Plagiarism is presenting someone else's words, ideas, structure or data as your own. Intent is not part of that definition, and this is the detail that catches people out: a panel asks whether the work misrepresents whose thinking it is, not whether you meant it to.
This post covers the seven forms it takes, the two that account for most genuine accidents, what a similarity score does and does not tell you, and a note-taking method that removes the problem at its source.
Every claim traced to a source
ThesisAI retrieves papers from indexed academic databases and verifies each inline citation against the source it came from, so attribution is built in as the draft is written rather than reconstructed afterwards.
Zacznij pisaćThe Seven Types
| Type | What it looks like |
|---|---|
| Verbatim | Copying text word for word without quotation marks, whether or not a citation follows |
| Mosaic (patchwriting) | Keeping the source's sentence structure and swapping in synonyms |
| Inadequate paraphrase | Genuinely rewording, but omitting the citation, so the idea reads as yours |
| Self-plagiarism | Reusing your own previously submitted or published work without disclosure |
| Source-based | Citing a source you did not read, or a fabricated one, or citing A when your information came from B quoting A |
| Structural | Reproducing another work's argument sequence or section architecture as your own framing |
| Data and image | Reusing figures, tables, code or datasets without permission and attribution |
Verbatim copying is the one everyone knows about and the one that almost never happens by accident. The two that produce most real cases are mosaic and inadequate paraphrase, and both are created at the note-taking stage rather than the writing stage.
Mosaic Plagiarism: The One That Feels Safe
Patchwriting feels like paraphrasing because you changed the words. You did not change the sentence. The structure, the clause order and the logical moves are still the author's, and a matching tool finds this easily because the distinctive nouns usually survive.
Source: "Rapid urbanisation in coastal regions has substantially increased exposure to storm surge, particularly where informal settlements have expanded into low-lying floodplains."
Weak (mosaic): Fast urban growth in coastal areas has significantly raised vulnerability to storm surge, especially where informal housing has spread into low-lying flood zones (Okonkwo, 2024).
Every content word was swapped and the citation is present, but the sentence is still the author's sentence. This is reported as plagiarism at most institutions.
Strong: Okonkwo (2024) identifies a specific mechanism behind rising storm surge damage: the risk is driven less by coastal population growth in itself than by where that growth lands, with informal settlements concentrated in the lowest-lying areas.
The sentence starts from the author, restructures the claim around what it actually argues, and adds an interpretive move that is the writer's own.
The test is mechanical. Cover the source, wait a few minutes, and write the point from memory. If you cannot, you have not understood it well enough to paraphrase it yet, and the honest options are to quote it directly or to go back and read it again. Our guide to paraphrasing works through the four-step method in more detail.
Self-Plagiarism Is Real and Catches Good Students
Reusing your own work without disclosure is a genuine offence at most institutions, and students are regularly surprised by it. The logic is that each submission claims to be new work produced for that assessment, so resubmitting material you were already credited for claims credit twice.
Common situations:
- Recycling a literature review from a coursework essay into your thesis
- Submitting substantially the same work to two modules
- Reusing your own published paper's methods section in your thesis without a note
- Building your thesis on your own earlier research proposal
The fix is nearly always disclosure rather than avoidance. Universities generally permit building on your earlier work where it is declared. State it plainly in your methods or introduction, cite your earlier work like any other source, and ask your supervisor in writing if the situation is unusual. A reused methods section with a sentence explaining the overlap is fine. The same section presented silently is not.
What a Similarity Score Actually Means
Turnitin and similar tools report a percentage of text matching their index. That number is not a plagiarism score, and treating it as one causes two opposite mistakes.
| Situation | Reality |
|---|---|
| A 30% score | Often entirely fine. Correctly quoted material, the reference list and standard methodological phrasing all match |
| A 3% score | Can still be a serious case. A single uncited paraphrased paragraph may match almost nothing |
| "Under 15% is safe" | Not a real rule. No threshold makes uncited work acceptable or correctly cited work suspect |
What matters is the match report, not the headline number. Open it, look at each highlighted passage, and ask whether that passage is quoted, cited and attributed correctly. A long unbroken match against a single source is the pattern that worries a marker, regardless of the total.
The Method: Fix It at the Note-Taking Stage
Nearly every accidental case traces to the same origin. Months earlier, you pasted a passage into a notes file meaning to rewrite it later, and by the time you wrote the chapter you could no longer tell which words were yours.
The solution is a colour or tag convention applied without exception from the first source you read.
Three-category note-taking
Quoted: exact words, in quotation marks, with author, year and page number. Never store source text without the marks.
Paraphrased: your wording of their point, with author, year and page. Written from memory with the source closed.
Mine: your own reaction, critique, connection or question. This is the material that becomes your contribution.
Two rules make it work. Record the page number at the moment you take the note, because reconstructing it later is the step everyone skips. And never paste source text into your draft file "to fix later" - the later self who reads it will not reliably know its origin.
Three habits that prevent the rest:
- Cite while drafting, not afterwards. A retrofitted citation pass is where attributions get lost.
- Read what you cite. If you found a source quoted in another paper, either read the original or cite it as quoted in the secondary source. Our post on primary vs secondary sources covers when that is acceptable.
- Attribute structure too. If your comparison framework comes from a published review, say so. Borrowed architecture needs credit as much as borrowed sentences.
Where AI Tools Fit
Two distinct risks come with generative tools, and they need different responses.
Fabricated references. Language models produce citations that look real and are not. Citing one is recorded as fabrication, and intent is not a defence. Open every source before it enters your reference list, and delete anything you cannot locate through a library catalogue or its DOI.
Undisclosed generated text. Submitting AI-drafted prose as your own unaided writing breaches most institutional policies, whether or not it triggers a similarity match. The policy question is about authorship, not detection, and AI detectors are unreliable in both directions, so neither passing nor failing one tells you much.
Disclosure and citation are different obligations here. Our guides to using AI ethically in academic writing and citing ChatGPT and AI tools cover which situation calls for which.
If You Are Accused
Do not respond in a panic, and do not begin by insisting it was accidental. Institutions handle this through a defined process, and there are sensible steps.
- Ask for the specific passages and the report in writing before you reply to anything
- Gather your notes, drafts and version history - a document history showing gradual development is strong evidence of your own process
- Check your institution's regulations for the procedure and your right to be accompanied
- Contact your students' union or academic advisor, who deal with these cases regularly
- If it was an error, say so directly and explain the mechanism. A specific account of how a citation was lost is more credible than a general denial
FAQs About Plagiarism
Can I plagiarise accidentally?
Yes, and most cases are exactly that. Intent affects the penalty at many institutions but not whether it counts as plagiarism. This is why the prevention has to sit in your note-taking habits rather than in care at the writing stage.
Do I need to cite common knowledge?
No, but the boundary is narrower than people assume. Common knowledge means facts an educated reader in your field would accept without a source. When in doubt, cite: an unnecessary citation costs nothing, and a missing one can cost a great deal.
How much can I quote before it is too much?
There is no numeric limit, but heavy quoting signals a different problem. Extended quotation usually means you are reporting the literature rather than analysing it. Quote when the exact wording matters; paraphrase when the idea matters.
Is it plagiarism if I cite the source but copy the words?
Yes, unless the copied words are in quotation marks. A citation tells the reader where the idea came from; quotation marks tell them the language is not yours. You need both.
Does a proofreader or editor count as plagiarism?
Language editing is normally permitted and often needs declaring. Substantive rewriting that changes your argument is a different matter. Check your institution's policy on third-party editing, and keep a record of what was changed.
Can I reuse my own literature review from a previous assignment?
Usually yes, with disclosure and a citation to the earlier work. Silently reusing it is self-plagiarism. Ask your supervisor in writing and keep the reply.
Do AI detectors prove plagiarism?
No. They are unreliable in both directions, producing false positives on non-native English writing in particular. A detector result is not evidence of misconduct, and your drafts and version history are far stronger evidence of your own authorship.
The formatting rules for attribution take an afternoon to learn. The habit that actually prevents plagiarism takes about a week to establish and then runs by itself: three categories of note, a page number recorded on the spot, and nothing pasted into a draft that is not clearly marked as someone else's. Build that once, and the question stops coming up. Our complete citation guide covers the mechanics once the attribution itself is sound.