A PhD candidate gets a similarity report of 18% and spends a weekend rewriting her entire literature review. The same week, a manuscript scoring 6% is returned by an editor with a formal plagiarism query. The outcomes look backwards because the percentage does not measure plagiarism. It measures how much of your wording matches the database the software happens to hold.
This article walks through a report the way an integrity officer reads one: passage by passage, asking what each match is and why it is there. It is also the best way to protect yourself before you submit.
What the software measures, and what it misses
Similarity checkers compare strings of consecutive words in your text against a corpus of articles, theses, web pages and previously submitted work. The output is a percentage plus a list of sources. The tools do not understand meaning, which leaves three blind spots:
- Idea plagiarism: taking someone's argument, framework or study design and restating it in your own words. Nothing matches, so the score stays near zero.
- Translation plagiarism: translating a passage from another language without attribution. Most tools do not compare across languages.
- Sources outside the corpus: print-only books, internal reports, undigitised theses.
The 6% manuscript fell into the second gap: part of its methods section was a near-literal translation of a paper in Spanish. A reviewer who had read the original spotted it immediately.
Legitimate matches you should leave alone
Before panicking, filter out overlap that was never a problem:
- Your reference list, where journal names and article titles match thousands of other papers.
- Direct quotations in quotation marks with a full citation.
- Standard technical phrases: “structural equation modelling”, “intraclass correlation coefficient”, the names of validated scales.
- Published questionnaire items, when you credit the instrument.
- Institutional boilerplate such as declaration pages and title page templates.
Most systems can exclude the bibliography and quoted material. Ask whoever ran your report whether those filters were on; an unfiltered 18% can drop to 7% once they are.
How to read each match
| Type of match | What it looks like | Concern | What to do |
|---|---|---|---|
| Scattered 5–8 word phrases | Many sources, each under 1% | Usually harmless | Spot-check, no rewrite needed |
| One long passage, one source | A single source at 3–5% or more | Look closely | Quote and cite properly, or synthesise in your own words |
| Match to your own earlier work | Source is your thesis or conference paper | Text recycling if undisclosed | Cite it and tell the editor |
| Match to a classmate's submission | Source listed as “student paper” | Serious | Explain, and produce your original drafts |
The rule of thumb: count the big sources, not the total. Fifteen sources at 1% each worry an editor far less than one source at 8%.
Self-plagiarism: the offence people don't think is one
Self-plagiarism, often called text recycling, means reusing your own published words, data or findings without saying so. Three forms come up again and again:
- Duplicate publication: sending the same results to two journals, or republishing a conference paper almost unchanged as a journal article without disclosure.
- Salami slicing: splitting one dataset into several thin papers, one variable each, to inflate a publication count.
- Recycled text: copying a methods section or background from an earlier paper.
The third form is treated most leniently. Many editors accept a methods description that mirrors an earlier paper when the procedure genuinely was the same, provided you cite it (“Samples were collected as described in…”). COPE's guidance on text recycling follows the same logic: transparency is what matters. The first two forms distort the scientific record, because readers think they are looking at two independent pieces of evidence when there is only one.
A common case is turning a master's or doctoral thesis chapter into an article. That is normal practice, but note in your cover letter that the paper derives from your thesis, and check whether the journal treats theses deposited in a university repository as prior publication.
Paraphrasing versus word-swapping
Swapping synonyms and reshuffling clauses to dodge the software is still plagiarism, because the structure of the argument and the order of ideas remain someone else's. Genuine paraphrase means:
- Reading the source, closing it, and writing what you understood from memory.
- Combining several sources into one sentence that carries your own view: “All three studies report…, but they diverge on…”.
- Citing the source anyway, even when no words overlap.
A quick test: place your sentence next to the original. If they have the same shape, with the same ideas in the same order, you have not paraphrased yet.
A pre-submission routine
- Capture sources while reading. Use Zotero or Mendeley and attach page numbers to every note. Most accidental plagiarism starts as a verbatim note that lost its quotation marks.
- Use two colours in your notes: one for copied wording, one for your own thoughts.
- Run the check on a near-final draft rather than chapter by chapter. Some systems store submissions and will flag your final version against your own earlier uploads.
- Read the report from the largest source down.
- Disclose proactively in the cover letter any related thesis or conference paper.
- Keep dated drafts so you can show how the text developed if questions arise.
If you are accused
Do not delete drafts and do not send a long emotional defence. Send the report you have read, a note on each major match (what it matches and why), drafts with version history, and your reading notes. Investigators usually tell careless citation from deliberate copying on exactly that evidence.
A useful task for this week: open your reading notes, find every verbatim passage that has no source attached, and attach one. That is where most of your real risk sits, not in the percentage.
Câu hỏi thường gặp
What similarity percentage is acceptable?
There is no universal threshold. Institutions and journals set their own reference points, and a human always reads the report. One source accounting for a large share is more concerning than a high total spread across many sources.
Is self-plagiarism really misconduct?
It can be, when you reuse published findings or text without disclosure. Reusing a methods description with a citation to your earlier paper is widely accepted; publishing the same results twice is not.
Can Turnitin detect translated plagiarism?
Similarity checkers generally do not compare across languages, so translated copying often scores low. It is still plagiarism, and reviewers who know the original literature frequently notice it.
Can I publish a chapter of my thesis as a journal article?
Usually yes. Say in your cover letter that the article derives from your thesis and check the journal's policy on theses held in institutional repositories.
Does paraphrasing tool output count as plagiarism?
If the tool only swaps words while keeping the original structure and you do not cite the source, yes. Synthesise ideas in your own words and cite the source regardless.
