Say how much AI was used

A label you can add to your own work, from writing and images to video and code, saying how much AI went into it. Seven tiers from fully human to fully automated, shown as one badge where people start reading.

Voluntary and self-declared. Nobody checks or certifies it, and it isn't a legal standard. The badge is your own word, like a byline. Free to use under the HAI Scale License, no sign-up.

HAI Scalev1.0
TierMost human first
HFully human HCHuman, checked M1Mixed, human-leaning M2Mixed, even split A1Mostly automated A2Almost fully automated AFFully automated

Self-declared by the maker. Not a certification or legal standard.

Get the right badge for where it's going

Every file

Find your tier, say where the badge will live, and take the snippet or file that fits.

Step 1 · Your tier
Already know it? Pick it directly
Step 2 · Where it's going
Background
Step 3 · Your badge

Choose a tier above to see your badge.

Let your agent add the badge

agents.md

Point a coding or writing agent at one file and it can work out the tier and insert the badge itself.

Using AI to choose or add the badge doesn't count as AI use. The tier describes how the content was made, not how the label got there. A hand-written essay that an agent labels for you is still H.

Paste this into your agent

Prompt
Read https://haiscale.amolven.com/agents.md and add the right HAI Scale badge to this project. Tell me which tier you chose and why.

Or link the instructions

One Markdown file with the tiers, the decision steps and the exact snippets for every placement.

Link
https://haiscale.amolven.com/agents.md

Open agents.md · also listed in llms.txt

Four rules

  1. 01

    Put it where people start

    Byline, title page, first seconds of a video, top of a README. A badge buried at the end discloses nothing.

  2. 02

    Label the finished work

    Judge what was published, not every draft along the way. If parts differ a lot, label the parts separately.

  3. 03

    When unsure, go one tier up

    If you're torn between two tiers, choose the more automated one. Over-disclosing costs nothing; under-disclosing costs trust.

  4. 04

    Don't restyle the badge

    Keep the colours and codes as they are so people learn to recognise them. Use the chip by default and the circle only where a chip can't fit.

What it looks like in place

Full examples
Blog post

Notes from a year of sourdough

The chip sits in the byline, next to who wrote it and when.

Photo
A2

No room for a label, so the circle goes in a corner.

README

weekend-planner

buildpassing licenseMIT HAI Scale: A1 — Mostly automated

In the badge row, alongside build status and licence.

Full files: blog post · README · plain-text forms

Every file

Download the kit

The chip is the default badge. The circle is for corners and icons; the dot is the circle without text, for anything under 24 pixels.

TierChipCircle
HH·Fully humanSVGSVG darkPNGPDFSVGPNGDot
HCHC·Human, checkedSVGSVG darkPNGPDFSVGPNGDot
M1M1·Mixed, human-leaningSVGSVG darkPNGPDFSVGPNGDot
M2M2·Mixed, even splitSVGSVG darkPNGPDFSVGPNGDot
A1A1·Mostly automatedSVGSVG darkPNGPDFSVGPNGDot
A2A2·Almost fully automatedSVGSVG darkPNGPDFSVGPNGDot
AFAF·Fully automatedSVGSVG darkPNGPDFSVGPNGDot

For machines

Add one tag to a page's <head> so tools can read the tier without parsing the badge:

<meta name="hai-scale" content="A1">

The full scale definition, with colours and URLs, is published as hai.json.

Stable URLs

Each tier has a permanent page and badge address you can link or hotlink:

https://haiscale.amolven.com/A1/
https://haiscale.amolven.com/badge/A1.svg
https://haiscale.amolven.com/badge/circle/A1.svg

Questions

Is this a legal or official standard?

No. It's a voluntary convention that one person started and anyone can use. It isn't certified, audited or recognised by any regulator, and it doesn't replace any disclosure a law, publisher or platform requires of you.

Does anyone verify the badge?

No. It is a self-declaration, like a byline. The badge says what the maker reports, and their reputation stands behind it.

What counts as AI?

Generative and machine-learning systems that produce or judge content: chat assistants, image and audio generators, coding agents, AI grammar and review tools. Deterministic tools such as spellcheck, calculators, filters and compilers do not count.

I used AI only to brainstorm. Which tier?

If none of its output ended up in the work, HC. If its ideas or wording shaped what you published, M1 or higher.

AI fixed my grammar and reworded a few sentences. Is that still HC?

Yes. HC covers checking and minor touch-ups that don't change the content: typos, grammar, simple rewording. Once AI writes new passages or changes what the work says, it is M1.

My project has parts with very different AI involvement.

Label the whole with the tier that best describes it overall, or label the parts separately: for example, text H and illustrations A2.

An AI picked my tier and added the badge. Does that change the tier?

No. The badge covers the content only. Using an AI to choose or insert the label is not counted.

Is a higher tier worse?

No. The scale describes how something was made, not how good it is. The point is that people can tell.

Can I use it commercially? Do I need permission?

Yes, and no permission needed. The HAI Scale License lets anyone use the scale for free, including on commercial work, with three conditions: label truthfully, keep the tiers and badges intact, and don't sell access to the scale itself. Charging for your own product that shows a badge is fine.

Will the tiers change?

This is version 1.0. The seven codes are meant to stay stable; wording may be refined. Changes will be versioned here.