Of the ten categories in Turnitin’s widely cited Plagiarism Spectrum, the first is also the bluntest: the clone, defined as “submitting another’s work, word-for-word, as one’s own” (Turnitin, 2015). It is the form of textual misappropriation that most readers picture when they hear the word plagiarism – a passage lifted intact, perhaps with the author’s name swapped at the top, perhaps with nothing changed at all.
It is also, by some distance, the form of plagiarism most easily caught. Modern text-matching systems were essentially designed to detect it; a competent reader can often spot it without any software at all. And yet, after more than two decades of widely deployed detection tools, near-verbatim copying remains stubbornly common in student assignments, in commercial web content, in self-published books and even, on occasion, in the speeches of heads of state.
This article asks why. It examines what clone plagiarism actually looks like in practice, why the technology that detects it works so reliably, and why – despite that reliability being widely understood – the behaviour persists across academic, journalistic and commercial writing.
What clone plagiarism is, precisely
The Plagiarism Spectrum places clone copying at severity level 9.5 out of 10 and notes that, in a survey of nearly 900 educators, it ranked among the most frequently encountered forms of unoriginal work. The defining characteristic is the absence of meaningful alteration: text is reproduced from a single identifiable source, without quotation marks, without citation, and without any attempt to translate the ideas into the writer’s own register.
It is worth distinguishing the clone from its near neighbours. The closely related “CTRL-C” category covers passages that are largely verbatim with minor changes – a deleted adverb here, a substituted conjunction there. “Find–Replace” plagiarism describes the slightly more strenuous practice of swapping individual words or phrases while leaving sentence architecture intact. Clone plagiarism sits at the unobscured end of this continuum: the source passage and the submitted passage are, for practical purposes, the same text.
In an empirical study of more than 1,000 university scripts using Turnitin, John Walker (2010) found that verbatim copying (which he defined as “copying material verbatim without citing the source”) was one of three dominant forms of plagiarism in business assignments, alongside what he called “sham paraphrasing” and “purloining” from classmates. Walker’s contribution was significant because his data measured what students actually did, not what they said they did – a distinction that, as we shall see, materially changes the picture.
Why detection is essentially trivial
To understand why clone plagiarism is the easiest variety to flag, it helps to know how text-matching software actually works. The underlying technique used by Turnitin, MOSS, Dolos and most of their peers is document fingerprinting – a method first formalised in the winnowing algorithm described by Schleimer, Wilkerson and Aiken at Stanford.
The procedure can be summarised in four steps:
- Tokenisation. The submitted document is normalised — whitespace collapsed, capitalisation flattened, punctuation stripped — to produce a clean stream of tokens.
- k-gram generation. The token stream is broken into overlapping subsequences of fixed length k (typically between 4 and 8 words for natural-language text). For a document of n tokens, this yields n−k+1 k-grams.
- Hashing and selection. Each k-gram is hashed, usually with a rolling hash function of the kind introduced by Rabin and Karp. The winnowing step then selects, from each sliding window of w consecutive hashes, the smallest value. These survivors form the document’s fingerprint.
- Indexing and matching. Fingerprints are compared against an index built from a reference corpus — for Turnitin, this includes billions of web pages, a vast archive of previously submitted student work, and licensed publisher content.
The mathematics of winnowing guarantee something important. If two documents share any contiguous run of at least w+k−1 tokens, the algorithm is guaranteed to detect it, irrespective of where in either document that run appears (Wikipedia, 2025). With typical parameters of k=5 and w=4, this means that any directly copied passage of roughly eight words or more will be flagged with certainty.
Eight words. That is an extraordinarily low threshold. A single moderately long sentence almost always exceeds it. A copied paragraph will produce dozens of overlapping fingerprint hits, each one independently traceable back to its source.
The implication is stark. For clone plagiarism – copying without alteration – the question is not really whether the software will detect it, but whether the source happens to be in the reference corpus. And given the scale of contemporary corpora (Turnitin’s database contains well over a billion student papers, supplemented by a current crawl of the open web), a source that is publicly accessible online or has previously been submitted at any participating institution is overwhelmingly likely to be there.
The student behaviour problem
If the technical case against clone copying is this clear-cut, the persistence of the behaviour becomes a genuine puzzle. The longitudinal data are useful here. In the only study to have administered the same plagiarism survey at the same institution every five years for two decades, Curtis and colleagues (2026) found a measurable downward trend in self-reported verbatim copying at Western Sydney University between 2004 and 2024 – but emphatically not its disappearance. Students reported plagiarising both naïvely (without recognising the behaviour as such) and knowingly, and the introduction of mandatory integrity training, honour codes and even generative AI detectors produced only modest deterrent effects.
Several recurring drivers emerge from the literature. The first is time pressure. The University of Pretoria’s academic integrity guidance lists “leaving writing until the last minute” as the single most common situational trigger for plagiarism, on the straightforward grounds that copying is faster than composing. Survey data from a Nigerian study by Babalola (2012) corroborate this: 59.7% of student respondents reported plagiarising specifically “in order to meet deadlines for assignments.”
The second is misjudged risk. A consistent finding across the literature, summarised by Park (2003) and reinforced by Babalola’s data, is that students who plagiarise tend to perceive that “there is little or no chance of getting caught and there is little or no punishment if they are indeed caught.” This perception is, in the case of clone copying, almost wholly mistaken – but it persists, and it persists asymmetrically. Walker’s empirical study of Turnitin reports found that students who were caught and confronted often did not modify their behaviour on the next assignment, suggesting that the deterrent effect of detection only materialises when consequences are visible and consistently applied.
The third driver is genuinely uneven understanding of what constitutes plagiarism. In a survey of 965 STEM undergraduates, Vieyra and Weaver (2023) found that the majority of first-year students admitted to some form of plagiarism and that recognition of specific types remained patchy even into the senior years. Crucially – and this is the part most relevant to clone copying specifically – students were generally good at identifying word-for-word copying as plagiarism in the abstract, but considerably worse at identifying it in scenario-based examples involving electronic sources. The intellectual category was understood; the behavioural application was not.
A fourth factor, identified by Jereb et al. (2018) in their comparative study of German and Slovene students, is the perceived ontological status of online text. Material on the open web is, for many students, psychologically closer to the public domain than to copyrighted scholarship — the belief, as Koul and colleagues put it, “that we as people have greater ownership of information than we have paid for.” Combined with the frictionless mechanics of select-copy-paste, this produces precisely the cut-and-paste behaviour that fingerprinting algorithms are best at catching.
The web writer’s version of the same problem
Clone plagiarism is not a uniquely academic phenomenon. It pervades commercial content production, where the incentive structure is different but the underlying behaviour is recognisable.
Google’s spam policies are explicit on this point. The company’s documentation describes scraping as “republishing content from other sites without adding any original content or value, or even citing the original source,” and lists it among the practices that may cause a site to “rank lower in results or not appear in results at all” (Google Search Central, 2025). The same policy targets “scaled content abuse,” which captures the practice of generating large volumes of unoriginal pages – increasingly, these days, with synonym-substitution or AI rewriting layered over scraped source text.
It is important to draw a careful distinction here. Google has, since at least Susan Moskwa’s 2008 clarification on the company’s developer blog, explicitly denied that there is a generic “duplicate content penalty” for, say, e-commerce sites that produce parameterised URLs of the same product page. The harm from clone-style copying on the open web is more specific: when Google’s systems detect that one page is a derivative copy of another, they cluster the duplicates and select a canonical version to surface in search results. The original source – the one with established crawl history, inbound links and topical authority – is heavily favoured. The copy receives, in practice, no traffic. From a commercial standpoint, the copied page is dead on arrival.
This rarely deters the behaviour entirely. Affiliate marketers, low-margin content farms and SEO operators continue to scrape, and a recurring justification echoes the student literature: the perceived probability of consequence is low, the marginal cost of producing the page is near zero, and even a small fraction of unsuppressed copies producing some traffic can be commercially worthwhile at sufficient scale. The economics of the long tail subsidise behaviour that, examined individually, looks irrational.
When the stakes are highest, the behaviour persists
Perhaps the most striking demonstration that clone plagiarism is rarely a matter of pure rational calculation is the regularity with which it surfaces in high-profile journalism and politics – contexts in which the costs of detection are catastrophic and the writers involved have every reason to know better.
The cases form an unflattering catalogue. In 1987, Joseph Biden’s presidential campaign collapsed after passages of his stump speech were shown to have been lifted, almost word for word, from a speech delivered earlier that year by the Labour leader Neil Kinnock – alongside the discovery of an earlier law-school incident in which Biden had reproduced five pages of a published law review article (NPR, 2016). In 2012, the journalist Fareed Zakaria was suspended by both Time magazine and CNN after admitting that paragraphs of his column on gun control closely tracked an essay by the historian Jill Lepore in The New Yorker; the side-by-side comparison published by The New York Times shows the kind of structural and lexical proximity that fingerprinting algorithms detect almost instantly. Senator John Walsh of Montana had his Army War College degree rescinded in 2014 after substantial portions of his final paper were found to have been copied. President George W. Bush’s 2010 memoir Decision Points was shown to contain passages lifted “almost verbatim and without attribution” from work by the journalist Ahmed Rashid, among others (BBC, 2010).
What is instructive about these cases is not the moral failing – that is unsurprising – but the implausibility of the cost-benefit calculation that produced the behaviour. Each of these writers operated in an environment in which a single competent reader was sufficient to surface the copy, in which the stakes of exposure were career-defining, and in which the source material was prominent enough that its provenance was discoverable by anyone with a search engine. The clone happened anyway.
Why people still do it
Synthesising the evidence, four explanations recur, and they apply with surprising consistency across student, commercial and professional writing.
The asymmetry between effort and reward feels too large to resist. Composing original prose is cognitively expensive; copying is cognitively trivial. When a deadline is imminent, when the writer believes the audience will not read carefully, or when the marginal value of the specific passage feels low, the temptation is real. This is not a moral observation but a behavioural one – and it is consistent with the finding by Songsriwittaya and colleagues, and others, that plagiarism correlates with “performance goals” rather than mastery goals.
Detection probability is consistently underestimated. Even sophisticated writers seem to operate under an intuition that their particular copy will not be the one that is checked, even when the technology to check it is universal and the algorithm is, as we have seen, mathematically guaranteed to flag any contiguous match above a low word threshold. This is a familiar cognitive bias – the gap between abstract awareness of risk and concrete behavioural caution.
The category of “the public web” is psychologically permissive. Web text feels, to many users, less authored than book text. It is cited less consistently in lay writing, copied more freely in everyday communication, and appears in contexts (memes, social media reposts, content aggregators) where attribution is conventionally lax. This makes it disproportionately the source for clone copying, even though it is also disproportionately the corpus against which detection systems compare.
Confession and lack of consequence form a feedback loop. Where institutions detect clone plagiarism but apply only soft consequences – a warning, a request to resubmit, a modest grade reduction – the behavioural lesson taught is that the cost of being caught is lower than the cost of writing the assignment honestly. This is not an argument for harsher punishment; it is an observation that the deterrent value of a detection system depends on what happens after detection, not merely on detection itself.
What this means for institutions and for writers
For educators, the implication is that detection technology alone is insufficient. The Curtis et al. longitudinal data show that academic integrity training, honour codes and detection software each produce small downward pressure on plagiarism rates, but none does so dramatically. The most effective institutional interventions appear to be those that combine clear policy, consistent enforcement, instruction in source use (particularly note-taking and paraphrase technique), and assessment design that genuinely rewards original engagement with material.
For writers – including students, web professionals and journalists – the practical conclusion is more straightforward. Clone copying is the variety of plagiarism that detection systems are most equipped to catch and that human readers are most equipped to recognise. It offers the smallest gain in time saved relative to its risk, because it is precisely the form of misappropriation that the technical and social infrastructure has been optimised over decades to surface. The writers who have suffered the most public consequences for plagiarism – the politicians, the columnists, the bestselling authors – are not, for the most part, people who paraphrased subtly. They are people who copied directly.
That this still happens, two decades into the era of universal text matching, is the most revealing fact in the whole literature. It tells us that plagiarism is not, fundamentally, a problem that detection alone can solve — because the people doing it are not, in most cases, making a careful estimate of detection probability. They are responding to deadline pressure, to underestimated risk, to the easy availability of source passages, and to the residual sense that this particular copy, this once, will go unnoticed.
It almost never does.
References and further reading:
- Babalola, Y. T. (2012) ‘Awareness and incidence of plagiarism among undergraduates in a Nigerian private university’, African Journal of Library, Archives and Information Science, 22(1), pp. 53–60.
- BBC News (2010) Pakistani journalist upset by George Bush ‘plagiarism’. Available at: http://www.bbc.co.uk/news/world-south-asia-11761333(opens in new tab) (Accessed: 12 May 2026).
- Curtis, G. J., Slade, C., Bretag, T. and McNeill, M. (2026) ‘Results from two decades of five-yearly plagiarism surveys’, International Journal for Educational Integrity, 22(7).
- Google Search Central (2025) Spam policies for Google web search. Available at: https://support.google.com/webmasters/answer/66358(opens in new tab) (Accessed: 12 May 2026).
- Haughney, C. (2012) ‘CNN and Time suspend journalist after admission of plagiarism’, The New York Times, 10 August.
- Jereb, E., Perc, M., Lämmlein, B., Jerebic, J., Urh, M., Podbregar, I. and Šprajc, P. (2018) ‘Factors influencing plagiarism in higher education: A comparison of German and Slovene students’, PLOS ONE, 13(8), e0202252.
- Moskwa, S. (2008) Demystifying the “duplicate content penalty”. Google Search Central Blog, 12 September.
- NPR (2016) A brief history of politicians lifting words from other sources. National Public Radio, 19 July.
- Park, C. (2003) ‘In other (people’s) words: Plagiarism by university students — literature and lessons’, Assessment and Evaluation in Higher Education, 28(5), pp. 471–488.
- Schleimer, S., Wilkerson, D. S. and Aiken, A. (2003) ‘Winnowing: Local algorithms for document fingerprinting’, in Proceedings of the 2003 ACM SIGMOD International Conference on Management of Data, pp. 76–85.
- Turnitin (2015) The Plagiarism Spectrum: Tagging ten types of unoriginal work. Turnitin LLC.
- Vieyra, M. L. and Weaver, K. D. (2023) ‘Exploring factors contributing to plagiarism as students enter STEM higher education classrooms’, Issues in Science and Technology Librarianship, 102.
- Walker, J. (2010) ‘Measuring plagiarism: Researching what students do, not what they say they do’, Studies in Higher Education, 35(1), pp. 41–59.