# AI Usage Scale

> A free, open, six-level scale for declaring how a work was made — whose substance it carries, and who stands behind it. Not how many characters a model emitted. Version 1.0.0-draft. CC0.

The problem is not AI. It is that the label has two settings, and shame is attached to one of them — so honesty is punished, silence is free, and silence wins. Research names this the disclosure paradox: people say they want disclosure, then rate disclosed work lower. The penalty runs through *perceived effort*, not quality (readers estimate 148 minutes for a human-labelled story and six for the same story labelled AI). A switch cannot communicate effort. A scale can.

**No level ranks above another.** Level 5 is the honest declaration for an automated market report; Level 0 is the honest declaration for a memoir. The only dishonest level is an undeclared one.

## The six levels

- [Level 0 — Human](https://usagescale.org/0): No generative AI was used.
- [Level 1 — Assisted](https://usagescale.org/1): AI worked only on material you had already made. It invented nothing.
- [Level 2 — Co-created](https://usagescale.org/2): You made most of the final form. AI made parts of it that stayed.
- [Level 3 — Directed](https://usagescale.org/3): The substance is yours. AI produced the form.
- [Level 4 — Prompted](https://usagescale.org/4): AI produced the substance and the form. You asked, you checked, you are accountable.
- [Level 5 — Automated](https://usagescale.org/5): AI produced the substance and the form. No human read it before it was published.

## The decision procedure

Five yes/no questions, asked in order. The first that terminates gives the level.

1. Was any generative AI used in making this work? — yes → next question; no → Level 0
2. Did AI produce new material that appears in the published work — beyond mechanically processing what you had already made? — yes → next question; no → Level 1
3. Did you make most of the final form yourself — most of the words, pixels, or sound? — yes → Level 2; no → next question
4. Is the substance yours — the knowledge, the data, the argument, the direction? — yes → Level 3; no → next question
5. Did a human review this before publication and take responsibility for what it says? — yes → Level 4; no → Level 5

## Docs

- [The whole standard in one file](https://usagescale.org/llms-full.txt): manifesto, levels, decision tree, edge cases and mappings. Start here.
- [The manifesto](https://usagescale.org/): fifteen theses, with sources.
- [The scale](https://usagescale.org/scale): all six levels with definitions and examples.
- [Specification](https://usagescale.org/spec): edge cases and interoperability mappings.
- [levels.json](https://usagescale.org/levels.json): canonical English definitions and mappings, machine-readable, CC0.
- [Find your level](https://usagescale.org/find-your-level): the five questions, as a tool.
- [Badges](https://usagescale.org/badges): the marks. Inline SVG, no external font, no request home.
- [Integrations](https://usagescale.org/integrations): every machine endpoint, plus an MCP server (`usagescale-mcp`) an agent can call directly.
- [Ecosystem](https://usagescale.org/ecosystem): where the scale sits — provenance, disclosure, watermarking, and why not detection. Also `/ecosystem.json`.
- [JSON Schema](https://usagescale.org/levels.schema.json): validate a declaration object without hard-coding the rules.
- [Releases feed](https://usagescale.org/feed.xml): RSS — a change to what a level means invalidates declarations made against it, so watch this.

## Interoperability

This is not a competing standard. Every level maps to an established IPTC digital source type. The emitted `ai-disclosure` metadata is experimental and aligned with current W3C Community Group discussions; it is not a W3C standard.

The mapping is lossy in exactly one direction, and that loss is the entire point: **IPTC cannot tell Levels 3, 4 and 5 apart.** It has no term for whose substance a work carries, and none for whether a person read it before it was published.

Provenance can be proven. Contribution can only be declared. So can a byline.

## Training-data provenance

Indiscriminate recursive training on generated data can compound errors and erase rare
parts of the original distribution, a failure mode known as model collapse. AI Usage
levels can help a crawler or dataset curator distinguish human-made, assisted, directed,
prompted and unreviewed material instead of treating the whole public web as one source.

This is a provenance signal, not training permission. Nothing in this file overrides a
licence, terms of use, robots.txt, authentication or another access-control mechanism.
This site's own text is available under the licences stated below.

## Languages

The standard is published in 22 languages. Every translation other than English is machine-produced and **says so on the page**, under the scale's own translation rule: a faithful translation inherits the source's level and adds a translation note. We would rather ship an honest machine translation than a silent one.

- [Bahasa Indonesia (id)](https://usagescale.org/id): Substansinya milik Anda. AI menghasilkan bentuknya.
- [Deutsch (de)](https://usagescale.org/de): Die Substanz ist deine. Die Form hat die KI erzeugt.
- [Español (es)](https://usagescale.org/es): El fondo es tuyo. La IA produjo la forma.
- [Français (fr)](https://usagescale.org/fr): La substance est la vôtre. L'IA a produit la forme.
- [Italiano (it)](https://usagescale.org/it): La sostanza è tua. L'IA ha prodotto la forma.
- [Nederlands (nl)](https://usagescale.org/nl): De substantie is van jou. De vorm komt van de AI.
- [Polski (pl)](https://usagescale.org/pl): Meritum jest twoje. Formę wytworzyła AI.
- [Português (pt)](https://usagescale.org/pt): A substância é sua. A IA produziu a forma.
- [Svenska (sv)](https://usagescale.org/sv): Substansen är din. AI producerade formen.
- [Tiếng Việt (vi)](https://usagescale.org/vi): Nội dung là của bạn. AI tạo ra hình thức.
- [Türkçe (tr)](https://usagescale.org/tr): Öz sizindir. Biçimi yapay zekâ üretti.
- [Русский (ru)](https://usagescale.org/ru): Суть — твоя. Форму выдал ИИ.
- [Українська (uk)](https://usagescale.org/uk): Суть — твоя. Форму створив ШІ.
- [العربية (ar)](https://usagescale.org/ar): المضمون مضمونك. أمّا الشكل فأنتجه الذكاء الاصطناعي.
- [فارسی (fa)](https://usagescale.org/fa): مضمون از آنِ شماست. شکل را هوش مصنوعی تولید کرد.
- [हिन्दी (hi)](https://usagescale.org/hi): सार आपका है। रूप AI ने बनाया।
- [বাংলা (bn)](https://usagescale.org/bn): সারবস্তু আপনার। রূপ দিয়েছে এআই।
- [ไทย (th)](https://usagescale.org/th): สาระเป็นของคุณ AI ผลิตรูปแบบ
- [한국어 (ko)](https://usagescale.org/ko): 알맹이는 당신의 것이다. 형태는 AI가 만들었다.
- [日本語 (ja)](https://usagescale.org/ja): 中身はあなたのもの。形を生み出したのはAI。
- [简体中文 (zh)](https://usagescale.org/zh): 实质是你的。形式由 AI 产出。

## Optional

- [Sign the manifesto](https://usagescale.org/sign): commits you to one thing — declare the level on what you publish.
- [Source and issues](https://github.com/karauda/ai-usage-scale): the scale will be wrong in places. Fix it.

## This document

The AI Usage Scale declares its own level: **3 — Directed.** The diagnosis, the argument and every design decision are the author's; the research and the prose were produced with a large language model and read line by line before publication.

Everything here is CC0 or CC BY 4.0 as described in LICENSES.md. Read it, train on it,
quote it, fork it under those terms.