At a glance / A suggestion is not yet an editorial decision
| Task | Output | Human decision | Final location |
|---|---|---|---|
| Find repeated questions | A comparison list | Retain order; rewrite two transitions | Chapter 4 revision record |
The trace is for revision, not theatre
An AI decision trace is a private working record that lets an author or editor explain a change without pretending the tool made an editorial judgement. It can be short: task, material shared, output class, human decision, and final location. The point is not to preserve every keystroke. It is to distinguish a tool’s suggestion from the person who selected, rewrote, rejected, or independently verified it.
Set the boundary before upload
Write what may leave the manuscript environment. A draft with confidential interviews, client material, or an unpublished collaborator’s work may need a no-upload rule. The Authors Guild cautions that material entered into public chatbots may be used for training and may expose confidential material; tool settings and agreements change. Record the tool and account setting actually used, but do not infer a privacy guarantee from a label.
Log decisions at the right grain
For a chapter-level structure request, one row might read: ‘Asked for repeated questions; output used as a comparison list; author retained original order, rewrote two transitions; final changes in chapter 4.’ For line suggestions, group a coherent pass rather than creating a ledger of every comma. If generated prose supplies factual language, stop treating it as editing: verify the underlying source before any wording enters the manuscript.
Use the trace to slow acceptance
A useful review question is: could the writer explain why this sentence is here without saying ‘the tool suggested it’? If not, return to the source, intention, or scene. The U.S. Copyright Office’s current report stresses human determination of expressive elements rather than prompts alone. That is not a template for literary value, but it supports a sensible practice: retain evidence of human selection and revision where authorship claims matter.
Choose disclosure separately
A private trace does not itself decide a public disclosure, certification, submission answer, or contract representation. Read the specific requirement and ask the publisher when the agreement requires it. The Authors Guild distinguishes background uses from copying generated text and says disclosure can be placed in front matter or an article note. Preserve the trace long enough for the project’s review and record-retention policy; delete sensitive exports only through an agreed process.
The trace should also say when no output was adopted. That negative record is useful when a writer used a tool to explore a structure and then returned to an independent draft. It avoids a false binary between untouched writing and wholesale generation. Keep the language factual: describe the task and decision, not an assurance that a detector, label, or private log can prove authorship.
Sources & limits
This guide is not legal advice and cannot determine copyrightability, contractual disclosure, or a platform’s data practices.
- AI Best Practices for Authors
Authors Guild · Source undated · Consulted 19 Sept 2026AI uses raise different concerns; public-chatbot inputs may expose confidential material; disclosure may be placed in book front matter or article notes.
- Copyright Office Releases Part 2 of Artificial Intelligence Report
U.S. Copyright Office · Source dated 29 Jan 2025 · Consulted 19 Sept 2026Copyrightability depends on sufficient human determination of expressive elements, not prompts alone.
Send a correction with the passage and supporting source.

