The rule is about authorship, and it is local.
Assistance is not the thing institutions police. Libraries employ writing tutors, faculty read drafts, statisticians are consulted, and none of that is contested anywhere. What is policed is authorship: whether the ideas, the argument and the decisions in a submitted piece are attributable to the person whose name is on it, in the way that particular course requires.
Where that boundary falls is a local question and not a universal one, which is why generic reassurance from any service is worth nothing. Your handbook and your syllabus settle it, and where they are genuinely ambiguous, an emailed question to the instructor settles it better still. That reply is worth keeping, because a written answer from the person who grades you is the strongest single document you can hold.
What this desk will and will not do.
A service that refuses nothing has told you something about the value of its promises. Ours are published, they sit alongside the desk's written terms, and they are short enough to read in a minute.
- It will read a rubric and explain what the top band actually asks for, in writing
- It will teach a method, work an example, and check whether your version of it holds
- It will locate and vet current sources, then verify that every citation survives being clicked
- It will edit for structure, argument and mechanics, and coach a draft toward the highest band it can honestly reach
- It will not invent a source, a quotation, a figure or a dataset, under any deadline
- It will not supply or endorse anything that misrepresents supervised clinical hours or a preceptor's sign-off
- It will not guarantee a grade, because nobody controls the grader
- It will not advise anybody on how to present work so that a similarity or writing check reads it differently, which is the subject of the next two sections and the one request we simply decline
How a similarity report is actually read.
A similarity system such as Turnitin compares strings in your document against a corpus of published work, web pages and previously submitted student papers, then reports where matches occurred and how much text they cover. That is the whole mechanism. It does not know what plagiarism is, it has no opinion about intent, and the headline figure it produces is a measurement rather than a finding.
Which is why experienced instructors open the breakdown instead of reading the number. Quoted material, reference lists, standard methodological phrasing, the assignment prompt reproduced at the top of the page and required template headings all generate matches in papers that are entirely the student's own, and a paper can therefore report substantial similarity while being beyond reproach. The reverse also holds: a low figure demonstrates only that no match was found in one corpus. What draws attention is the pattern rather than the total, and specifically whether matched text forms continuous unquoted passages tracing back to a single source.
An AI indicator is a guess about text, not a record of events.
A writing-origin indicator works by measuring statistical properties of prose and comparing them against patterns typical of machine-generated text. It has no access to how a document came into being. It did not watch you write, and it cannot, in principle, establish authorship; it can only report that a passage resembles or does not resemble a distribution. Such a measure can be wrong in both directions, and it is wrong most often for writers whose style is unusually plain, unusually formulaic, or produced in a second language.
This column will not tell anybody how to make writing read differently to such a check, and a desk that offers to is selling you a risk it has priced at nothing. The advice is useless quite apart from being dishonest, because the question in an integrity meeting is never how the text scored. It is what evidence exists that the work is yours, and no amount of rewording produces evidence. Keeping it, however, is straightforward, and it is the subject of the rest of this column.
Provenance, which answers the question for you.
Provenance means being able to show how a document came to exist. It costs nothing while you are working and it cannot be reconstructed afterward, which is the entire argument for the habits below. None of them assumes you will ever be asked.
Should a question arrive, these are what answer it, and they answer it far better than any argument about a tool. Marked-up drafts count here too, which is a quiet secondary benefit of working with an editor: the correspondence and the tracked revisions are themselves a record of the work being built. That, rather than rescue, is the ordinary business of the copy desk.
- Draft in one file from the beginning, in software that keeps version history, and let that history accumulate
- Never paste a finished piece into a fresh empty document, since doing so discards the record of how it was built
- Keep the outline, the reading notes and the annotated sources in the same folder as the draft
- Keep the marked-up copies you were sent by a tutor, a writing center or an editor, with their dates intact
- Keep the instructor's replies to any question you asked about what was permitted
- Save your own rubric self-score, which shows you working against the criteria before submitting
Questions to the desk.
Does a high similarity score mean a paper was plagiarized?
No. A similarity system reports where text matches a corpus, nothing more. Quotations, reference lists, standard methodological phrasing and reproduced assignment instructions all raise the figure in honest work. What matters is the breakdown: whether matched text forms continuous unquoted passages from one source, or is scattered across citations, templates and common phrases. The number alone settles nothing in either direction.
Can an AI detector prove who wrote something?
No, and it does not claim to. Such a tool measures whether prose resembles patterns typical of machine-generated text and reports a likelihood. It has no record of how a document was produced, so it cannot establish authorship and can be mistaken in both directions. That is precisely why documentary provenance, rather than argument about the tool, is what resolves a question about a paper.
What should I do if I am asked about a submission?
Reply promptly and ask what specifically is in question, since the answer determines what is relevant. Then offer your file history, outline, notes, annotated sources and any correspondence about the assignment, and be ready to talk through your argument. Do not argue about the detection tool itself. Institutions have procedures for this and using them calmly is the correct response, including asking about representation or appeal rights.
Is using a tutor or an editor against the rules?
Tutoring, editing, research counsel and placement support are long established and ordinary at every level of study, and most handbooks say so explicitly. How far a student may go beyond those is a policy question that only your own institution answers. Read the handbook first, ask the instructor where it is unclear, and keep the reply. Whatever a service says about itself, look for the list of things it refuses.
