AI Usage Standard (AIUS)

TL;DR

AIUS is a small, rule-driven standard for declaring AI involvement in a written artifact (a blog post, a project page, a paper). Every artifact declares six dimensions of contribution at four levels each. A fixed derivation rule turns those declarations into one of five tiers, rendered as a glance-readable bar-meter badge. The same input always produces the same badge, regardless of who or what implements the spec.

Live example

The badge below renders directly from a frontmatter declaration. Same component, same data path used on every post and project on this site.

Why this exists

Existing options each fall short. CRediT covers human authorship roles only. ICMJE-family policies require qualitative AI disclosure in Methods sections, with no glance-readable signal. AI Usage Cards use boolean flags across three dimensions, with no derived top-level tier. AICAT is fine-grained but unadopted. C2PA / Content Credentials focuses on image provenance, not text or code authorship.

AIUS combines AI Usage Cards' dimensional approach with explicit level gradation and a deterministic top-level tier derivation, producing both an at-a-glance signal and inspectable detail.

The Constitution

Article 1 — Dimensions (fixed, six)

1. ideation — topic and scope generation
2. research — sources, literature, fact-finding
3. drafting — first-pass prose production
4. editing — polish, tone, grammar
5. code — code snippets, scripts, programs
6. visuals — diagrams, illustrations, generated imagery

Adding or removing a dimension constitutes a v2.0 break.

Article 2 — Levels (fixed, four)

Label	Integer	Meaning
none	0	Human only; no AI involvement
assist	1	AI offered suggestions; human accepted or rejected each
collab	2	Human and AI iterated jointly
lead	3	AI produced; human approved or refined

Article 3 — Tier Derivation (deterministic function)

input:  dims = {ideation, research, drafting, editing, code, visuals} ∈ {0,1,2,3}^6
intermediate:
  m       = max(dims)
  n_lead  = count(d ∈ dims | d == 3)
  n_none  = count(d ∈ dims | d == 0)

rule (apply in order, first match wins):
  n_none == 6                       → human-only
  m == 1                            → ai-assisted
  m == 2                            → co-authored
  m == 3 and n_lead <  4            → ai-led
  m == 3 and n_lead >= 4            → ai-generated

Article 4 — Mandatory Dimension Declaration

Every artifact declares all six dimensions explicitly. No defaults. A missing dimension is a build error.

Article 5 — Mandatory Model Attribution

If max(dims) ≥ 1, the models list must be non-empty. Each entry is an official model version identifier (e.g., claude-opus-4-7, gpt-5, gemini-2.5-pro). If max(dims) == 0, models is optional and ignored.

Article 6 — Tier Override Prohibition

The author must not set the tier manually. Tier is always the output of the derivation function in Article 3. A frontmatter tier field is a build error.

Article 7 — Glyph and Label Mapping (fixed)

Tier	Glyph	Label
human-only	◌	Human-Only
ai-assisted	◔	AI-Assisted
co-authored	◑	Co-Authored
ai-led	◕	AI-Led
ai-generated	●	AI-Generated

Canonical labels are the English strings exactly as listed. Localized variants are out of scope for v1.0; future versions may add them without breaking adopters.

Article 8 — Variants

Two variants are recognized: full (the bar meter, intended for post body or project header) and mini (single-line summary, intended for cards and list pages). Adding a third variant is a v2.0 break.

Article 9 — Licenses

• Spec text: CC-BY-4.0. Cite as: Disclosure follows AIUS v1.0 (https://serkan.ai/standards/ai-usage).
• Reference component code: MIT.

Article 10 — Versioning

• Backward-compatible additions → minor (v1.x).
• Backward-incompatible changes → major (v2.0).

Adopters may declare aius_version in their frontmatter to pin to a version. Default is 1.0.

JSON Schema

A machine-readable schema for the ai frontmatter object is available at /standards/ai-usage/schema.json. Validate adopter frontmatter against it directly.

Examples gallery

Human-only

Pure-human note. No AI involved.

AI-Assisted

Spell-check and source-finding assistance.

Co-Authored

Joint iteration on ideas, research, and prose.

AI-Led

AI produced the draft, code, and editing; human directed scope and reviewed.

AI-Generated

Every dimension led by AI, with human only approving the final result.

Adopting AIUS in your site

1. Read this spec.
2. Validate frontmatter against schema.json.
3. Implement deriveTier() per Article 3, or copy the reference TypeScript implementation.
4. Render a badge per Article 7's glyph and label table.
5. Cite as: Disclosure follows AIUS v1.0 (https://serkan.ai/standards/ai-usage).

A reference Astro implementation lives at github.com/serkanaltuntas/ai-usage-badge.

Versioning and history

v1.0 (2026-05-03) — initial release. Six dimensions, four levels, five tiers, two variants. Constitution articles 1–10.

License

This standard's text is licensed CC-BY-4.0. Reference component code is licensed MIT. Both licenses are available in the mirror repo.