Writing an AI Disclosure Statement for Coursework
Most schools now have a line in their integrity policy saying that machine assistance must be acknowledged. Far fewer say what the acknowledgement should look like, so students improvise: a sentence in the footer, a vague "I used AI for research", or nothing at all. An AI disclosure statement fills that gap. It is a short, structured account, written by the student and submitted with the work, of what assistance was used, for which tasks, and what the student did with the result.
A teacher cannot tell from a finished essay how it was made; the site's essay on assessment when product severs from process explains why no inspection of the artifact restores that link. A disclosure statement asks the student to report the process directly, on the record, at submission. It gives the teacher something to read instead of something to suspect, and it is the student's first exhibit should a question arise. The Academic Integrity hub treats it as routine.
The short answer. An AI disclosure statement names each tool used, states the task it served, records what the student kept, changed or rejected, and says how the output was checked. It runs from one sentence for a school essay to a page for a dissertation, sits in the front matter, and is signed and dated.
What the statement is for
The first purpose is evidentiary. Detection software estimates whether text resembles machine output; it does not establish who did the work. Liang and colleagues showed in 2023 that widely used detectors flagged a majority of essays by non-native English writers as machine generated while passing native speakers' work. A disclosure statement is a different kind of witness: a first-person account, made at the time, that can be checked against the draft trail and tested in the short exchange the oral exam piece describes.
The second purpose is pedagogical. Writing the statement makes the student describe their own process, the metacognitive act that writing to think argued is the real purpose of school writing. Collins, Brown and Holum's 1991 essay on making thinking visible named articulation as a core move of apprenticeship; the statement is articulation applied to the tools.
The third purpose is fairness. A student who discloses accurately and is then questioned has proof of good faith, which is why the piece on what a student can show when falsely accused of using AI puts the statement first.
What an AI disclosure statement should contain
The statement answers five questions, in whatever order the school prefers.
- Which tools. The name and version of each system used.
- For what tasks. Outlining, summarising a source, explaining a concept, drafting a passage, rewording, checking grammar. Each task gets its own line, since a school may permit some and forbid others.
- What was kept. Which parts of the submitted work contain machine text, verbatim or lightly edited, located by section or paragraph.
- What was rejected or changed, and why. This is the line that carries the learning. A student who can say why an output was wrong has shown command of the material.
- What was checked. How the student verified anything the model asserted. Language models produce confident falsehoods; a statement that records no checking has recorded a risk.
Add a dated signature, and expect the statement to agree with the school's other process evidence. A paragraph declared as machine-drafted should appear in the document's version history as a single paste; one declared as the student's own should appear as composition over time. Where the two disagree the teacher has a question to ask.
Levels of assistance and what must be declared
Not every use needs the same treatment, and a good policy says so. UNESCO's 2023 guidance on generative AI in education and research asks institutions to set explicit rules on acceptable use, and the QAA's sector resources on generative artificial intelligence take the same line for UK higher education. The last column describes the pattern in published guidance; the local policy governs, and the working policy template shows where the statement fits.
| Level of assistance | Example | What must be declared | Usually permitted for graded work? |
|---|---|---|---|
| None | Student wrote and revised alone | One line stating that no generative tools were used | Yes |
| Mechanical | Spell check, grammar suggestions, citation formatting | Usually nothing; name the tool if the policy asks | Yes |
| Explanation and study | Asking a model to explain a concept before writing | Tool, topic, and that no output entered the text | Usually yes |
| Feedback on a draft | Model comments on the student's own paragraph | Tool, what was asked, which suggestions were acted on | Often yes, sometimes restricted |
| Outline or structure | Model proposes a plan the student writes to | Tool, the prompt, how far the plan was followed | Varies by assignment |
| Passage generation | Model drafts sentences that appear in the work | Tool, exact passages, edits made, checking done | Rarely, and only when the brief says so |
The bottom row is where policies bite. Cambridge's guidance states that submitting unacknowledged AI-generated content in a summative assessment as one's own work is academic misconduct unless the brief says otherwise. The offence is the missing acknowledgement, which is what the statement supplies. A disclosed passage may cost marks; an undisclosed one is a different category of problem.
How long it should be, and where it goes
Length follows the level of use. For an essay with no assistance, one signed sentence. Where a model helped with outlining or feedback, three to six sentences. For a dissertation or extended project, a section of its own, half a page to a page, often with a table by chapter. A statement longer than the assistance warrants reads as padding; one shorter reads as concealment.
The statement sits in the front matter, after the title and before the body, where a plagiarism declaration traditionally goes. A copy also belongs in the process portfolio, beside the drafts and versions it describes. On Moodle or Canvas it can be a required field on the submission form; it must be dated and not editable afterwards. Read across a class, statements also show which tasks students delegate when work is set as homework.
The statement is distinct from citation. Where a specific output is quoted or closely paraphrased, the rules on citing AI assistance in academic work apply at the point of use; the statement is the whole-document summary.
Two templates a student can copy
Bracketed placeholders are replaced with the actual tool and version.
A short statement for a school essay
Declaration of AI use. In preparing this essay I used [tool and version] for [task, for example: generating three counterarguments to my thesis in section 2]. I kept [for example: one of the three, rewritten in my own words] and rejected the others because [reason]. I checked [for example: the two dates the model supplied, against the textbook]. No other generative tools were used and all other text is my own. Signed [name], [date].
For a student who used nothing: "No generative AI tools were used in preparing this work. Signed [name], [date]." That sentence turns silence into a declaration.
A longer statement for a dissertation or project
Statement on the use of generative AI.
>
Tools used. [Tool A, version], accessed via [interface], between [dates]. [Tool B, version] for [narrow purpose].
>
Purposes. Tool A was used to (1) explain [concept] during the literature review, with no output entering the text; (2) comment on a draft of section 4.2. Tool B checked the grammar of the final draft.
>
Text derived from a tool. Paragraphs 2 and 3 of section 4.2 began as machine-generated drafts responding to my prompt "[prompt]". I rewrote both substantially; the surviving machine phrasing is the sentence beginning "[first words]", cited in the text.
>
Verification. Every reference the tools proposed was traced to its source; [number] could not be found and were discarded. Figures in section 4.2 were checked against [dataset].
>
Declaration. The analysis, argument and conclusions are my own. I have not used generative tools for any purpose not listed above. Signed [name], [date].
The longer template asks for prompts and locations so an examiner can reproduce what was done.
Frequently asked questions
Do I need an AI disclosure statement if I only used a grammar checker?
Usually not for the mechanical tier, though many schools ask students to name the tool anyway. The safe rule is to read the assignment brief and, if it is silent, write one sentence naming the tool and the task. A sentence costs nothing; an omission can cost a great deal.
Can a teacher penalise me for disclosing AI use?
A teacher can apply the assignment's rules to disclosed use: if passage generation was forbidden and you declare it, marks may follow. What disclosure removes is the more serious charge of misrepresentation. Under most policies, including Cambridge's, the misconduct lies in the missing acknowledgement.
How detailed should the disclosure be?
Detailed enough that a reader could reconstruct what you did: which tool, which task, which parts of the text, what you changed and what you checked. For a school essay that is a few sentences; for a dissertation, a page with a table by chapter. If unsure whether to include a use, include it.
Is a disclosure statement the same as citing AI?
No. Citation acknowledges a specific output at the point where it is used, following the style guide's rules. The disclosure statement summarises all assistance across the whole document. Substantial work containing machine-derived text normally needs both: a citation where the text appears and a statement at the front.
Where this leaves a school
A school that requires an AI disclosure statement with every substantial piece of coursework has stopped asking teachers to infer process from product, which the finished text cannot support, and started asking students to report process directly, which they can. The Academic Integrity hub gathers the wider argument; the piece on version history as evidence of authorship is the natural next read.
The statement pairs with a citation practice, described in how to cite AI assistance in academic work, so that the front-matter declaration and the reference list tell one story.