An AI disclosure statement is a short note that says which AI tool you used, what you used it for, and that a person checked the result. One to three sentences is enough. Journals and most universities now require one; most workplaces do not yet, but a one-line statement costs nothing and saves an awkward conversation later.
On this page
- What an AI disclosure statement is
- When you need one (by context)
- The four levels of AI use
- What to include
- 8 AI disclosure statement examples
- Template you can copy
- Where to put it
- Mistakes that make it useless
- FAQ
What is an AI disclosure statement?
It is a declaration, written by the person responsible for a piece of work, that generative AI took part in making it. It answers three questions a reader would ask if they could: which tool, for what, and who stands behind the result. It is not a disclaimer ("this may contain errors") and it is not a citation of a specific output, although you may need those too.
The clearest formal version comes from academic publishing. The ICMJE recommendations say chatbots cannot be authors, that journals should ask at submission whether AI was used, and that humans remain responsible for any material that involved AI. Elsevier's policy goes further and gives authors the statement title to use and a wording template. The same logic transfers to a client report, a blog post or an internal memo: name the tool, say what it did, own the output.
When do you need an AI disclosure statement?
It depends on who is reading and what rule they work under. The table below is the short version. Where a rule exists, it is linked, and you should read the rule itself before relying on it.
| Context | Is a statement required? | Who says so |
|---|---|---|
| Journal article or conference paper | Yes, for anything beyond grammar and spelling. In the cover letter and in the manuscript. | ICMJE, Elsevier and most publishers |
| Grant peer review (NIH) | Not a disclosure question: use is prohibited, because uploading an application breaks confidentiality. | NIH NOT-OD-23-149 |
| Student coursework | Usually yes. Check the course or university policy; the instructor decides. | Your institution's policy |
| Chatbot or voice agent talking to the public in the EU | Yes. People must be told they are dealing with a machine. Deepfakes and AI text published on matters of public interest must be labeled. | EU AI Act transparency rules, applying from August 2026 |
| Reviews and testimonials | Disclosure does not save you. Fake reviews from people who do not exist, including AI-generated ones, are banned in the US. | FTC rule on fake reviews |
| Blog posts, marketing copy, web content | Not required by Google. Google says it rewards helpful content however it is produced, and suggests a disclosure where a reader might reasonably ask how it was made. | Google Search guidance |
| Client deliverables (consulting, agency, legal, accounting) | Check the contract and engagement letter. Many now ask for it; some forbid uploading client data to public tools. | Your contract and your firm's policy |
| Internal documents | Whatever your company AI policy says. If you have no policy, disclose anything beyond grammar and spelling until you do. | Your employer |
Two things are missing from most guides. First, none of the publishing rules say how much AI use triggers a statement. Second, none of them give wording for work outside a journal. The next two sections fix both.
The four levels of AI use, and what each one needs
Decide your level first. The statement writes itself after that.
- Level 0: grammar, spelling, punctuation. No statement. Elsevier says this in plain words, and no other publisher we checked disagrees.
- Level 1: assisted. AI helped you search, summarize sources, outline, generate ideas or check your reasoning. You wrote the words. One sentence naming the tool and the task.
- Level 2: drafted. AI produced a first draft, a translation, a table or a figure that you then edited and verified. Name the tool, the parts it drafted, and what you checked.
- Level 3: generated. The output went out largely as produced, or it is synthetic media (a voice, an image, a video of a person). Name the tool, say it is AI-generated, and put the label on the item itself rather than in the footer. In the EU this is where the AI Act labeling duties bite.
What to include in an AI disclosure statement
Five parts. Most statements that fail leave out the third or fourth.
- The tool and version. "ChatGPT (GPT-5)", "Claude Sonnet 4.5", "Gemini 2.5 Pro", "Microsoft 365 Copilot". A date works where the vendor does not publish a version.
- What it did. Pick verbs that match the level: checked, summarized, outlined, drafted, translated, generated.
- What it did not do. "No data was analyzed by the tool." "No client material was uploaded." This is the line reviewers and clients care about most.
- Who checked. A person reviewed and edited the output and takes responsibility. Name the role if not the person.
- Where it lives. Acknowledgments and cover letter for a paper; methods if AI touched data or figures; the foot of a report; the "About" or byline area of a web page; the caption for an image.
8 AI disclosure statement examples
Each example is written to be pasted and edited. Replace the bracketed parts.
1. Journal manuscript (Elsevier-style declaration)
Declaration of generative AI and AI-assisted technologies in the manuscript preparation process. During the preparation of this work the authors used ChatGPT (GPT-5, OpenAI) to improve the readability of the Introduction and Discussion. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.
2. Methods section, where AI touched the data
We used Claude Sonnet 4.5 (Anthropic) to produce first-pass thematic codes for 212 interview transcripts using the prompt reproduced in Appendix B. Two authors independently reviewed every code against the source text; 31 codes were changed and 9 removed. The model did not have access to participant identifiers.
3. Student essay or thesis
I used Gemini (Google) to generate an outline and to summarize three of the twelve sources cited in this essay. I read all twelve sources in full, wrote every sentence of the essay myself, and checked each summary against the original. This use is permitted under the course AI policy dated [date].
4. Blog post or article on a company website
How this article was made: a member of our team wrote the brief and the outline, used Claude to produce a first draft, then rewrote and fact-checked every section. All figures link to their original source. Last reviewed [date].
5. Client report (consulting or agency)
AI use in this report: Microsoft 365 Copilot, inside our tenant, was used to summarize meeting transcripts and to draft section 4. No client data was uploaded to any public AI service. All findings, figures and recommendations were verified by the engagement team, who are responsible for them.
6. Internal memo or board paper
Prepared with ChatGPT Enterprise (Level 2: drafted, then edited). Source documents: Q3 management accounts and the sales pipeline export. Reviewed by [role] on [date].
7. Customer-facing chatbot or voice agent
You are chatting with an automated assistant powered by AI. It can make mistakes. Type "agent" at any time to reach a person. (Show this before the first exchange, not buried in a footer. In the EU, telling people they are interacting with a machine is a legal duty from August 2026.)
8. AI-generated image, audio or video
Image generated with [tool] from a text prompt; no photograph was used. The person shown does not exist. For a cloned voice: This narration uses an AI voice trained with the speaker's written consent. Put the label on or next to the item, and in the file's metadata if the platform supports it.
AI disclosure statement template (copy and fill in)
Pick the level, then fill the blanks. Delete any sentence that does not apply.
AI disclosure (Level [1 / 2 / 3])
Tool: [name, version or date, vendor]
Used for: [outline / summaries / first draft of sections X / translation / figure Y / generated images]
Not used for: [data analysis / anything involving client or personal data / final wording]
Checked by: [name or role] reviewed and edited all AI-assisted content and takes responsibility for it.
Prompt and output kept: [yes, available on request / attached as Appendix X]
Date: [date]If you work in a company, add this template to your AI acceptable use policy so everyone writes the same thing. A policy that asks for "transparency" without a form to fill in gets ignored, as anyone who has run a policy rollout knows.
Where does the statement go?
For a paper: ICMJE says the cover letter and the manuscript, with writing assistance in the acknowledgments and data or figure work in the methods. For APA-style student work, the APA Style blog asks you to describe the use in the method section or introduction, include the prompt, and cite the tool in the reference list (author OpenAI, year, title, version, URL). For a web page, the byline area or a short "How this was made" note near the end. For a report, the page after the executive summary or the final page, with a line in the transmittal email. For a chatbot, before the first message.
AI transparency statement vs AI disclosure statement vs disclaimer
People search for all three and mean slightly different things. A disclosure statement is about one piece of work: this paper, this report. An AI transparency statement is about an organization: a standing page that says which AI systems you run, what they do, what data they see and how to reach a human. Public bodies in several countries now publish these. A disclaimer is a liability line ("may contain errors, check important information") and does not tell anyone what you actually did. A good disclosure can sit under a transparency statement and may include a disclaimer, but a disclaimer on its own is not a disclosure.
Mistakes that make an AI disclosure useless
- Naming "AI" without the tool. "AI was used" tells a reader nothing. Name it.
- Listing the tool as an author or co-author. Every major publisher forbids it.
- Disclosing Level 0 use at length while hiding Level 2 use. Readers notice the mismatch.
- Skipping "what it did not do." The absence of that line is what makes a client ask whether their data went into a public model.
- No named checker. "The output was reviewed" needs a name or a role after it.
- Putting a deepfake label in the page footer instead of on the media.
- Uploading confidential material to a public tool and then writing an honest disclosure about it. The disclosure does not undo the upload. Read what shadow AI looks like if that sounds familiar.
Need the policy that goes with the statement? AI Setup for Your Company gives you a written AI policy, the disclosure wording above built in, and the rollout notes, set up for your company rather than a generic template.
FAQ
Do I need to disclose using ChatGPT for grammar and spelling?
No, under the publishing rules we checked. Elsevier states that basic checks of grammar, spelling and punctuation do not need a declaration, and the ICMJE guidance targets AI-assisted writing and analysis rather than proofreading. If your employer, course or client says otherwise, their rule applies. When you cannot tell whether your use crosses the line, write one sentence and move on.
Is an AI disclosure statement legally required?
Sometimes. In the EU, transparency duties under the AI Act apply from August 2026: people must be told when they interact with an AI system, and deepfakes and AI text on matters of public interest must be labeled. In the US there is no general federal duty, but the FTC bans fake AI-generated reviews outright, and several states have their own rules. Publishers and universities impose their own requirements by policy, not law.
Can I list ChatGPT or Claude as an author?
No. The ICMJE recommendations say chatbots should not be listed as authors because they cannot take responsibility for the work, and Elsevier, Springer Nature, IEEE and the other major publishers say the same. Credit the tool in a disclosure or acknowledgment, cite specific outputs where your style guide asks for it, and keep the human names on the byline.
How do I cite ChatGPT in APA Style?
The APA Style blog gives this reference format: OpenAI. (2023). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/chat. Cite it in text as (OpenAI, 2023). Describe how you used the tool in your method section or introduction, include the prompt, and add long transcripts as an appendix, because the tool gives a different answer each time.
Does disclosing AI use hurt Google rankings?
Google's published guidance says it rewards high-quality content whatever the production method, and that it does not penalize AI involvement as such. Thin, unoriginal content ranks badly whether a person or a model wrote it. A disclosure line does not change that either way. What matters is whether the page is accurate, specific and useful to the reader.
Where does the AI disclosure go in a research paper?
In two places, per ICMJE: the cover letter and the manuscript itself. Writing assistance goes in the acknowledgments. Any use for data collection, analysis or figures goes in the methods, with the prompt. Elsevier asks for a separate titled declaration at submission. Check the specific journal's author guidelines, since some also require a line in the author questionnaire.
Leave a comment: