Introduction
People who want to be honest about using AI often get stuck on the wording. They aren't sure whether a disclosure is needed, how much to say, or whether saying anything will make the work look worse.
This walkthrough takes three people through the same five steps and ends with the statement each one writes. The people, their work, and the rules they work under are invented. The rules are typical of ones in use and aren't quoted from any real institution.
The Starting Point
Marcus is an independent consultant. He has written a twenty-page report for a client on why its customer complaints rose last year. The analysis and recommendations are his. He used an AI assistant to tighten the wording throughout and to cut the executive summary from two pages to one.
Ms. Okafor teaches middle school science. She gave an AI tool her unit objectives on photosynthesis and asked for a ten-question quiz. She then went through the draft, replaced two questions, fixed one wrong entry in the answer key, and simplified the wording of several items.
Wei is a master's student writing a literature summary for a seminar. He had read six of his twelve papers closely. For the other six, he asked an AI assistant for summaries to get started, then read each paper and rewrote the summaries himself. He also asked the assistant to suggest ways of grouping the twelve papers into themes, and used one of its suggestions.
Walking Through the Decision
Step 1: Find the rule that applies, or note that none is written
Each person looks for the written rule first.
Marcus rereads his contract. It doesn't mention AI. It has a confidentiality clause, which raises a separate question about whether client material may go into an AI tool at all. He settled that with the client before he started. On disclosure, the contract is silent.
Ms. Okafor's district has a staff policy. It says staff may use district-approved AI tools to prepare teaching materials, that staff remain responsible for the accuracy of what they give students, and that no student information may be entered. It doesn't require disclosure.
Wei's seminar syllabus has a statement on AI. Students may use AI tools to find and summarize sources. All submitted text must be the student's own writing. Any AI use must be acknowledged in a note at the end of the paper, naming the tool and what it was used for.
So one person has no rule, one has a rule that permits the use and asks for nothing more, and one has a rule that requires a note.
Step 2: Say what the AI did and what the person did
Before deciding anything, each person writes down the division of work in plain terms.
| What the AI did | What the person did |
|---|
| Marcus | Edited wording; shortened the summary | The analysis, the recommendations, the full draft, review of every edit |
| Ms. Okafor | Drafted ten questions and an answer key | Set the objectives, replaced two questions, corrected the key, adjusted the wording |
| Wei | Drafted summaries of six papers; suggested theme groupings | Read all twelve papers, wrote all the final text, chose the grouping |
A person who can't fill in the right-hand column has found a problem with the work itself.
Step 3: Apply the test of whether the reader would feel misled
The test is whether the person relying on the work would feel misled if they learned later how it was made.
Marcus's client is paying for his analysis, and the analysis is his. Many clients wouldn't mind that software polished the prose. Some would want to know, and Marcus can't tell which kind this client is. A one-sentence statement costs him little, so he decides to include one and to ask the client how they'd like this handled in future.
Ms. Okafor's students rely on the quiz being accurate and matched to what she taught. How the first draft was produced doesn't change what they're relying on, and her policy asks for no disclosure. Her colleagues are a different audience. The quiz goes into a shared department folder, and a teacher who reuses it would want to know which parts she checked. She decides on a short note in the file.
Wei doesn't need the test. His syllabus requires a note.
Step 4: Draft a disclosure of one to two sentences
Each statement names the tool, the use, and the human review. None of them apologizes or argues.
Wei checks his against the other half of his rule. The syllabus says all submitted text must be his own writing. Because he rewrote every summary after reading the papers, he can say so truthfully. If he had kept the AI's summaries, the honest disclosure would have described a violation, and the fix would have been to rewrite the summaries.
Step 5: State who answers for errors
In all three cases the answer is the person.
If Marcus's shortened summary now overstates a finding, the client will hold Marcus to it. If a quiz question has two defensible answers, Ms. Okafor's policy makes her responsible. If Wei's paper misdescribes a study, the mark and any integrity question are his.
Publishing bodies have put this in writing for authors of research papers. The Committee on Publication Ethics says authors are "fully responsible for the content of their manuscript, even those parts produced by an AI tool" (Committee on Publication Ethics 2023). The International Committee of Medical Journal Editors says humans "are responsible for any submitted material that included the use of AI-assisted technologies" (ICMJE 2026). Neither rule binds a consultant, a teacher, or a seminar student. The reasoning carries over: the note tells readers about the method and leaves the responsibility where it was.
Key Considerations
The common mistake is to treat a disclosure as a confession. People who see it that way either avoid disclosing or write something defensive, such as "I only used AI a little." A disclosure is a description of method, in the same family as naming the software used to analyze data or thanking a colleague who read a draft. It says what was done so that readers can judge the work with that in view.
A second mistake runs the other way: treating the disclosure as a shield. "AI-assisted" at the bottom of a document doesn't lower the standard the document has to meet.
How much to say depends on the rule and the reader. Where no rule exists, one or two sentences that a reader could check against the work are enough. A general description, such as "an AI writing assistant," usually suffices unless the rule asks for the product's name.
Summary
Each person found the rule or its absence, separated the AI's work from their own, asked what the reader would expect to know, and wrote a short description of method. The three statements follow.
Marcus, in the client report: "I wrote this report and its analysis. I used an AI writing assistant to edit the wording for clarity and length, and I reviewed every change."
Ms. Okafor, in the shared file: "First draft of this quiz was generated with the district-approved AI tool from my unit objectives. I replaced questions 4 and 9, corrected the answer key for question 6, and checked every item against the unit."
Wei, at the end of the seminar paper: "AI use: I used an AI assistant to produce first summaries of six of the twelve papers and to suggest ways of grouping them. I read all twelve papers, and all text in this paper is my own."
- Tool. Each statement says what kind of tool was used: a writing assistant, the district-approved tool, an AI assistant.
- Use. Each says what the tool did: edited wording, generated a first draft, produced first summaries and suggested groupings.
- Human review. Each says what the person did afterward: reviewed every change, replaced and corrected specific items, read every paper and wrote the final text.
References
- Committee on Publication Ethics. 2023. "Authorship and AI Tools." COPE position statement. First issued February 2023.
- International Committee of Medical Journal Editors. 2026. Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals. Updated January 2026.