Conformance

The three conformance classes, publisher and validator conformance, and Directory verification.

Status Stable

This specification is published and current. Future phases will add a formal validator-output schema, certification badges, and an open test suite. See specification conventions for status definitions.

Last updated:

Purpose

This page is the single source for what it means to "conform" to the AI Discovery Files specification. It defines the three conformance classes, the requirements a publisher MUST meet to claim each class, and the requirements a validator MUST meet to assess publishers.

1. What conformance means

Conformance is the relationship between something — a website's published files, or a tool that tests them — and the requirements stated in this specification. A publisher's website conforms to a class when it meets the file set and validation requirements for that class. A validator conforms when it correctly assesses publishers against the same requirements.

The specification distinguishes two kinds of conformance:

Both kinds are independent. A publisher can claim conformance without using a particular validator. A validator can be conformant without being operated by any particular publisher. The two meet when a validator tests a publisher and the result is recorded.

2. The three conformance classes

Conformance is defined at three progressive levels. Higher classes include all the requirements of lower classes; you cannot claim Recommended conformance without also meeting Essential.

Essential

2 files. Minimum viable AI visibility.

  • llms.txt — business identity and context
  • ai.txt — AI usage permissions

A publisher MUST publish both files at the website's root. Both files MUST conform to their respective specifications. The business name in llms.txt (the H1 heading) MUST be consistent with any name used elsewhere on the site.

Suitable for: sites that have not previously published AI Discovery Files and want a minimal, defensible starting point.

6 files. Solid foundation for AI visibility.

A publisher MUST publish all six files at the website's root. Each file MUST conform to its specification. The publisher MUST maintain identity consistency across files: the business name in identity.json, llms.txt, and brand.txt MUST be the same canonical form (or correctly cross-referenced via alternateNames). Permissions in ai.json MUST NOT contradict declarations in ai.txt.

Suitable for: businesses serious about AI visibility, organisations whose AI representation matters for trust, and any publisher who wants AI systems to find structured information they can reliably extract.

Complete

10 files. Full coverage.

  • All Recommended files (above), plus:
  • llm.txt — redirect variant for compatibility
  • llms.html — human-readable HTML mirror
  • developer-ai.txt — developer-facing technical context
  • robots-ai.txt — AI-specific crawler directives

A publisher MUST publish all ten files at the website's root. Each file MUST conform to its specification. The relationships defined in the Interoperability Guide MUST hold (e.g. llm.txt SHOULD redirect to llms.txt; robots-ai.txt directives MUST NOT contradict ai.txt permissions). Identity and permissions consistency requirements from Recommended also apply.

Suitable for: organisations with a comprehensive AI visibility programme, multi-product publishers, and sites that want maximum surface area for AI systems to find authoritative information.

3. Publisher conformance

A publisher conforms to class N when, at the time of assessment, the following all hold:

  1. File set: all files required by class N are accessible at the website's root over HTTPS, return HTTP 200, and serve the appropriate media type for each file.
  2. Per-file validity: each file passes the validation rules defined in its individual specification page (see the specifications hub for links).
  3. Identity consistency: business name and primary URL are identical across files (or correctly cross-referenced via alternateNames).
  4. No contradictions: the precedence and conflict-resolution rules in the Interoperability Guide are satisfied.

Conformance is assessed at a point in time. A publisher conforming today may stop conforming tomorrow if files change, become inaccessible, or develop contradictions. Publishers SHOULD periodically re-check.

Class membership is monotonic: a Complete-conforming publisher also conforms to Recommended and Essential.

4. Validator conformance

A validator conforms to this specification when it can correctly assess publisher conformance to any of the three classes. A conformant validator MUST:

  1. Identify the file set required for the class being tested.
  2. Run each spec's validation rules against each present file.
  3. Detect identity inconsistencies (business name mismatches not reconciled via alternateNames).
  4. Detect contradictions per the Interoperability Guide.
  5. Report the result deterministically — the same input MUST yield the same output across runs.
  6. Distinguish "publisher conforms to class N" from "publisher does not conform to class N" without ambiguity.

A validator MAY additionally produce diagnostic output (warnings, suggestions, partial-pass details) but the core conformance decision MUST be a definite pass/fail per class.

The AI Visibility Checker is the reference implementation. The standard validator-output schema is published at validator-output.schema.json. Conformant validators emit output structured per that schema, enabling cross-validator comparability.

5. Self-declaration

A publisher MAY self-declare conformance to a named class without third-party verification. Self-declaration is a public claim made by the publisher (typically in marketing materials, technical documentation, or a website footer) that their site meets the requirements of class N.

Self-declaration carries no certification mark and no badge. The publisher takes responsibility for the accuracy of the claim. Anyone reading the claim can independently verify it by running a conformant validator against the publisher's site.

6. Directory verification

The AI Visibility Directory is the canonical conformance registry for this specification. A publisher MAY submit their site to the Directory, which tests the site using the reference implementation (the AI Visibility Checker) and records the result. The Directory's formal role, verification process, re-check cadence, dispute resolution, and removal policy are documented in detail at The Conformance Registry.

Directory-verified sites are listed publicly with their assessed conformance class. Verification is rechecked on a quarterly cadence; sites that drift below their declared class are notified and downgraded.

Directory verification is the only path to a verification badge. Self-declared publishers may use the phrase "AI Discovery Files conformant" in their own materials per the licensing policy, but MUST NOT display Directory-issued badges unless the Directory has verified them.

7. Certification marks

The Directory issues three visual certification badges for verified publishers, one per conformance class. Each badge is a single SVG served from a stable URL on this site, and the recommended embed snippet links the badge to a per-publisher public verification record at /directory/verify/<slug>/. Verified publishers obtain a copy-paste embed snippet from their Directory dashboard.

Display rules, sizing guidance, anti-fraud measures, and the full embed pattern are documented at Conformance Certification Badges. The summary version:

8. Changes to conformance classes

Conformance class definitions are STABLE. Once published, the file set required by a class will not change in a backwards-incompatible way within a MAJOR version. Adding a new file to a class is a breaking change because previously-conforming publishers stop conforming; such changes will only happen at MAJOR-version boundaries with deprecation notice.

The names of the classes — Essential, Recommended, Complete — are part of the specification's public surface and will not be renamed.

References