AI Usage Scale

An independent, open vocabulary for declaring how generative AI contributed to a work and whether a person reviewed it before publication.

Website · Definitions · Find your level · Specification · Project status

The scale describes a process. A number is a publisher’s self-declaration, not a quality score, authorship certificate, detector result or legal compliance assessment. An absent declaration means the process is unknown.

Six levels

Level	Name	Description
0	Human	No generative AI was used.
1	Assisted	AI supported research or mechanically processed human material; no new model-originated material remains.
2	Co-created	The human made most of the final form; AI-created parts remain.
3	Directed	Original human substance materially determines the work; AI supplies most of the form, with meaningful human review.
4	Prompted	AI supplies substance and form; a human meaningfully reviews the work before publication.
5	Automated	AI supplies substance and form; no human reviews the output before publication.

Read the complete definitions and edge cases before declaring a level. Text, image, audio, video, code and data may have different levels; an optional surfaces object records those differences.

Known boundary: the current decision tree can suggest Level 3 before checking human review, although the definition requires it. That suggestion needs verification against the definition. Resolving the unreviewed, human-substance case requires an explicit specification decision; this revision preserves the existing level meanings.

Use a declaration

Choose a definition and link to it. For example, for a work that satisfies Level 3:

AI Usage Scale: Level 3 (Directed) — https://usagescale.org/3

Each level page includes disclosure sentences and self-contained SVG marks. Downloaded or inline marks need no request to this website. Hosted badge URLs remain available; loading them sends a request to the hosting provider.

Optional project metadata:

<meta name="ai-usage" content="3">
<meta name="ai-usage-standard" content="https://usagescale.org">
<link rel="ai-disclosure" href="https://usagescale.org/3">

These names are project conventions. For JSON-LD, identify the extension explicitly:

{
  "@context": {
    "@vocab": "https://schema.org/",
    "aiUsageLevel": "https://usagescale.org/spec#aiUsageLevel"
  },
  "@type": "CreativeWork",
  "aiUsageLevel": 3
}

aiUsageLevel is not a registered Schema.org property. The vocabulary itself uses the existing DefinedTermSet and DefinedTerm types. Neither representation creates a dedicated Google Search feature.

A declaration grants no training permission. Licensing, consent, terms and access controls are separate questions.

Status and evidence

The published vocabulary is version 1.3.1. English is the normative source. Core pages are available in 22 languages; the 21 translations are labelled as machine translations pending fluent human review. Some developer and policy references remain in English.

As of 10 October 2026, two CHAOSS repositories reference the project in related-tools lists. A diarie backlog records deferred interest. This audit verified no independent, public production declaration; references and interest are not deployments or endorsements. See the evidence.

This is an independent proposal. Suggested IPTC mappings, conditional W3C Community Group text mappings and C2PA comparisons do not imply acceptance by those organisations. The CG's 9 October 2026 report uses three text values; Levels 1 and 2 need additional scope/review information. Schema.org issue #3391 is a relevant public discussion; submissions and pilots are tracked publicly.

The current website revision was produced with AI assistance and awaits human editorial review. It does not automatically declare its own content to be Level 3.

Machine-readable files

Endpoint	Purpose
levels.json	Definitions, questions, edge cases and suggested mappings; localised copies at /<lang>/levels.json.
spec.schema.json	JSON Schema for the vocabulary document.
levels.schema.json	JSON Schema for a declaration.
llms.txt / llms-full.txt	Optional plain-text references for tools.
feed.xml	Published release history.

spec/levels.json is the canonical source. Pin a release when reproducing a declaration. The schemas validate structure and consistency, not the truth of a publisher’s claim.

The repository includes a local MCP server. The package is not currently published on npm; install it from source or the archive attached to release v1.3.1. It exposes classification, level lookup and the full vocabulary over stdio.

Development

Requires Node.js 22.12 or later; Node.js 24 is also supported.

npm ci
npm run dev
npm run verify

verify checks Astro types, builds the static site and generated assets, then validates the vocabulary, translations, classifier, schemas and shipped HTML.

For browser checks:

npx playwright install chromium
npm run test:browser

The browser script serves dist/ locally and checks layout, keyboard operation, the wizard, themes and accessibility. With an existing Chrome installation, use BROWSER_CHANNEL=chrome npm run test:browser.

The site is static Astro, with local fonts, a light/dark theme and Arabic/Persian directionality. All six definitions and the decision procedure remain readable without JavaScript. CI verifies pull requests; production deployment uses Cloudflare Pages.

Contribute and licence

Useful contributions include a documented workflow that exposes a boundary, an independent integration, or a translation reviewed against the English source. See Contributing, Governance and Roadmap.

Maintained by Konrad Karauda. Decisions and changes are recorded in public issues and pull requests.

• Scale definitions and SVG marks: CC0-1.0.
• Website explanatory prose and documentation: CC BY 4.0.
• Website and MCP code: MIT.
• IBM Plex Mono and Spectral fonts: SIL Open Font License 1.1, with notices in src/fonts/.

See LICENSES.md for scope and CITATION.cff for citation metadata. The original manifesto is retained as a historical document; it is no longer the homepage.