Disclosing AI Use to Your Institution, Honestly and Specifically
Every institution that has addressed AI use in written work has landed somewhere different. Some permit it freely for drafting and outlining. Some allow it only for editing an already-finished piece of your own writing. Some ban it outright for any part of the process. A growing number sit in between and simply require you to disclose what you used and how, leaving the judgment about whether that use was appropriate to a reader who now has the information to make it. This piece cannot tell you which of those rules applies to you, because it genuinely varies enormously and changes over time even within a single institution. What it can do is make the disclosure itself easier to write honestly, once you know your policy requires or rewards one.
Find your actual policy before you write anything else
The single most important step in this entire process happens before you touch your document: locate your institution's actual, current, written policy on AI use, whether that is a university-wide academic integrity statement, a course syllabus, a journal's author guidelines, or an employer's internal policy. Do not assume your last course's rule carries over to this one, and do not assume a friend's experience at a different institution tells you anything about yours — this is exactly the area where policies diverge the most, sometimes between two courses in the same department. If the policy is ambiguous about a specific use you are considering, ask the instructor, editor, or supervisor directly rather than guessing in the direction that is convenient for you. A five-minute question up front is cheaper than a difficult conversation after submission.
Why "I used AI a little" is not a useful statement
A vague disclosure fails the one job a disclosure is supposed to do: give your reader enough information to evaluate the work appropriately. "I used AI a little" could mean you ran a spell-check-adjacent grammar pass on a finished draft, or it could mean an AI tool wrote three of your six sections and you lightly edited them — two enormously different situations that a vague statement makes indistinguishable. It is also not falsifiable or checkable by the reader, which quietly shifts all of the trust burden onto their willingness to take your word for it. A specific disclosure protects you precisely because it is checkable: it names what you actually did clearly enough that a skeptical reader could ask a follow-up question and get a coherent answer, rather than a shrug.
Turning your process into a specific, quantified statement
Vagueness is usually a symptom of not having actually tallied what happened, not a deliberate choice to obscure it. The words-to-pages arithmetic gives you a concrete way to quantify the honest answer instead of reaching for hedge words. Take a 3,000-word report, formatted double-spaced, which converts to about 12 pages using the standard page-count baseline. If you used an AI tool only to propose section headings early on — pure structure, discarded and rewritten in your own words once you started drafting — the honest, specific, checkable claim is that zero of those 12 pages contain AI-generated prose that made it into the final document, and you can say exactly that. Contrast that with a report where you asked the tool to draft entire sections that you then edited rather than replaced: at that point you cannot honestly claim zero pages, and the specific, defensible move is to estimate as best you can which sections were AI-drafted-then-edited and name them, rather than reaching for an unquantified "some assistance" that hides more than it reveals.
What belongs in a disclosure statement
A genuinely useful disclosure statement, regardless of the exact format your institution asks for, tends to answer four questions plainly: which tool or type of tool you used (a general-purpose language model, a grammar checker, a specific research assistant); what you used it for, stated as specifically as you can (outlining a structure, explaining a concept you were stuck on, checking readability of a finished draft, drafting specific named sections); how much of the final document reflects that use, quantified where you reasonably can be, as in the page-count example above; and what you did to verify anything factual the tool contributed, since a disclosure that names AI-assisted research without a verification step invites exactly the follow-up question you would rather answer up front. None of these four require you to admit to anything the policy prohibits — they simply require you to describe what actually happened instead of what sounds acceptable in the abstract.
Disclosure is not a workaround for prohibited use
It is worth being direct about one thing disclosure does not do: it does not turn a use your institution's policy prohibits into an acceptable one. If your policy bans AI-generated prose in submitted work and you disclose that you used it anyway, you have been honest about a violation, not excused from it. Disclosure exists for the large space of uses that are permitted, encouraged, or judged case-by-case, where the reader genuinely needs the information to evaluate your work fairly. Treat disclosure as a tool for accurately describing legitimate assistance, and treat a policy that says "not permitted" as meaning exactly that, regardless of how thoroughly you plan to document what you did instead.
The same report, a heavier-use version
It helps to see the quantified approach applied to a messier, more realistic case than the clean zero-pages example above. Suppose the same 3,000-word, 12-page report also included a section where you asked the tool to draft a paragraph explaining a technical process, which you then substantially rewrote in your own words after checking its claims against a source — versus a different section where you accepted a generated paragraph with only light copyediting. The first case is closer to the structure-and-explanation use this site generally treats as defensible, since the surviving prose is yours even though a generated paragraph helped you understand the process; a reasonable disclosure might describe it as "used to help draft an initial explanation, rewritten and verified before inclusion." The second case is the one that needs a more direct accounting: if a section's prose is still substantially the tool's sentences, disclose that specifically — which section, roughly how much of it, and what if any verification you did on its claims — rather than folding it into the same general sentence as the first, lighter use. Lumping different intensities of use into one vague disclosure line is exactly the pattern that makes a statement unhelpful, even when every individual fact in it is technically true.
Disclosure formats vary by context, and that is fine
What a disclosure statement looks like differs by where it is going. A course assignment might want a short paragraph at the end of the document, following a template the instructor provided. A thesis or dissertation might require a dedicated methods-adjacent section describing AI use across the whole project rather than per-chapter. A journal submission might require a specific checkbox-style declaration plus free-text detail, distinct from the acknowledgments section. An employer might want nothing more than a note in a project log. None of these formats change the underlying content that makes a disclosure useful — what you used, for what, how much of it survived, and what you verified — they just change where that content needs to live and how formally it needs to be phrased. Write the honest, specific version first in plain language, then adapt its wording to whatever template or format is actually required.
When policies change mid-project
Long projects — a thesis, a multi-semester research program, a manuscript under review for months — sometimes outlast the policy that was in effect when the work began, as institutions update their guidance in response to how the tools are actually being used. If you notice a policy has changed partway through a project, the safe move is to check with whoever set the policy about which version applies to your submission, rather than assuming the older, more permissive version still covers you. Keeping your own running note of what you used and when, even informally, makes this conversation far easier, because you will not be reconstructing your process from memory under time pressure.
The habit that makes this easy
The best time to prepare an accurate disclosure is throughout the project, not the night before submission. Keep a simple, ongoing note of every substantive AI interaction as it happens: what you asked for, what kind of output you got, and whether any of it survived into your document in any form. By the time you need to write the disclosure statement itself, you are summarizing a record you already have rather than trying to honestly reconstruct a process you only half remember. That habit costs almost nothing to maintain and removes the single biggest reason people default to a vague statement — not because they intend to hide anything, but because an accurate accounting feels effortful to produce after the fact. Make it effortless by keeping the record as you go, and specific disclosure stops being a burden and becomes just the last thing on a checklist you already have.