Chapter 13Before you share
Writing the note
Look again at the Vanderbilt line. It was set out like a formal citation and gave a date, but it answered only one question: where the words came from. Readers wanted to know what the people sending it had done. A useful note says three things: what the AI tool did, what you did, and that you answer for the result. APA’s own example is a single sentence: “I used ChatGPT (OpenAI, 2025) to edit my paper and provide feedback on clarity and flow.”1
For a report at work, a note might read: AI use: I used an AI tool to draft this summary. I checked the figures against the source report and rewrote the conclusions, and I’m responsible for the result. (An example written for this page.) It names the tool’s part, names yours, and claims the work. If your workplace or publisher has its own wording, use theirs.
Where should you say it? Wherever your reader would want to know. That is a judgment, not a formula. A spelling check needs no note. A summary a client will rely on, a paper for a class, or anything a reader might trust differently if they knew, probably does. A fair test, in the spirit of the tax-advisor experiment: would this reader feel misled if they learned it from someone else?
A last look before you share
Before AI-assisted work goes to someone else, run down a short list. It is the book’s advice, gathered from this chapter and the ones around it:
- The facts, numbers, names and quotes are checked (Chapter 12).
- Every source it cites has been opened and read (Chapter 12).
- Nothing private is in it that shouldn’t travel (Chapter 10).
- It is cut to what this reader needs, with the main point first.
- It says what it is, what was checked and what it doesn’t cover.
- It says how AI helped, where the reader would want to know.
- It says what you mean, and you would sign it (Chapter 14).
None of this takes long for an everyday email. For a report a client will act on, it is most of the work, and it is the part no AI tool can do for you.
Try it: tick off the list, then build a short note on how AI helped.
Your turn
Try it. Ask any AI chat tool you already use for a short email about a real task, such as telling a team about a schedule change, or take something you recently drafted with AI help. Count the words. Mark the line that answers each of the three questions, and strike every line that answers none of them. Count again. Then add one line saying what it is and what you checked, and one saying how AI helped. On paper, do the same with a long email from your own inbox. Did the cut version lose anything a reader needed, or turn up a gap?
Think about it. Think of the people who read your work: a manager, clients, students, a patient’s family. Which of them would want to know that AI helped, and what would they want to know about it? How would they feel if they heard it from someone else?
So what do you owe the person who reads an AI-assisted draft? The work that makes it theirs: cut to what they need, the main point first, checked, with a plain word about what it is and how AI helped. That leaves a harder question, the one the Vanderbilt apology raised. Some work shouldn’t be handed to an AI tool at all, however well you check and label it. Which work is that? That is the next chapter.
Check in
Try answering in your own words, then open each one to compare.read the answer to compare.
Why does putting the main point first and cutting the rest help a reader?
Busy readers scan. In Morkes and Nielsen’s 1997 studies, most participants scanned unfamiliar text before reading it, and a version of a website cut to about half the words scored 58 percent higher on measured usability. What comes first is what a reader is most likely to see.
Why do publishing-ethics groups say an AI tool can’t be listed as an author?
Because it can’t take responsibility for the work, as COPE puts it. Authors disclose how they used the tool instead, and they remain responsible for everything in the work, including the parts the tool produced.
An HR manager used an AI tool to draft a one-page summary of a new leave policy for staff. She checked it against the policy itself, but not against state law. What should the top of her email say?
What it is (a summary, not the policy itself), what was checked (the policy text), and what it doesn’t cover (state law). Then the main change first, and a line saying an AI tool drafted the summary and she checked it.
Your AI-assisted summary for a client quotes a sentence and gives a percentage from an industry report. Before you send it, how do you check each one? (This goes back to Chapter 12.)
Match the check to the claim. Open the report and find the sentence word for word, and find the percentage there too, redoing any math. If you can’t find either one, cut it.
A colleague says, “Research shows people trust you less if you admit AI helped, so I never mention it.” Is that a sound conclusion?
Only half. In Schilke and Reimann’s experiments, disclosing did lower trust, even with careful wording. But being exposed by someone else lowered it further, and many schools, journals and employers require disclosure anyway. The better answer is to make the work worth trusting and say plainly what you did.
Notes (1)
- 1Timothy McAdoo, Samantha Denneny, and Chelsea Lee, “Citing generative AI in APA Style: Part 1—Reference formats,” APA Style Blog, September 9, 2025, https://apastyle.apa.org/blog/cite-generative-ai-references (accessed October 27, 2026).
Chapter 13 · page 4 of 4
Turn pages with the ← and → keys.