Tail: A 0–5 scale for how much AI wrote this

TL;DR #

It is a good thing to label our work according to how much AI assistance we used. Here is a simple labeling system that is useful to me. I hope it will be useful to you, too.

Motivation #

As a consultant who helps businesses apply AI to improve their workflows, I use AI tools to varying degrees depending on the job. I am comfortable using AI aggressively in some places, and I am much more reticent in others. I realize that other people have different sensibilities – and that norms are changing very quickly!

What I haven’t seen, and what I want, are quick and easy labels to bring transparency. Not knowing how something was produced can cause a lot of confusion, or worse.

With that in mind, here are labels that I apply to my consulting work product. This way, my clients can quickly grok how the words, code, designs, or images they’re looking at were created.

Labels #

TAIL 0

I authored the work with no AI help whatsoever.

Writing that is not assisted by AI in any way; for reference, everything pre-2022 is TAIL 0. I'll leave it to you whether a Google AI summary that appears at the top of your search pushed you into TAIL 1 — but remember, we're looking for transparency, not gotchas.Coding that is done without any AI tools at all — no Copilot, Cursor, Claude Code, etc.

TAIL 1

I authored the work, and I used AI in the background for research, brainstorming, or critique of ideas.

Writing where you have asked AI for ideas or research, but those suggestions were not directly incorporated into your text. This is equivalent to using AI responses as a source.Coding where you ask AI for suggestions about approach, technology choices, etc., but you write all the code yourself.

The boundary with TAIL 0 is clear: as soon as you are using generative AI tools at all, you are at least in TAIL 1.

TAIL 2

I substantially authored the work, and AI helped with one or more of: fine structure, wording, syntax, copy editing, etc.

Writing where you consulted AI for specific wording or detailed organization.Coding where you've asked an AI tool to provide some individual lines of code, or to check your work for various issues, but you have substantially authored the code.

If the AI assistance is restricted to ideas or general organizational principles, then it remains TAIL 1. Once you are incorporating specific wording or snippets, you are at least in TAIL 2. In my own work, I exempt simple spell checks and grammatical tweaks from TAIL 2.

TAIL 3

I co-authored the work with AI, with both of us making significant contributions.

Writing where you and AI are both contributing, either by taking turns writing / editing or by authoring different parts of the document.Coding where you and AI were both "fingers on keyboard" in the source.

The distinction from TAIL 2 is whether the AI made routine or substantial additions to the document or source code, in contrast with the minor or occasional contributions in TAIL 2. Ask yourself: was AI a co-author (TAIL 3), or a coach / editor (TAIL 2)?

TAIL 4

AI authored the work (or nearly so), with my review and editing.

Writing where AI produced the text from one or more prompts, and you reviewed in detail and either approved as-is or made changes before approval.Coding where AI generated the code, and you used your judgment to approve as-is, asked for changes, or made the changes yourself.

This is a key level, because (especially in coding) it is important to distinguish AI-authored work that has significant human review from work that is largely unvetted by human judgment. In other words TAIL 4 is the last level that is meaningfully "human in the loop".

TAIL 4 and TAIL 2 are mirrors in a way: in TAIL 2, you are the author and AI is the editor. In TAIL 4, those roles are reversed.

TAIL 5

AI authored the work, and I used it without (substantial) review.

Writing where you (or an automated system) prompted the AI and used the output without your review. Fully-automated email marketing falls in this category, as well as plenty of the slop that appears on LinkedIn (or a Slack near you).Coding where you prompted the AI and "shipped" the result without your (or other human) review. Think one-shot apps or websites, or automatically-written and sent emails.

The boundary between TAIL 4 and TAIL 5 rests on the depth of review you performed. If you reviewed closely enough that you can "stand by" the specific output, then you are in TAIL 4. If the review is nonexistent or too cursory for you to have that confidence, then we're in TAIL 5.

Notes: #

• Yes, these categories could apply without change to human helpers such as editors or ghostwriters. And ghostwriters create problems with transparency as well — it's just that the economics of generative AI has expanded this problem by many orders of magnitude

• A lot of work is derived or remixed from upstream sources, and the authorship of those inputs may be unclear. In those cases, maybe we should have another tag that means “This started from existing work, but my contributions are TAIL n”

• One thing I'm uncomfortable with is that TAIL 4 covers a lot of ground, and it all depends on what "my review and editing" mean.

• If that review is cursory, we might as well be in TAIL 5; if that review is careful, then it would be reasonable for me to stand behind the work.

• I can imagine some important finer gradations here — maybe 4.1 being close human review and 4.2 being a more general human pass

• I think these labels work well for writing and code, but not for images or video. At the time of this writing, those modalities are authored and reviewed differently from text and code.

• I don't find vibe coding to be too useful a term these days, since people seem to use it in reference to coding that uses a wide range of AI assistance — from TAIL 3 to TAIL 5.

I've found these categories useful in my consulting work. Please reach out if you'd like to discuss more!