Disclosure Without Dread: Drafting an AI Acknowledgement That Satisfies a UK Examiner

A decade ago, the acknowledgements page of a UK dissertation was a place for thanking librarians, long-suffering partners and the participant who answered the recruitment advert at the eleventh hour. Today it is increasingly a place where examiners look for something else: an honest account of the tools that helped produce the work. Institutions have moved quickly, and many postgraduates are still working from guidance written two or three academic years ago.

What follows is not a template to be copied wholesale. Policies differ between institutions, faculties and even individual modules, and the only authoritative version is the one attached to your own programme handbook. It is, instead, a way of thinking about disclosure so that whatever you write can survive a second reading by someone who is paid to be sceptical.

Why the statement exists at all

An AI acknowledgement is not a confession and it is not a penalty. It is a record of method, and it belongs alongside the other disclosures a dissertation already makes: which software performed the statistical analysis, which archive supplied the documents, which transcription service handled the interviews. Examiners are not, in the main, interested in whether a language model was open on a second monitor. They are interested in whether the intellectual work in front of them is attributable to the candidate.

The UK Quality Code for Higher Education places the responsibility for academic integrity with the institution, but it does not prescribe a single national form of words. That is why the sector currently contains a spectrum: some universities require a formal declaration in the front matter, others a structured table of tools, others a checkbox on the submission portal with an attached narrative, and a handful still expect nothing unless a module leader asks.

Reading your own rulebook properly

Before drafting a single sentence, establish which document governs you. Three layers usually apply, and they can conflict.

• The institutional academic integrity policy, which sets out the categories of permitted and prohibited use.
• The faculty or school guidance, which often adds subject-specific rules — for example, restrictions on uploading unpublished data to third-party services in laboratory disciplines.
• The module or programme handbook, which occasionally imposes stricter requirements than either of the above, particularly on taught masters dissertations.

Where two documents disagree, the stricter one is the safer guide, and a short email to your supervisor confirming your reading is worth the two minutes it takes.

Assistance and authorship: where the line sits

The disclosure question is often framed as "did you use AI?", which is the wrong question. Almost everyone now uses it in some form, if only through a grammar checker embedded in a word processor. The question that actually matters is whether the intellectual contribution remains yours: the framing of the research problem, the selection and interpretation of evidence, the argument advanced in each chapter, and the claims you would be prepared to defend in a viva.

In practice, that distinction separates activity into two recognisable groups.

Generally regarded as legitimate, subject to declaration: using a tool to suggest sentence-level edits where the meaning is already yours; generating practice questions on a topic you have read; debugging code; producing a first-pass summary of an article you have read in full and then verified; exploring possible search terms before a systematic review.

Generally regarded as misconduct: submitting prose generated in response to your research question as your own analysis; asking a tool to produce references or paraphrases of sources you have not consulted; using a model to reword text specifically to evade similarity detection; uploading confidential or personally identifiable participant data to a service not approved by your ethics committee.

The grey zone sits in the middle, and it is wider than most students assume. Using a model to propose a paragraph structure for a section you then write yourself sits closer to legitimate editing. Using it to draft that section and then altering the vocabulary sits closer to misconduct, however much editing follows.

The elements a defensible statement tends to contain

A statement that examiners find persuasive is specific, bounded and verifiable. Six components recur:

1. The tool, named precisely, with version or provider where known.
2. The purpose, described functionally rather than euphemistically.
3. The extent, indicating which chapters or tasks were affected and roughly how much of the work.
4. The human contribution, making clear what you did that the tool did not.
5. The verification step, explaining how outputs were checked against sources or data.
6. The date or period of use, particularly where tools changed capability mid-project.

A workable example, to be adapted rather than reproduced:

I used [tool, version] during the drafting of Chapters Two and Three as a structural aid. I supplied my own notes and asked it to propose alternative paragraph orders; I selected and rewrote the structures I adopted. It was also used to check grammar and sentence clarity in approximately 20 per cent of the final text. All citations were located and verified by me in the original sources. No generated text was submitted as analysis, and no primary data were uploaded. Used between March and August 2026.

Wording that tends to attract scrutiny

Certain formulations are read as evasive, and evasiveness is what prompts a referral rather than the use itself. Phrases such as "AI was used appropriately", "some assistance with editing was obtained" or "the tool helped throughout" supply no scope and no verification, and they sit awkwardly beside a version history that shows heavy generated text.

Two further errors are worth flagging. The first is naming a tool you cannot evidence, perhaps because a friend recommended a service you never actually opened. The second is producing a statement that contradicts your own draft history — a chapter written in a fortnight, in a register markedly different from your earlier work, accompanied by a declaration that reads "grammar only".

Keeping a record you could produce on request

Disclosure is retrospective, so the evidence has to exist before you need it. Keep a short working log — dated, one line per session — of what you asked, what you used, and what you discarded. Retain your document version history rather than overwriting a single file. If you have used a tool for anything beyond proofreading, keep the prompts. None of this is normally submitted, but it is the difference between a query that is resolved in a week and one that occupies a term.

A five-minute check before submission

Re-read your statement against your own chapters and ask whether a reader comparing the two would find it consistent. Confirm the tool names are accurate. Confirm the sections named are the sections affected. Confirm the verification sentence describes something you actually did. Then check the format your institution requires — front matter, portal field or separate form — because a perfect statement in the wrong place is simply an omission.