Every major body that has actually had to rule on AI and authorship has converged on the same two-part answer: an AI tool cannot be an author, and using one to produce appreciable content requires disclosure. That convergence is useful, because it means you do not have to invent your own personal ethics of AI-assisted work from scratch — you can borrow a standard that has already been tested against real disputes.
What the actual authorities say
The U.S. Copyright Office’s 2023 policy statement requires applicants to disclose the inclusion of AI-generated content in a work submitted for registration and to give a brief explanation of the human author’s contributions; separately, AI-generated content that is more than de minimis should be explicitly excluded from the claim. Applicants “should not list an AI technology or the company that provided it as an author or co-author simply because they used it when creating their work” — authorship claims must be limited to the human-authored portions (U.S. Copyright Office, “Copyright Registration Guidance,” 2023). The Committee on Publication Ethics reached the same structural conclusion for academic publishing in its 2023 position statement: AI tools “cannot meet the requirements for authorship as they cannot take responsibility for the submitted work,” cannot hold copyright or manage conflicts of interest, and any appreciable AI use in a manuscript’s writing, data analysis, or images must be disclosed, typically in the methods or acknowledgments section, naming the specific tool (COPE, “Authorship and AI Tools,” 2023). A 2026 cross-publisher review of Emerald, Frontiers, Sage, and Taylor & Francis author guidelines found the same “human accountability” standard across those four publishers, regardless of how permissive each publisher’s tone was about AI use itself (Chigwada & Ngulube, Frontiers in Research Metrics and Analytics, 2026). Other major houses publish similar disclosure rules in their own author guidelines — check the journal or publisher you are submitting to rather than assuming one house’s policy covers another.
Notice what these bodies are not saying. None of them say AI-assisted work is dishonest, inferior, or something to hide. They say the same thing in slightly different words: a human stays the accountable author, and disclosure applies once AI use is more than trivial (spell-check and routine grammar tools are broadly treated as exempt) — a rule built for scholarly publishing and legal registration, but one that generalizes cleanly to everyday professional and creative work.
The everyday version of the same rule
Translated out of legal and academic language, the practical norm for most everyday AI-assisted work is:
- You remain the accountable author. If the work is wrong, offensive, or causes a problem, that is your responsibility to fix and answer for, regardless of which tool produced the specific sentence.
- Disclose appreciable use where the distinction matters. A quick grammar pass does not need a footnote. A client deliverable substantially drafted by AI, a school assignment, or a professional document where the reader is relying on your unaided expertise generally does.
- Never claim credentialed expertise you did not exercise. If a document implies specialized judgment (legal, medical, financial, technical) was applied by you personally, and AI did the substantive reasoning instead of you verifying it, that is the specific case disclosure exists to prevent.
Do not present AI-generated analysis, code, or advice in a specialized domain as your own verified expert judgment unless you actually reviewed and stand behind every substantive claim. The disclosure norm is not satisfied by a generic “AI was used” footnote if the underlying content was never actually checked by the named human author.
What this is not: a plagiarism panic
It is easy for authorship guidance to tip into treating any AI assistance as something shameful to confess. That overcorrection causes its own damage — it pushes disclosure underground, because people reasonably fear being penalized for normal, increasingly universal tool use rather than rewarded for honesty about it. The actual authorities above do not ask for that. COPE’s position asks only that AI use be disclosed in the methods or a similar section of the paper, which is better read as routine reporting than as confession, and multiple publishers now provide a standard field for exactly this purpose rather than treating it as an exceptional admission. Treat disclosure the same way: a normal, unremarkable line in a document’s process notes, not an apology.
A disclosure decision table
| Context | Disclose? | How |
|---|---|---|
| Internal draft you will fully rewrite yourself | No | — |
| Client deliverable substantially AI-drafted, then reviewed by you | Yes | A brief process note: “Drafted with AI assistance, reviewed and finalized by [name].” |
| School assignment | Yes, per your institution’s specific policy | Follow the exact disclosure format your school requires; policies vary and the format itself matters |
| Published or publicly attributed writing | Yes, if AI contributed appreciable content | Name the tool and describe the human contribution, following the COPE/publisher pattern above |
| Code generated with AI assistance, reviewed and tested by you | Often expected in professional settings | A comment or commit note identifying AI-assisted sections, per your team’s convention |
| Specialized professional advice (legal, medical, financial) | Always, and only after independent verification | State clearly what was AI-assisted and what you personally verified |
The full AI-assisted authorship disclosure template gives you ready-to-use disclosure language for each of these contexts.
When there is no formal policy to follow
Most everyday work does not come with a publisher’s author guidelines or a school’s explicit AI policy attached. In that gap, default to the same underlying test the authorities above use, scaled down: would the person relying on this work want to know that AI contributed appreciably to it, and would they feel misled if they found out later without having been told? If the honest answer is yes, disclose, even without a formal template to follow. A short, plain sentence is enough in most everyday contexts — the goal is honesty, not bureaucratic process.
Teams without a written AI-use policy tend to develop an inconsistent, ad hoc norm by default, where some people disclose and others don’t, and nobody has actually agreed on what counts as “appreciable.” That inconsistency is itself a source of the friction that authorship disputes cause later — a colleague who finds out a heavily AI-drafted deliverable was presented without any disclosure will reasonably feel misled, even if no rule was technically broken. If you manage a team, what to tell your team when you automate covers the parallel problem of disclosure at the process level; pair it with a simple, team-agreed threshold for when individual deliverables need a disclosure line, so the norm does not depend on each person’s individual judgment call every time.
Where this connects to skill and craft
Disclosure norms exist alongside, not instead of, the judgment about how much of a piece of work should be AI-generated in the first place. Keep your taste: AI and creative practice covers the craft side of that question — how much AI assistance still lets you develop and exercise your own creative judgment — while this article covers the accountability side: whatever mix you land on, name it honestly. The two questions are related but distinct: a heavily AI-assisted piece that is honestly disclosed is a different situation, ethically, than a lightly AI-assisted piece presented as fully unaided. What not to delegate to AI covers the upstream decision of how much to delegate in the first place; this article covers what to do once you have delegated something appreciable.
Try it today
Look at the last piece of AI-assisted work you produced that went to someone else — a client, a manager, a teacher, a reader. Using the decision table above, decide honestly whether it needed disclosure and whether it got it. If it did not and should have, that is worth a short, unremarkable follow-up note now, using the language in the downloadable template, rather than a bigger problem later if the gap surfaces on its own.
