trace-label

TRACE AI LABEL v0.1

Transparent Reporting of AI Content & Editing


Why TRACE?

AI is everywhere in publishing. But right now, it’s hard to tell what’s been lightly edited by a model, what’s been co-authored, and what’s been fully generated. This leaves audiences guessing, creators exposed to skepticism, and platforms vulnerable to regulatory pressure.

TRACE aims to fix this with a simple, neutral labeling system. Think of it like a nutrition label for AI use: lightweight, consistent, and easy to adopt.


Core Principles

  1. Simplicity first — a single badge like [TRACE U3] tells the whole story at a glance.
  2. Neutral, not judgmental — labels explain usage, not quality.
  3. Self-label today, platform support tomorrow — the spec starts with creators, then grows into integrations.
  4. Tiered clarity — six clear levels (U0–U5) cover the full spectrum of AI involvement.

The Levels (U0–U5)


Using TRACE Labels

The simplest way to start is with self-labeling:

View the Styleguide for standardization of visual implementation

Optional suffixes add more transparency:

Example label with multiple suffixes:
[TRACE U4 +PR +SR]


JSON Metadata (for platforms & devs)

Every badge can be backed by structured metadata.
Lightweight, optional at first — but ready for automation.

{
  "trace": {
    "version": "0.1",
    "level": "U3",
    "modality": "text",
    "human": {
      "review": true,
      "authors": ["Jason S."]
    },
    "date": "2025-09-10",
    "ai": {
      "role": "coauthor",
      "method": "iterative",
      "models": ["disclosed"]
    },
    "safety": ["copyright"]
  }
}

Field-by-field guidance

Tip: v0.1 is intentionally lightweight. Start with the fields above. Platforms can add richer, optional detail using the extension block below.


Optional extension block (for richer detail)

To support power users and platform integrations without burdening creators, add an optional ext namespace:

{
  "trace": {
    "version": "0.1",
    "level": "U3",
    "modality": "text",
    "human": { "review": true },
    "date": "2025-09-10",
    "ai": { "role": "coauthor", "method": "iterative", "models": ["disclosed"] },
    "safety": ["copyright"],

    "ext": {
      "apps": [
        { "name": "ChatGPT", "provider": "OpenAI", "version": "2025-09", "mode": "web" }
      ],
      "models": [
        {
          "name": "gpt-5-thinking",
          "provider": "OpenAI",
          "version": "2025-08-15",
          "id": "openai:gpt-5-thinking:2025-08-15",
          "usage": { "share": "majority", "purpose": "drafting" }
        }
      ],
      "pipeline": [
        { "step": "transcription", "app": "Whisper", "version": "large-v3" },
        { "step": "summarization", "model": "gpt-5-thinking", "iterations": 3 }
      ],
      "data": { "sources": ["original", "licensed"], "retrieval": { "used": false } },
      "prompt": {
        "disclosure": "summary",
        "summary": "Draft first-pass newsletter from transcript; keep jokes; 700–900 words."
      },
      "provenance": { "type": "none" }
    }
  }
}

Schema choices that help adoption


For Creators


For Platforms & Engineers


Adoption Path


Contribute

TRACE is open and versioned. v0.1 is deliberately simple — the spec will evolve with community input.

Ways to contribute:

Open issues, propose updates, or join the discussion in GitHub Issues.


Transparency only works if we do it together. Start by labeling your next post [TRACE Ux].