Every check on this tool has a reason. Every dimension maps to a documented behaviour of AI recommendation systems. This page explains the criteria, the framework they are built on, and what the score does — and does not — tell you.
llms.txt is a plain-text file placed at the root of a website —
yourdomain.com/llms.txt — that provides structured, machine-readable
context about the entity behind the site. It is the intentional signal layer
between your website and AI systems that read it.
The format was proposed in 2024 and has since been adopted by publishers,
researchers, and practitioners building for AI-mediated discovery. Unlike
robots.txt (which tells crawlers what not to read) and
sitemap.xml (which tells crawlers where to go), llms.txt
tells AI systems who you are, what you do, and how to interpret your content.
llms.txt reduces ambiguity at the point of AI interpretation.
The shift from search-result visibility to AI-answer visibility is structural,
not cosmetic. In search, a business needed to rank in a list. In AI-mediated
discovery, a business needs to be recommended — which requires the AI
to have sufficient confidence in the entity's identity, authority, and relevance.
llms.txt is how you declare that confidence-building data directly.
This tool validates whether your llms.txt file is structured to
support that confidence-building process. It does not test what AI systems
currently believe about you — that is Phase 1. It tests whether your declaration
is complete, well-formed, and internally consistent.
The 8 dimensions are not arbitrary categories. They map to the distinct types of signals AI recommendation systems use to evaluate whether an entity is trustworthy, authoritative, and relevant enough to surface in a synthesised response.
Three of the dimensions — Identity Clarity, Topical Authority, and Cross-Source Trust — map directly to the three pillars of the ESC Framework (Entity Clarity · Semantic Authority · Cross-Source Trust), published independently at ShodhDynamics.com. The remaining five dimensions address the structural and operational requirements that make the ESC signals readable and usable.
The dimension weights reflect this logic: checks that establish foundational identity carry more points than checks that refine or extend it. Failing Identity Clarity checks costs more than failing optional enrichment checks — because the consequence in AI systems is correspondingly more severe.
| DIMENSION | ESC MAPPING | ROLE IN AI RECOMMENDATION |
|---|---|---|
| Structure | Prerequisite | File must be parseable. Malformed files are skipped silently. |
| Identity Clarity | E — Entity Clarity | AI must unambiguously identify the entity before it can recommend it. |
| Content Precision | E — Entity Clarity | AI extracts service and summary data to answer "what does this business do." |
| Cross-Source Trust | C — Cross-Source Trust | Multiple corroborating sources reduce AI hallucination risk on entity facts. |
| Relationship Completeness | Structural | Author/publisher chains establish content ownership — critical for citation. |
| Topical Authority | S — Semantic Authority | Topic and term declarations map the entity's expertise domain explicitly. |
| Content Navigability | S — Semantic Authority | Articles and featured pages give AI a content inventory to reference. |
| Temporal Currency | Recency Signal | Publication dates signal an active entity — stale content lowers AI confidence. |
Below is every dimension, what it measures, why AI systems require it, and the specific checks used to evaluate it. Point values reflect relative importance in the AI recommendation candidacy model.
Validates that the file follows the expected section format and ordering. AI parsers and LLM context ingestion pipelines read llms.txt files programmatically. A file with missing sections or incorrect ordering may be partially parsed or ignored entirely — silently, without error.
AI systems require this because: structured files allow deterministic extraction of entity signals. A well-ordered file with all required sections ensures the AI receives a complete, unambiguous context package, not a partial one.
The highest-weighted dimension. Validates that the entity is unambiguously declared — correct type (Person or Organization), substantive description (minimum 120 characters), geographic service area, and canonical IDs. Identity Clarity is the prerequisite for every other signal in the file.
AI systems require this because: entity disambiguation is the first step in any knowledge graph resolution process. If an AI cannot confidently identify who the entity is, it cannot confidently recommend them. Ambiguous or thin entity declarations result in the AI defaulting to competitors with clearer identity signals.
Validates the quality and substance of descriptive content — the SUMMARY section and, where applicable, the SERVICES section. The SUMMARY must be a substantive entity description (minimum 150 characters), not a tagline or marketing headline.
AI systems require this because: when a user asks "what does [business] do?", the AI extracts the answer from the SUMMARY and service descriptions in the llms.txt file. A 101-character summary is still a tagline. A tagline tells the AI almost nothing actionable about the entity's actual function. The test detects tagline patterns specifically.
Validates the presence of sameAs profile declarations — the Also at: lines in the PRIORITY ENTITY section. Minimum three external profile URLs required. These declarations tell AI systems where to find corroborating information about the entity across independent sources.
AI systems require this because: hallucination risk on entity facts decreases when multiple independent sources corroborate the same claims. An entity that exists only on its own website provides no corroboration signal. An entity with declared LinkedIn, GitHub, and Gravatar profiles gives the AI three independent verification points, which increases confidence in entity claims and reduces fabrication risk.
Validates the ENTITY RELATIONSHIPS section — specifically author chains, publisher chains, and person-organisation links. These relationships map content ownership and organisational structure in a form AI systems can parse and cite.
AI systems require this because: when an AI cites content, it attributes authorship. If no author chain is declared, the AI cannot attribute the content — which reduces citation likelihood and may attribute the content to no one or to the wrong entity. Publisher chains establish organisational endorsement of content claims. Person-organisation links are critical for personal brand entities where the individual and the business are distinct but related.
Validates CORE TOPICS, KEY TERMS, and FRAMEWORKS declarations. Core topics must number between 3 and 15 — fewer signals insufficient expertise; more than 15 dilutes the authority signal. Key terms must include descriptions, not just names. Frameworks declaration is weighted to reflect the value of original IP in AI authority inference.
AI systems require this because: topical authority is how AI systems decide which entity is the most credible answer to a domain-specific question. A business that explicitly maps its expertise through structured topic and term declarations gives the AI a navigable expertise model. Without it, the AI defaults to whichever entity it has the most general knowledge about — typically the most prominent competitor, not the most relevant expert.
Validates ARTICLES, CANONICAL IDS, and FEATURED PAGES sections. Requires a minimum of 3 articles with descriptions. Each article should carry a description so AI systems can summarise the content without reading the full page. Canonical IDs anchor the entity graph and allow AI systems to reference specific content objects with precision.
AI systems require this because: AI recommendation depends on having a navigable content inventory. A business with no declared articles or pages gives the AI no evidence of content production — which reduces perceived authority regardless of what actually exists on the site. The articles section is not about indexing; it is about making the evidence of expertise visible to AI in a structured, parseable form.
Validates that the most recently published article has a publication date within the last 90 days. Requires ISO 8601 date format in Published: fields. Stale content — the most recent article older than 90 days — is flagged with a note that the entity may appear inactive to AI systems.
AI systems require this because: recency is a proxy for entity activity. An entity that has not published in over three months sends a lower confidence signal than one that published last week. AI systems trained on time-stamped corpora have implicit temporal weighting. Declaring recent, dated content is a direct input into that recency signal — and the absence of dates removes the signal entirely.
The ESC Framework was developed independently by Anurag Gupta at ShodhDynamics.com to describe the three structural conditions AI systems appear to require before surfacing an entity in a synthesised recommendation response. It is a practitioner-developed framework grounded in observable AI behaviour, not a formal academic standard.
| PILLAR | WHAT IT DESCRIBES | FAILURE CONSEQUENCE |
|---|---|---|
| E — Entity Clarity | The AI can unambiguously identify who the entity is, what type it is, where it operates, and what it does. Name, type, description, geography, and canonical IDs all contribute. | AI either misidentifies the entity, conflates it with a similarly-named entity, or omits it in favour of a more clearly declared competitor. |
| S — Semantic Authority | The AI associates the entity with a specific, coherent domain of expertise. Topics, terms, frameworks, and structured content inventory all contribute to authority inference. | AI treats the entity as a generic provider in a broad category rather than a specialist. Generic positioning loses to specialist positioning in AI recommendation decisions. |
| C — Cross-Source Trust | The AI can verify entity claims against multiple independent sources. Profile declarations, external citations, and third-party mentions all contribute to corroboration density. | AI assigns lower confidence to unverified claims. Entities with low cross-source corroboration are more likely to be misrepresented or omitted. Hallucination risk increases. |
The three pillars are interdependent. Entity Clarity is the prerequisite — without it, Semantic Authority and Cross-Source Trust cannot attach to a stable entity reference. Semantic Authority without Cross-Source Trust produces an entity that appears expert but unverified. Cross-Source Trust without Entity Clarity produces corroboration that cannot be attributed.
A complete, well-formed llms.txt addresses all three pillars
simultaneously, which is why the scoring model weights Identity Clarity checks
highest — they establish the foundation on which the other signals rest.
The score is a declaration quality index — it measures how completely and
correctly your llms.txt file communicates entity signals to AI
systems. It is not a guarantee of AI recommendation; it is a measure of
whether the prerequisite conditions for recommendation candidacy are in place.
llms.txt file provides AI systems with a complete, unambiguous, corroborated entity context. Recommendation candidacy conditions are in place — AI perception is now the variable (Phase 1).Honest scope definition is part of a credible methodology. The following are explicitly outside the scope of the Phase 0 validator — not because they are unimportant, but because they require different tooling, different data sources, or are addressed in Phase 1.
llms.txt file.
llms.txt declarations. Inconsistency between declared
and actual profile data is a separate audit — also addressed in Phase 1.
The methodology behind this tool draws on published frameworks, academic research, and practitioner documentation. The links below are the primary sources behind the scoring model.