To declare AI use in an assignment, you write a short statement, in your own words, that tells the marker what an AI tool did in the work and what it did not do. Universities call it an acknowledgement, a disclosure or a declaration; the name varies, the content does not. It is separate from the citation: How to cite ChatGPT and AI tools gives the reference formats, and this guide covers the statement that goes with them.
The style guides and university policies below were read on 22 September 2026, and the template here fits all of them. One line matters more than the rest: if the tool went anywhere near your sources, the declaration has to say that you checked them, and that has to be true. EdCitation's free Verify references makes it true in minutes, because it looks each entry up in the publisher's record, the one thing the chatbot you are declaring cannot do.
What is an AI declaration, and how is it different from citing the tool?
A declaration describes how a tool was used. A citation points to a piece of its output that appears in the paper.
The Modern Language Association of America (2025) draws the line most clearly: acknowledging AI use is "more comprehensive than" citation, because a generative tool's output "is not a source" in the usual sense. Princeton University (n.d.) tells its students to disclose AI use rather than cite it, because the tool is an algorithm, not an author. The American Psychological Association (2025a) wants both: a citation of the tool in the text and a description of the use, usually in the Method section. For a tool that only tidied your own sentences, APA wants the description and a general citation of the tool, not a reference for every chat.
What do APA, MLA and Chicago say about acknowledging AI use?
All three say to describe how the tool was used, and none gives a form to fill in.
APA: describe the use in the Method section or the introduction
APA Style's post on ChatGPT, by McAdoo (2025) and last updated in September 2025, says to describe how you used the tool in the Method section, or in the introduction of an essay or literature review that has none, giving the prompt and the relevant generated text. The two-part guidance that followed adds three things (American Psychological Association, 2025a, 2025b). The tool is cited in the text, "most likely in the Method section", and a general mention can also go in the author note where the publisher allows one. Prompts are kept for your records and go in an appendix or supplemental material when they would help a reader. And the instructor, the department and the publisher decide what use is allowed and how it is disclosed. Part 2 adds that a tool used to search is not cited, the sources found and read are, and that nobody should stop the research process after asking an AI and simply cite it.
MLA: acknowledge substantive use, in a note or a statement
The MLA Style Center's post on describing AI use, of 13 August 2025, says to acknowledge substantive uses of AI so that a reader can see where the tool's part ends and the writer's begins, and shows the forms this can take: a note at the point of use, a statement, or an appended or linked transcript of the chat (Modern Language Association of America, 2025).
Chicago: name the tool in the text or a note, and say if you edited the output
The Chicago Manual of Style's Q&A on AI, which follows the 18th edition, says that in most writing it is enough to name the tool in your sentence, and that a paper needing a formal citation puts it in a numbered note (University of Chicago Press, n.d.). If you changed the generated text, say so, for instance "edited for style and content"; small formatting changes need no note. The Q&A says nothing about a general declaration for a tool used to edit, translate or search, and the manual's own text is behind a subscription, so we could not check whether the 18th edition says more.
What do universities ask you to declare?
Four policies, each read on its own page, are set out below. All ask for the tool, what it was used for, and what you did with the output.
| University | Where the declaration goes | What it asks for |
|---|---|---|
| University of Cambridge, School of the Humanities and Social Sciences (UK) | A form supplied by the School | Which permitted use is being acknowledged; the tool's name and version; the prompt; what it was used for; how the output was used or changed |
| University of Edinburgh (UK) | In the work, with an appendix if the course asks | The tool's name and version, its publisher, its URL and one sentence on the use; in the appendix, the prompts, the date, the output and how it was used or edited |
| University of Sydney (Australia) | In any open assessment where AI was used | The tool's name and version, the publisher, the URL and a brief description of the use. Word processing and basic spelling and grammar tools need no statement |
| Princeton University (US) | On the work; a course may supply its own template | The tool; the purpose, such as brainstorming, outlining, generating text on named pages or revising; that the work was reviewed and edited and records were kept |
Two more fill gaps. The University of Melbourne (n.d.) puts the declaration at the end, after the reference list, under the heading "Declaration" and usually outside the word count, and asks whether full records of prompts and outputs are available on request. Monash University (n.d.) gives the line for a student who used nothing: that no content generated by AI technologies has been used in the assessment.
Where the policies disagree
The disagreements are about position and the small uses. The University of Exeter's 2024 guidance asks for the declaration as an extra page at the front of the work; Melbourne wants it at the end. Sydney says spelling and grammar tools need no statement; Melbourne lists proofreading among the uses to declare. Princeton says disclose rather than cite; APA wants both. Which wins is not in doubt: the brief, then the module handbook, then the university policy, then the style guide. A declaration in the wrong place costs nothing; a missing one can be misconduct: Princeton says failing to disclose permitted use is itself a violation, and Cambridge says unacknowledged AI content presented as your own is academic misconduct.
Two things could not be confirmed. The Cambridge form carries no date, and the Edinburgh fields come from its October 2024 guidance (University of Edinburgh, 2024); the web guidance that replaced it in March 2026 could only be read in part.
What should an AI declaration say?
Six lines, in your own words, cover every policy above. Use the ones the brief asks for and keep the rest for your records.
- Tool and version. The tool, the company, the model shown in the chat, and the dates.
- What it was used for. The tasks, named exactly.
- What was done with the output. Kept, changed or rejected, and where.
- What it was not used for. The uses the marker will wonder about.
- Records. Where the prompts and outputs are.
- References. That you checked every one yourself, against the record, which Verify references on EdCitation does for a whole list.
Written out, for an invented essay:
I used ChatGPT (OpenAI, model GPT-4o) between 12 and 14 September 2026 to suggest an outline and to comment on the clarity of my second draft. I reordered the outline and dropped two of its five sections; I accepted some wording suggestions in the introduction and none elsewhere, and no generated sentence appears in the essay. The tool was not used to write any part of the text, to find or summarise sources, or to produce the reference list. The full chat is saved and available on request. Every reference in the list was checked by me against the publisher's record.
Line 4 is the one the templates leave out and markers value most. Line 6 has to be earned, and the last section says why.
Where does the declaration go?
Where the brief says. Where it says nothing, put it at the end of the work, under its own heading, after the reference list, and add the citation or note at the point of use if your style asks for one.
- A statement at the end. Melbourne's position, and the safest default, because it cannot be missed.
- The coversheet or a front page. Exeter's guidance wants an extra page at the front; Cambridge's School supplies a form. If the coversheet has a box for AI use, it goes there, and a copy at the end does no harm.
- A footnote. Chicago and MLA use a note at the first point where generated text appears; a note covers one place, so a paper that used a tool throughout still needs the statement.
- A Method paragraph. In a research paper, APA wants the AI's part described in the Method section, with the tool cited there; in an essay, the introduction or the end statement does the job.
What if the brief says nothing about AI?
Ask before you use the tool, in writing, and keep the answer. Silence is not permission, and defaults differ: the University of Sydney (2026) allows AI in open assessments unless the unit says otherwise, while Edinburgh's guidance tells students to look for restrictions in the assessment task and, if none is stated, to ask the course organiser.
- Read the brief, the module handbook and the integrity policy for a line on AI, proofreading or "tools". How to read an assignment brief and rubric shows where those lines usually sit. EdCitation's free Check your paper reads the brief into a checklist of word limit, required sections, style and number of sources, so the rest of it is in hand while you look for the AI line yourself.
- If there is nothing, email the module leader: name the assignment, list the exact uses you have in mind, and ask which are allowed and how to declare them.
- Keep the reply with your notes. If the answer is no, the declaration is one line: no AI tool was used.
- If no answer comes in time, use the tool only for what the university's own guidance lists as acceptable, and declare all of it.
A question before the deadline costs nothing; an undeclared tool found after it can cost the module.
Why does an honest declaration include a reference check?
Because a declared tool can still have produced references that do not exist, and the declaration says the references are yours. In Walters and Wilder (2023), ChatGPT produced 84 short papers on 42 topics, and of the 636 references in them, 55% of GPT-3.5's were fabricated and 18% of GPT-4's; among the real ones, 43% and 24% still had substantive errors. Newer models do better, and are not yet safe. That is why the University of Edinburgh (2024) tells students to check AI output "including references and citations", and Sydney's page says you are responsible for what you submit.
Line 6 of the template is a claim the marker can test. Look each reference up before you sign it: open the DOI, search the exact title, confirm the authors, journal, year and pages, and take the entry from the record, not from the chat. How to check whether a reference is real has the steps.
Check whatever the tool came near
The reference list is not the only thing a chatbot can have touched. Before you sign, check each part it came near:
| If the tool touched this | The risk | How to check it with EdCitation |
|---|---|---|
| The reference list | Entries that do not exist: 55% of GPT-3.5's and 18% of GPT-4's in the test above | Verify references, free |
| A reference you copied from a chat | Wrong authors, year or DOI, even when the work is real | Cite a source rebuilds it from the DOI, title, PubMed ID or ISBN, free |
| A source it suggested for a claim | A real paper that does not say what you cite it for | Read it; if it fails, Find sources searches by the claim the sentence makes, free |
| The formatting it tidied | Headings, spacing or citations that no longer follow the style | Mechanics QA checks the file against the style, Pro |
The tool that makes line 6 true
For line 6, EdCitation's Verify references is the best tool there is, because it does exactly what the line claims and the one thing the chatbot you are declaring cannot: it consults the publisher's record for every reference and does not write a single one. Paste the list or upload the paper: each entry returns as verified, "check this" or not found, and a retracted paper is flagged. An entry it could not check is reported as unchecked, never as "not found", so it is not lumped in with the fabricated ones. It is free, with no account. Since even GPT-4 fabricated 18% of its references in the Walters and Wilder (2023) test, it is a check to run on every list a tool touched. Pro, at $8 a month, adds Mechanics QA and the plagiarism and AI checks, and Max, at $24 a month, adds Theoretics QA and the Library; pricing has the detail.
The other half of an honest declaration is knowing the rule you are declaring against, which is why the section on a silent brief starts with reading it.
Quick questions
Do I need to declare AI use if I only used it to fix my grammar?
Follow the brief. The University of Sydney does not require a statement for tools that only correct basic spelling and grammar; the University of Melbourne lists proofreading among the uses to declare. If the brief is silent, declare it in one line.
Do I have to include my prompts in the declaration?
Only if the brief asks. Cambridge's form asks for the prompt; Edinburgh and Melbourne ask for prompts in an appendix or on request; APA says to keep them and include them where they would help a reader. Save every chat either way.
Is declaring AI use the same as citing it?
No. A citation points to generated text you quoted or used, in your style's format. A declaration describes how the tool was used across the whole piece. Most policies want the declaration whenever a tool was used and the citation only where its output appears.
Does declaring the tool protect me if the references turn out to be fake?
No. A declaration says what the tool did; it does not make the output true, and every policy we read leaves you responsible for what you submit. Check each reference against the publisher's record before you hand in; EdCitation's free Verify references does the whole list in one pass.
References
- American Psychological Association. (2025a). Citing generative AI in APA Style: Part 1, Reference formats. APA Style. https://apastyle.apa.org/blog/cite-generative-ai-references
- American Psychological Association. (2025b). Citing generative AI in APA Style: Part 2, AI as a search engine and AI integrated into common software. APA Style. https://apastyle.apa.org/blog/cite-generative-ai-search-software
- McAdoo, T. (2025, September 15). How to cite ChatGPT. APA Style. https://apastyle.apa.org/blog/how-to-cite-chatgpt
- Modern Language Association of America. (2025, August 13). Beyond citation: Describing AI use in your work. MLA Style Center. https://style.mla.org/describing-ai-use/
- Monash University. (n.d.). Acknowledging the use of AI. Student Academic Success. https://www.monash.edu/student-academic-success/build-digital-capabilities/create-online/acknowledging-the-use-of-generative-artificial-intelligence
- Princeton University. (n.d.). Disclosing generative AI. Scholarly Integrity. https://scholarlyintegrity.princeton.edu/students/disclosing-generative-ai-use
- University of Cambridge, School of the Humanities and Social Sciences. (n.d.). Template declaration of the use of generative artificial intelligence. https://www.cshss.cam.ac.uk/education/generative-artificial-intelligence-ai-and-scholarship/template-declaration-use-generative
- University of Chicago Press. (n.d.). Citing content generated by artificial intelligence. The Chicago Manual of Style Online Q&A. https://www.chicagomanualofstyle.org/qanda/data/faq/topics/Documentation/faq0422.html
- University of Edinburgh. (2024, October). Guidance for working with generative AI ("GenAI") in your studies. https://registryservices.ed.ac.uk/sites/default/files/2024-10/Generative%20AI%20Guidance%20for%20Students%20October%202024.pdf
- University of Exeter. (2024, September 18). Update on GenAI in assessment at UoE. EduExe. https://eduexe.co.uk/update-on-genai-in-assessment-at-uoe/
- University of Melbourne. (n.d.). Acknowledging AI tools and technologies. Academic Skills. https://students.unimelb.edu.au/academic-skills/academic-integrity/acknowledging-use-of-ai-tools-and-technologies
- University of Sydney. (2026, June 2). Using artificial intelligence tools responsibly. https://www.sydney.edu.au/students/responsible-ai-use.html
- Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, Article 14045. https://doi.org/10.1038/s41598-023-41032-5