute canonical URL aligned with the final public document address.
Why it matters: Canonical metadata helps distinguish the preferred document from duplicates and technical URL variants.
Title element
AI ReadinessWhat is evaluated: A non-empty HTML title element.
Why it matters: The document title provides a concise identity for browsers, search systems and other metadata consumers.
Meta description
AI ReadinessWhat is evaluated: A non-empty page description meta tag.
Why it matters: A description gives machine consumers a concise summary of the page without replacing the primary content.
Robots meta
AI ReadinessWhat is evaluated: Indexing directives published in the page-level robots meta tag.
Why it matters: A noindex directive can prevent search discovery regardless of otherwise strong technical signals.
Open Graph metadata
AI ReadinessWhat is evaluated: Core Open Graph properties for the page identity and preview.
Why it matters: Consistent preview metadata makes the page identity more portable across systems that consume Open Graph.
X-Robots-Tag
AI ReadinessWhat is evaluated: Indexing directives published in the final HTTP response headers.
Why it matters: X-Robots-Tag can restrict indexing independently of the HTML robots meta tag.
Internal pages: publication consistency
AI Readiness when evaluatedWhat is evaluated: Sampled sitemap URLs are checked for agreement between their publication signals and preferred page address.
Why it matters: The sitemap should describe public pages consistently with the pages themselves.
Internal pages: basic metadata
AI Readiness when evaluatedWhat is evaluated: The sample checks document title and description presence plus duplicate titles between sampled pages.
Why it matters: Page identity and concise document context should exist beyond the homepage.
Technical accessibility
Homepage response
AI ReadinessWhat is evaluated: A usable public HTML response from the site homepage.
Why it matters: A failed or non-HTML homepage does not provide a defensible base document for the rest of the technical assessment.
robots.txt
AI ReadinessWhat is evaluated: Availability and basic readability of the public robots.txt file.
Why it matters: robots.txt is the standard publication point for crawl rules and Sitemap discovery.
sitemap.xml
AI ReadinessWhat is evaluated: Availability and basic readability of the public XML sitemap.
Why it matters: A sitemap helps automated systems discover public URLs without implying that those URLs will be indexed or ranked.
Document HTTP headers
AI ReadinessWhat is evaluated: An interpretable HTML Content-Type and related document-delivery headers.
Why it matters: Incorrect response metadata can prevent reliable automated interpretation even when HTML is present.
Content without mandatory JavaScript
AI ReadinessWhat is evaluated: Useful page content present in the received HTML without executing external JavaScript.
Why it matters: Not every automated consumer executes client-side JavaScript before reading a page.
Internal pages: accessibility
AI Readiness when evaluatedWhat is evaluated: A bounded sample of up to two sitemap URLs is checked for usable HTML and useful initial content.
Why it matters: A successful homepage does not prove that internal site pages are equally reachable to automated systems.
Delivery Reliability
Redirect chain
Delivery ReliabilityWhat is evaluated: The public redirect path, including origin changes and protocol downgrades.
Why it matters: Redirects determine the actual document URL and can cross trust or transport boundaries before content is reached.
HTTPS on the final URL
Delivery ReliabilityWhat is evaluated: Delivery of the final page over HTTPS.
Why it matters: HTTPS protects response integrity and confidentiality between the site and the client.
HSTS
Delivery ReliabilityWhat is evaluated: Strict-Transport-Security on HTTPS responses.
Why it matters: HSTS tells supporting browsers to use HTTPS for subsequent requests and reduces downgrade exposure.
Content-Security-Policy
Delivery ReliabilityWhat is evaluated: A Content-Security-Policy response header.
Why it matters: CSP constrains allowed resource and execution sources and can reduce the impact of content-injection attacks.
Framing protection
Delivery ReliabilityWhat is evaluated: Public restrictions on embedding the page in frames.
Why it matters: frame-ancestors or X-Frame-Options can reduce clickjacking and unwanted framing of the interface.
X-Content-Type-Options
Delivery ReliabilityWhat is evaluated: The nosniff response directive.
Why it matters: nosniff tells browsers not to reinterpret resources as a different content type from the declared Content-Type.
Referrer-Policy
Delivery ReliabilityWhat is evaluated: A policy controlling referrer information sent to other origins.
Why it matters: Referrer-Policy limits unintended disclosure of source URLs and their parameters to external recipients.
Permissions-Policy
DiagnosticWhat is evaluated: A policy restricting selected browser capabilities.
Why it matters: Permissions-Policy can reduce unnecessary access to browser features, although it is diagnostic and does not affect AI Readiness.
Mixed content in HTML
Delivery ReliabilityWhat is evaluated: Explicit HTTP resource references inside an HTTPS page.
Why it matters: Insecure subresources can be blocked by browsers and weaken the integrity of an otherwise HTTPS page.
Form action safety
Delivery ReliabilityWhat is evaluated: Forms on HTTPS pages that submit to an insecure HTTP URL.
Why it matters: Submitting form data over HTTP can expose or alter user input in transit.
Optional machine-readable signals
Browser WebMCP
OptionalWhat is evaluated: Statically detectable browser-side WebMCP tool registration in HTML or available inline script.
Why it matters: Browser WebMCP exposes page-level tools to browser agents and is distinct from a remote MCP server. It remains optional and does not affect AI Readiness.
llms.txt
OptionalWhat is evaluated: Availability of the proposed /llms.txt machine-context document.
Why it matters: llms.txt can provide additional machine-readable navigation context, but its presence does not guarantee discovery, indexing or citation.
llms-full.txt
OptionalWhat is evaluated: Availability of an extended proposed LLM context document.
Why it matters: llms-full.txt can expose more machine-readable context, but it remains optional and does not guarantee use by an AI system.
Agent interaction semantics
OptionalWhat is evaluated: Static coverage of native interactive controls, accessible names and selected ARIA roles and states.
Why it matters: Browser agents can use accessible names, roles and states to interpret actions, but this diagnostic is not a ranking factor and does not replace a browser accessibility audit.
Remote MCP server
OptionalWhat is evaluated: The conventional /mcp/ endpoint is probed with server/discover for MCP 2026-07-28; tools/list is read only when the server advertises tools capability.
Why it matters: Remote MCP is a server-side agent interface, distinct from Browser WebMCP. Its presence is an optional capability and does not affect AI Readiness.
AI-commerce: Product data
Catalog discovery and the completeness of machine-readable descriptions for available products.
Confirmed store context
Contextual criterionWhat is evaluated: An unambiguous public catalog and product-page context for the separate commerce assessment.
When it applies: The remaining commerce criteria apply only after a store context has been confirmed.
Product catalog
Stable criterionWhat is evaluated: Discoverability of public pages for individual products from the website signals already available.
Product structured data
Stable criterionWhat is evaluated: Readable Product or ProductGroup markup with the offer associated with the product.
Name, description and image
Stable criterionWhat is evaluated: The core descriptive product fields in the machine-readable representation.
Price and currency
Stable criterionWhat is evaluated: An unambiguous purchase price and currency code for the specific product offer.
Availability
Stable criterionWhat is evaluated: Machine-readable availability of the specific product offer.
Brand
Contextual criterionWhat is evaluated: Brand information where it is applicable and published for the product.
Product identifiers
Stable criterionWhat is evaluated: A supported product identifier, such as SKU, GTIN or MPN.
Product URL and canonical
Stable criterionWhat is evaluated: A consistent canonical URL for the public product page.
Variant mapping
Contextual criterionWhat is evaluated: The machine-readable relationship between product variants and their common product model.
When it applies: Applies only when variants or ProductGroup are detected.
Product card completeness
Contextual criterionWhat is evaluated: A summary of how many core product fields could be assessed in the bounded sample.
When it applies: Depends on the availability and unambiguous interpretation of the selected product pages.
AI-commerce: Consistency and commerce policies
Agreement between visible product content and markup, shipping and returns, policy links and published feeds.
Visible/structured name consistency
Contextual criterionWhat is evaluated: The visible product name compared with the name in unambiguous structured markup.
Visible/JSON-LD price consistency
Contextual criterionWhat is evaluated: The visible purchase price compared with the structured price for the same offer.
Visible/structured currency consistency
Contextual criterionWhat is evaluated: The visible currency compared with the currency code in the structured offer.
Visible/JSON-LD availability consistency
Contextual criterionWhat is evaluated: The visible availability compared with the value in structured data.
Shipping information
Stable criterionWhat is evaluated: A public machine-readable description of shipping conditions.
Return policy
Stable criterionWhat is evaluated: A public machine-readable description of return conditions.
Commerce policy links
Contextual criterionWhat is evaluated: The validity of published shipping and return-policy links without fetching their destination pages.
When it applies: Applies only to explicitly published policy links.
Product feed discovery
Contextual criterionWhat is evaluated: An explicitly published public product feed without fetching its content.
When it applies: The absence of an optional feed is not a critical error.
Product feed freshness
Contextual criterionWhat is evaluated: An available update-time signal for the published product feed.
When it applies: Applies only when a feed is published and its update signal is unambiguous.
AI-commerce: Agentic discovery
Optional public UCP, A2A, OpenAPI and MCP declarations without executing operations.
UCP profile
Experimental criterionWhat is evaluated: The structure of an explicitly published versioned Universal Commerce Protocol profile.
When it applies: The absence of UCP is not a critical error; operations are not executed.
A2A Agent Card
Experimental criterionWhat is evaluated: The structure of an explicitly published public A2A Agent Card.
When it applies: The absence of A2A is not a critical error; agent operations are not executed.
OpenAPI discovery
Contextual criterionWhat is evaluated: An explicit OpenAPI declaration in a published agent document.
When it applies: The linked specification is not fetched, and its absence is not a critical error.
MCP declaration
Experimental criterionWhat is evaluated: An explicit MCP server declaration in a published agent document.
When it applies: MCP tools are not called; the absence of a declaration is not a critical error.
Agent endpoint safety
Contextual criterionWhat is evaluated: Whether explicitly published agent addresses are suitable for public use and avoid obviously private or credential-bearing values.
When it applies: Applies only to endpoints published in the agent documents being checked.
Version history
- 1.0 — 3 August 2026. Initial reproducible technical scale.
- 2.0 — 3 August 2026. Expanded semantics, AI discovery and metadata coverage.
- 3.0 — 3 August 2026. Stricter context gates, delivery security inside the general technical score and separate experimental coverage.
- 4.0 — 8 August 2026. AI Readiness separated from Delivery Reliability.
- 4.1 — 11 August 2026. Crawler-policy handling was clarified and agent interaction semantics were added as optional diagnostics.
- 4.2 — 14 August 2026. Added bounded internal-page readiness sampling from sitemap URLs.
- 4.3 — 18 September 2026. Separated Browser WebMCP from Remote MCP and added bounded Remote MCP discovery without tool execution or any AI Readiness score change.