MLI Methodology

I1 — Organization identity declaration

What it measures: Whether the organization is declared in machine-readable form via JSON-LD Organization or LocalBusiness schema, with verified identity links.

Why it matters: An index has to resolve the name on a page to an entity — one durable record, distinct from the organizations it collides with. Without a first-party Organization declaration, that resolution is assembled from whatever third-party material exists: aggregator listings, review platforms, social profiles, news coverage. With one, the site supplies the record itself, and sameAs supplies the reconciliation instruction that ties it to public records already held. This is what the vocabulary is specified for and what crawler operators document consuming. It is an index-stage function, unaffected by what a model reads during a live fetch.

Public-interest weight: High. Community organizations without an Organization declaration are more likely to be characterized by aggregators and directories than by themselves, and are more exposed to name collision — a legal aid clinic sharing a name with a commercial firm, a coalition merged with its own member organizations.

Commercial weight: High. Multi-location brands, franchises, and post-rebrand entities are the common failure case: several plausible records, no declared authority over which is current.

Signals

• JSON-LD contains Organization, LocalBusiness, NGO, or another organization subtype
• Properties present: name, url, logo, sameAs, description, address
• sameAs points to verified profiles or official records

I2 — Service-type specificity Very high public-interest stakes

What it measures: Whether the page declares the correct schema.org service-type for what the organization actually does — not just generic Organization.

Why it matters: An agent answering "where can I get free legal help in Maryland in Spanish" needs a service-type axis to filter on. Generic Organization schema offers none. LegalService does. The same applies to GovernmentService, SocialService, MedicalClinic, and dozens of other subtypes. This is index-stage and action-stage work: the criterion is scored for what a declared subtype makes possible, not for a measured ranking effect. A subtype carrying filterable properties is what lets an agent return the service itself rather than a page that discusses it.

The same mechanism shapes commercial discovery. An agent answering "find me a same-day plumber in Oakland" needs Plumber or HomeAndConstructionBusiness — not a generic Organization.

Public-interest weight: Very high. Without a specific service-type schema, clinics rank below service-typed peers in service-type-filtered queries and sit in the long tail of undifferentiated generic Organization results. The cost lands hardest on task completion: a person on a filing deadline who receives a landing page instead of the eligibility rule has an answer-shaped object, not an answer.

Commercial weight: High. The same mechanism governs whether a booking, comparison, or purchase agent can assemble an offer from the site at all, or has to fall back on a typed competitor.

Signals

• jsonLdTypes includes a service-specific type appropriate to the page content
• Required properties for that subtype (e.g., LegalService should have serviceType, areaServed, provider)

I3 — Source authority and credentials Very high public-interest stakes

What it measures: Whether the site declares the authority signals that distinguish authoritative voices from predatory ones — accreditation, parent organization, professional credentials, government TLD, free-access status.

Why it matters: Machine legibility alone doesn't distinguish a BIA-accredited immigration legal services organization from a notario. (Notarios are unlicensed practitioners who exploit a name collision with Latin American notarios públicos to advertise legal services they aren't qualified to provide.) Commercial immigration service providers — both licensed and unlicensed — invest in search marketing in ways that underfunded community legal aid organizations typically cannot. Declaring accreditation, parent organization, credentials, and free-access status makes the distinction available in machine-readable form rather than leaving it to be inferred from prose — and available to the person reading the answer, who can follow accreditedBy to the accrediting body. Nothing here guarantees an agent weights those signals; the criterion scores whether the distinction is declared at all.

Public-interest weight: Very high. This is the criterion that keeps the rest of the pillar from being a legibility contest. Without declared authority, the qualified and the unqualified are equally well-formed records, and an agent has nothing in the data to separate them.

Commercial weight: Medium. Licensure, certification, and parent-company signals matter in regulated sectors — finance, healthcare, trades — and are largely inert elsewhere; score honestly rather than rewarding decorative award properties.

Signals

• Schema for hasCredential, accreditedBy, parentOrganization, funder, award
• isAccessibleForFree: true where applicable
• TLD signals (.gov, .edu, .org consistent with mission)
• HTML links to accrediting bodies (BIA, ABA, state bars, accreditation councils)
• nonprofitStatus for tax-exempt orgs

R1 — Agent contract

What it measures: Whether the site declares an explicit contract for AI agents and crawlers.

Why it matters: Without robots.txt agent directives, an agent has no signal whether it's welcome, what it can fetch, and what's off-limits. llms.txt provides an AI-readable site index. sitemap.xml aids discovery. These are the most direct findability signals and are missing from most under-resourced sites.

Signals

• robots.txt accessibility (200 response at /robots.txt)
• Presence of agent-specific User-agents (GPTBot, ClaudeBot, Google-Extended, PerplexityBot, anthropic-ai, cohere-ai)
• llms.txt or llms-full.txt at root
• sitemap.xml declared in robots.txt or accessible at root

R2 — Initial HTML completeness

What it measures: Whether substantive content — text, forms, action paths — is present in the initial HTML, or gated behind JavaScript or third-party iframes.

Why it matters: Agents that don't execute JavaScript miss content added post-load. SPAs that render to empty shells appear blank to crawlers that don't execute JavaScript. Forms that live in third-party iframes can't be filled by an agent operating on the parent page. Sites built on frameworks that render primarily client-side, and pages whose action paths live in third-party iframes, often fail this criterion. Server-rendered pages with native forms typically pass it.

Signals

• Semantic element count in initial HTML
• Form count (forms in initial HTML, not iframe-embedded)
• Iframe count and what's inside them
• noscriptHasContent (substantial content in <noscript> indicates JS dependency)

R3 — Multilingual reach Very high public-interest stakes

What it measures: Whether the site declares its language(s), serves alternate translations via hreflang, and declares language availability in service schema.

Why it matters: A site without explicit language declarations cannot be reliably identified or prioritized in language-filtered queries. Agents must infer language from content, creating slower, lower-confidence matches that rank below sites with declared language metadata. The single most important signal for the Notice-to-Appear scenario, and for any cross-language service-finding case. Community legal aid clinics frequently translate content but fail to mark translations with hreflang, losing ranking advantages in language-specific queries and reducing discoverability to speakers of those languages.

Commercial sites face the same penalty in cross-language queries. A SaaS company serving Latin American customers without hreflang on its Spanish docs ranks below competitors that have it — even when its product is better.

Signals

• <html lang="..."> set
• <link rel="alternate" hreflang="..."> to translated pages
• availableLanguage property in service schema
• contactPoint.availableLanguage for service contacts

Skip-if: Organization is genuinely English-only and its served audience is too. This rarely applies in US service contexts and should be a documented exception.

S1 — Page type and breadcrumb

What it measures: Whether the page declares its schema.org type (Article, WebPage, ContactPage, FAQPage, Service, etc.) and provides BreadcrumbList for hierarchical context.

Why it matters: An agent extracting an answer needs to know what kind of page it's reading. A FAQPage has different extraction logic than a Service page. BreadcrumbList lets the agent place the page in site context.

Signals

• jsonLdTypes includes a page-level type matching the actual content
• BreadcrumbList present with correctly-ordered items

S2 — Heading hierarchy as content map

What it measures: Whether headings form a parseable outline an agent can use to navigate the page's content.

Why it matters: Headings are how agents map a page. Multiple H1s, skipped levels, or non-descriptive headings make extraction significantly harder. Headings with stable IDs are deep-linkable, which improves citation precision.

Signals

• h1Count (single descriptive H1 expected)
• Logical H2/H3 progression; no skipped levels
• headingHierarchyIssue (multiple H1s, or H3s without H2s)
• headingsWithId ratio

S3 — Semantic density

What it measures: Ratio of semantic landmarks (main, nav, header, footer, section, article, aside) to non-semantic div/span elements.

Why it matters: Div soup is harder to parse than semantic HTML. Agents that map pages by landmarks get nothing from a page that's all <div>. This is where MX and accessibility most clearly converge — semantic landmarks serve both screen readers and agents.

Signals

• Semantic element count vs total div/span count
• Ratio: semanticElements / (semanticElements + totalDivSpan)
• Presence of all four major landmarks (main, nav, header, footer)
• nav[aria-label] for distinguishing multiple navs

C1 — Machine-readable dates

What it measures: Whether content has datePublished and dateModified in JSON-LD, plus appropriate <time datetime> elements.

Why it matters: Agents need to decide whether to surface information or treat it as stale. Without machine-readable dates, they guess from prose ("Updated: January 2025") — which is unreliable — or assume worst-case staleness and skip the page.

Signals

• datePublished in JSON-LD
• dateModified in JSON-LD
• <time datetime="..."> elements in HTML (ISO 8601 format)

C2 — Time-sensitive markup Very high public-interest stakes

What it measures: Whether deadlines, hours, event dates, and validity periods are machine-readable.

Why it matters: A Notice to Appear recipient needs deadline awareness. A clinic's walk-in hours, a registration deadline, a court date, an application window — all must be machine-readable or agents can't filter or alert appropriately. The same mechanism governs commercial time-sensitive content: a retailer's flash sale, a conference's early-bird deadline, a SaaS webinar registration. "We are open Tuesdays 9–12" in prose can't be matched by a query like "who can I see this Tuesday morning?" — the agent has nothing to filter on.

Signals

• Event schema with startDate/endDate
• validThrough on Offer / Service
• OpeningHoursSpecification on LocalBusiness
• eventStatus for cancelled/postponed events

Skip-if: Page has no time-sensitive content (e.g., a static About page with no hours, dates, or deadlines).

C3 — Eligibility, cost, and service availability Very high public-interest stakes

What it measures: Whether audience, cost, jurisdiction, and availability of services are declared in machine-readable form.

Why it matters: An agent filtering for "free immigration help in Maryland in Spanish" needs four facts: service type (I2), free/low-cost, jurisdiction, and language (R3 covers language; this criterion covers cost and jurisdiction). An agent filtering for "EV charging open after 10pm in Austin" or "gluten-free meal delivery serving Brooklyn" needs the same filterable structure — service type, jurisdiction, hours, and dietary or product attributes — declared in machine-readable form. Without this markup, the page can't be filtered on cost or jurisdiction — and ranks below pages that can be, in service-discovery queries. That is index-stage and action-stage filtering: whether the page can be selected against a constraint and then acted on. It is not a claim about how often a model cites the page while composing an answer.

Signals

• audience with audienceType, geographicArea, eligibility qualifications
• areaServed (city, state, county, jurisdiction)
• isAccessibleForFree: true for free services; priceSpecification or price: 0; sliding-scale via priceRange
• availableLanguage at service level (overlaps with R3)
• OfferCatalog for orgs with multiple service offerings
• Operational status: businessStatus (BusinessOperating, TemporarilyClosed, PermanentlyClosed) or Offer.availability (InStock, LimitedAvailability, SoldOut)

Skip-if: Page is purely informational with no service offered (e.g., a glossary, a news article).