Plagiarism is presenting someone else's words, ideas, data or structure as your own without giving them credit. It includes copying text word for word, patchwriting (swapping a few words in a source sentence), paraphrasing without a citation and, under many institutions' rules, reusing your own earlier work without saying so. Quoting with quotation marks and a citation, or genuinely restating an idea in your own words and citing where it came from, is not plagiarism.
- Credit is what matters: copied words need quotation marks and a citation; borrowed ideas need a citation.
- Changing a few words in someone else's sentence (patchwriting) is still plagiarism, even with a citation, because the expression is still theirs.
- Common knowledge, such as widely known dates and facts, does not need a citation; specific findings, figures and arguments do.
- Rules on AI-written text differ between schools, journals and employers, so check the policy that applies to you.
- A similarity percentage measures matching text, not guilt; a person has to look at what matched and why.
What plagiarism actually means
Most academic and publishing definitions share the same core: using another person's work without acknowledging it, so that a reader would reasonably assume it is yours. "Work" covers sentences and phrases, but also ideas, arguments, findings, data, images and code.
Intent is not required. Many cases come from careless note-taking: a passage copied into notes, its source forgotten, later pasted into an essay as the writer's own summary. Institutions often treat accidental cases more leniently, but both count.
Plagiarism is not the same as copyright infringement. Copyright is a legal right over the expression of a work; plagiarism is an academic and ethical rule about attribution. You can plagiarise a public-domain text, and you can infringe copyright while crediting the author, for example by republishing a whole article without permission.
The main types of plagiarism
| Type | What it looks like | How to avoid it |
|---|---|---|
| Direct copying | Sentences or paragraphs copied word for word without quotation marks or a citation. | Put copied words in quotation marks and cite the source, or rewrite the idea in your own words and cite it. |
| Mosaic plagiarism (patchwriting) | Source sentences kept in the same order and structure, with some words replaced by synonyms or phrases rearranged. | Close the source, write from your understanding, then compare and cite. |
| Paraphrasing without credit | Someone else's idea, argument or finding restated in genuinely new words, but with no citation. | Cite every source whose ideas you use, even when none of the wording is theirs. |
| Self-plagiarism | Submitting or publishing work you already submitted or published elsewhere, without disclosure or permission. | Ask before reusing, cite your earlier work, and follow the rules on duplicate publication. |
| Undisclosed outside help | Text written by another person or an AI tool presented as entirely your own where the rules require disclosure. | Read the applicable policy and disclose assistance as it requires. |
Self-plagiarism
Reusing your own words can feel harmless, but the reader, teacher or editor is told the work is new when it is not. Resubmitting an essay from another course usually breaks the rules unless both instructors agree. In research, duplicate publication can make one result look like two, and copyright in an earlier article may belong to its publisher. Citing your earlier work and reusing only what is necessary, such as a standard methods description, is the usual solution, but journals differ on what they accept.
Ideas, expression and common knowledge
Ideas need credit, not only words. If a specific argument, interpretation or result came from a source, cite it, however thoroughly you have reworded it.
Common knowledge does not need a citation. Facts that are widely known, easy to verify and undisputed can be stated without a reference, such as that the Second World War ended in 1945 or that Paris is the capital of France. What counts depends on the audience: a fact every economist knows may still need a citation for general readers. Specific numbers, recent findings and anything reasonably disputed should be cited. When in doubt, cite.
How to quote, paraphrase and cite correctly
The passage below was written for this guide and attributed to a fictional source, "Lee (2021)", to show the citation format.
The original passage
Cities that plant trees along their streets often see lower summer temperatures at ground level, because the canopy blocks direct sunlight and water evaporating from the leaves cools the surrounding air. The effect is strongest in dense neighbourhoods where buildings and paving absorb the most heat.
Patchwritten version (plagiarism)
Cities that put trees along their roads frequently experience reduced summer temperatures at street level, since the canopy stops direct sunlight and water evaporating from leaves cools the nearby air. The effect is greatest in crowded areas where buildings and pavements soak up the most heat (Lee, 2021).
This has a citation, but it is still plagiarism: the structure, the order of ideas and most of the phrasing are Lee's, with only individual words swapped.
Acceptable paraphrase with citation
Street trees can make city neighbourhoods noticeably cooler in summer. Their shade keeps sunlight off the ground, and the moisture they release through their leaves lowers the air temperature, with the biggest benefit in tightly built areas full of heat-absorbing surfaces (Lee, 2021).
The idea is credited, and the sentences are new: different structure, emphasis and vocabulary.
Direct quotation
Lee (2021) notes that the cooling effect "is strongest in dense neighbourhoods where buildings and paving absorb the most heat".
Quote when the exact wording matters, such as a definition or a claim you want to analyse, and keep quotations short.
A reliable paraphrasing method
- Read the passage until you understand it well enough to explain it to someone else.
- Close or hide the source.
- Write the idea from memory, in your own sentence structure, as you would naturally say it.
- Compare your version with the original. If whole phrases or the same sentence pattern survived, rewrite or switch to a quotation.
- Add the citation in the style your course or publisher requires, such as APA, MLA, Chicago, Harvard or a numbered style.
AI-generated text and disclosure
Text produced by a chatbot raises a different question from classic plagiarism. No human author's words were taken, yet submitting it as entirely your own can still misrepresent who did the work. AI tools can also reproduce phrasing from their training data and invent sources, which creates ordinary plagiarism and citation problems on top.
There is no single rule. Universities, courses, journals, publishers and employers have adopted different policies, from outright bans, through use permitted with disclosure, to use allowed for specific tasks such as grammar checking. Some ask you to name the tool and describe how you used it. Read the policy that actually applies to you, and if it is unclear, ask in writing.
If you are reviewing someone else's work, remember that AI detectors estimate how similar a text is to machine-generated writing; they are not proof. Our AI detector runs in the browser using an open RoBERTa model trained on the RAID benchmark, with a second model for borderline scores, and like every detector it can be wrong. Use its result as a reason to have a conversation, not as evidence on its own.
How plagiarism checkers work
A plagiarism checker does not understand authorship. It finds text that matches or closely resembles text in other documents, and tools differ mainly in what they compare against and how they define "resembles".
What they compare against
- Web and database search. Commercial services compare a submission against indexed web pages, licensed articles and books and, in some services, previously submitted student papers. This finds sources you did not know about, but only if they are in the index.
- Sources you provide. Other tools compare your text only against documents you give them, which is useful when you know the candidate sources, such as an essay's reading list or the sources in your own notes.
Shingling and fingerprints
To find copied passages quickly, checkers break text into overlapping sequences of words called shingles or n-grams. The sentence "street trees cool city air" produces the 3-word shingles "street trees cool", "trees cool city" and "cool city air". Each shingle is normalised (lower case, no punctuation) and turned into a number with a hash function, and a subset of those hashes becomes the document's fingerprint; winnowing, described by Schleimer, Wilkerson and Aiken in 2003, is a well-known way to choose them. Documents that share many fingerprint hashes share runs of identical wording.
Semantic similarity for paraphrase
Exact shingles miss patchwriting, because swapping synonyms breaks the word sequences. Newer checkers also convert each sentence into a sentence embedding, a list of numbers produced by a language model that represents its meaning. Sentences with similar meaning end up close together, measured with cosine similarity, even when they share few words. This catches reworded passages but also flags sentences that make the same point independently, so its results need more human judgement.
Our plagiarism checker combines both approaches against sources you paste in: it highlights exact matches of five or more consecutive words and uses sentence embeddings to flag likely paraphrases. It does not search the web, and the comparison runs in your browser.
What a similarity percentage means
The headline number in a plagiarism report is the share of your text that matched something in the comparison set. It is a measure of overlap, not a finding of plagiarism.
- Matches can be legitimate. Properly quoted and cited passages, a reference list, standard phrases ("the results suggest that"), technical terms, names of laws or organisations, and a required assignment template all produce matches.
- A low score can hide a problem. Copying from a source that is not in the comparison set, or translating a source from another language, may produce no match at all. Unattributed ideas in genuinely new words also will not show up.
- There is no universal acceptable percentage. What matters is what matched: a few scattered common phrases are normal, while one continuous uncited paragraph is a problem even if it is only a small share of a long document.
- Settings change the number. Excluding quotations, the bibliography or short matches can move the score a lot.
How to check your own work before submitting
- Make sure every quotation has quotation marks and a citation, and every borrowed idea is cited.
- Compare paraphrases side by side with the source; rewrite any that follow its sentence pattern.
- Run the draft through a plagiarism checker with your main sources and read every highlighted passage, not just the percentage.
- Check that every reference exists and says what you claim, especially if an AI tool was involved.
- Disclose AI assistance or reused earlier work if your rules require it.
Frequently asked questions
Is paraphrasing plagiarism?
Paraphrasing is not plagiarism when the wording and structure are genuinely yours and you cite the source of the idea. It becomes plagiarism if you leave out the citation, or if your "paraphrase" keeps the original's structure and most of its phrasing with a few words changed.
How many words in a row count as plagiarism?
There is no official number. Checkers use thresholds such as five or more matching words to decide what to highlight, but plagiarism is about whether distinctive wording or ideas were used without credit. A single distinctive phrase can need quotation marks, while a long common phrase may not.
Can I be accused of plagiarism for using ChatGPT?
It depends on the rules you are working under. Some institutions treat undisclosed AI use as misconduct, while others allow it with disclosure or for limited tasks. Read the policy that applies to you.
What similarity score is acceptable?
No score is automatically acceptable or unacceptable. Reviewers look at what matched. A higher score made up of cited quotations and references can be perfectly fine, while a low score containing one copied, uncited paragraph is not.