Alpha Fresh

AI Usage Standard (AIUS)

Deterministic AI-involvement disclosure standard for written artifacts. Six dimensions × four levels → five-tier glance-readable badge, with strict frontmatter validation. Live across this site; mirror repo for adopters.

Usable but unstable. Breaking changes expected.

Started
May 2026
Updated
May 2026

Gathering feedback — open an issue with your thoughts.

Open in repo ↗
Roadmap5 / 8 — 54% · effort-weighted
  • Spec v1.0 — 10-article constitution
  • Astro components (full + mini variants)
  • Zod schema + build-time validation
  • /standards/ai-usage canonical page
  • GitHub mirror repo (Phase 1)
  • Phase 2 — npm package (@serkanaltuntas/ai-usage-badge) — after first external adopter or 6 months; Astro/React/Vue components
  • Phase 2 — JSON Schema CLI validator
  • Phase 3 — localized spec pages (/standards/ai-usage/tr, /fr) — after npm package stabilizes; canonical English unchanged

What

AIUS is a versioned, deterministic standard for declaring AI involvement in published artifacts (blog posts, project pages, papers, READMEs). Every artifact carries an <AIUsage /> badge derived from a frontmatter declaration of six dimensions, each at one of four levels, plus a list of models used.

The canonical spec lives at /standards/ai-usage. Anyone may adopt it under CC-BY-4.0 with attribution.

Why

Existing options each fall short:

  • CRediT covers human authorship roles only.
  • ICMJE / Nature / Springer / Elsevier policies require qualitative AI disclosure in Methods sections — no glance-readable signal.
  • AI Usage Cards (Wahle et al. 2023) is the closest prior art but uses boolean flags across three dimensions; AIUS extends to six dimensions × four levels with deterministic tier derivation.
  • C2PA / Content Credentials focuses on image provenance, not text or code authorship.

AIUS fills the gap of “at-a-glance disclosure with rule-based determinism” for written artifacts. A reader sees the bar meter and immediately understands the AI-involvement profile without reading prose. An adopter implements the spec from the canonical text and gets identical behavior across sites.

How

Three layers, all open-source:

  1. Spec layer — the constitution at /standards/ai-usage plus a machine-readable schema.json. CC-BY-4.0.
  2. Logic layer — a pure tier-derivation function and a frozen glyph table, fully unit-tested. MIT.
  3. Component layer<AIUsage /> (full bar meter) and <AIUsageMini /> (one-liner) Astro components, plus Zod-validated frontmatter that fails the build on any deviation from the spec.

Phase 2 plan

After Phase 1 stabilizes (or the first external adopter shows up):

  • Publish an npm package with Astro, React, and Vue components.
  • Ship a CLI that validates ai frontmatter against schema.json.
  • Tighten the JSON Schema’s allOf constraint with dependentRequired if validators support it.

Phase 3 plan

Once the npm package stabilizes and external adopters exist:

  • Add localized variants of the canonical page (/standards/ai-usage/tr, /standards/ai-usage/fr, etc.). The English version stays canonical; localized pages are translations, not redefinitions.

Repo

Mirror repo: github.com/serkanaltuntas/ai-usage-badge. Contains the spec text in markdown form, the JSON Schema, both license files, and a self-contained Astro reference example.

astrotypescriptzod