What a declaration has to answer
- Which tool, and which version or date of use.
- What task it was used for — searching, summarising, drafting, checking language.
- What you did with the output: verified, edited, rejected, or reproduced.
- What remained entirely human: judgement, interpretation, conclusions.
Vagueness is what draws scrutiny. "AI was used to assist with this dissertation" says nothing; "a language model was used on 12 March to summarise twelve abstracts, each of which was then read in full and corrected by the author" says everything.
The AI Assessment Scale
Many UK institutions now describe permitted AI use on a five-level scale. Check your own department's wording, but the levels generally run:
- No AI — the assessment is completed without generative tools.
- AI for planning — brainstorming, structuring, or idea generation; no AI-generated text in the submission.
- AI for collaboration — AI supports drafting or editing in parts of the work, with the contribution disclosed.
- Full AI — AI is used throughout, with critical evaluation of its output by the student.
- AI exploration — AI use is itself part of what is assessed.
Name the level your work sits at, then give the specifics. A level without detail is half a declaration.
US syllabus statements
US courses more often use a three-tier framing: AI prohibited, AI permitted with disclosure, or AI encouraged. The same rule applies — state which tier the course sets, and describe your actual use within it.
ICMJE and COPE: AI is never an author
Both bodies hold that authorship requires accountability, and a tool cannot be accountable. AI cannot be listed as an author or co-author, cannot be cited as a source of authority, and cannot carry responsibility for accuracy. Where AI contributed, it belongs in an acknowledgement or methods statement, not the author line.
You remain answerable for every sentence, every reference, and every number — including anything a tool produced and you accepted.
A worked example
"In preparing this dissertation I used a generative AI assistant at AI Assessment Scale level 2. Between 3 and 19 March 2026 it was used to summarise open-access abstracts identified through Europe PMC and to suggest alternative phrasings for the discussion. Every summary was checked against the source article by the author. No AI-generated text appears verbatim in the submitted work. Study selection, data extraction, risk-of-bias assessment, interpretation and conclusions are entirely the author's own. AI is not an author of this work."
What not to do
- Do not describe tools as undetectable, or seek to avoid detection — that reframes ordinary use as misconduct.
- Do not disclose at the end from memory. Write it as you work.
- Do not cite an AI tool as evidence for a clinical claim; cite the underlying study you verified.
- Do not assume silence is permitted. If your department's policy is unclear, ask in writing.