Why this exists
This site tells people to disclose AI assistance on a dozen pages and has never helped anyone write the sentence. The instruction is easy to give and genuinely awkward to follow — most people are not sure what counts as assistance, how specific to be, or whether admitting to a grammar check makes things worse.
The most common mistake is vagueness. “AI was used in the preparation of this work” describes nothing, and a reader who wanted detail now has to ask for it. Naming what the tool did — outlined, drafted, translated, corrected — is both more useful and easier to stand behind.
What counts as assistance
Wider than most people assume, and the boundary is set by your institution rather than by the tool. Grammar correction, predictive text, translation and summarising a source all sit inside some policies and outside others. Our answer on what counts as AI assistance works through where the rule actually lives, and academic integrity shows the same piece of work judged under three different policies with three different outcomes.
The rule that binds you is usually narrower than the university-wide policy people tend to read. A syllabus AI policy sits on top of it and can be stricter, and in US institutions an honour code may govern instead, which shifts the interpretation onto you rather than onto a list of prohibited acts. What a finished declaration should contain is set out in AI disclosure statement, with three versions of the same statement compared.
If you are drafting a statement because you have already been questioned about a detection score, that is a different situation and the priority order changes — start with what to do when you are accused.