Most academics are familiar with plagiarism detection software. Students submit assignments through it. Journals use it to screen manuscripts. Universities use it to investigate allegations of academic misconduct. In many institutions, plagiarism checking has become a routine part of academic life.
The purpose appears straightforward. Compare a document against a large database of published content and identify similarities that may require further investigation.
At least, that was the original intention. Today, I wonder whether plagiarism detection software is increasingly being used for a very different purpose. Not to identify plagiarism. But to remove evidence of it. And, perhaps, that is how it is being predominantly use these days?
Why These Tools Were Created
The development of plagiarism detection software was a logical response to a genuine problem. As digital content became more accessible, it became easier for students and researchers to copy material from books, articles, websites and other sources. Manually checking every submission became impossible.
Technology offered a solution. Software could compare a submitted document against millions of sources and produce a similarity report. This could be reviewed to determine whether the similarities were legitimate or whether they represented plagiarism. Importantly, these systems were never designed to make decisions. They were designed to assist human judgement.
A similarity score is not proof of misconduct. It is simply an indicator that further investigation may be required. That distinction remains important.
The Emergence of Similarity Scores
Over time, something interesting happened. The conversation shifted from plagiarism to similarity. Rather than asking whether a piece of work was original, attention increasingly focused on a percentage. What is the similarity score? Is it below 20%, 15%, 10% … or whatever similarity figure you set as your benchmark?
Many academics will have had discussions that revolve around these numbers. In some cases, students become more concerned about their similarity score than about whether they have properly acknowledged the work of others. In fact, some journals, even reputable ones, state that the paper being submitted should be below a certain plagiarism threshold. This ignores the fact that a paper with a 5% similarity score could be plagiarised and a paper with a 30% similarity score may not be. Moreover, it provides a target for somebody who is operating in an unethical way.
The percentage becomes the objective. And once a metric becomes the objective, behaviour changes and, more often than not, it is no longer a good metric.
A Different Way of Using the Software
Consider a student who has copied sections of text from various sources. Years ago, they may have submitted the work and hoped that nobody noticed. Today, they can upload the document to a plagiarism checker before submission. If the similarity score is too high, they can modify the text, run the check again, and repeat the process until the score falls below an acceptable threshold.
The objective is no longer to produce original work. The objective becomes reducing the similarity score. The software is still functioning exactly as designed. It is accurately identifying similarities. However, the way the software is being used is different to the purpose for which it was originally designed.
Now, instead of helping identify plagiarism, it can become part of a workflow designed to disguise it.
The Growth of Commercial Services
This phenomenon extends beyond individual students. Search the internet it is not hard to find many commercial services that advertise similarity reduction, plagiarism removal or assistance with producing content that can pass plagiarism checks. Some services explicitly promise low similarity scores.
Others offer rewriting services that claim to preserve meaning while altering wording sufficiently to avoid detection. The marketing itself is revealing. These businesses are not promoting originality. They are promoting successful navigation of a compliance system.
In effect, a market has emerged around avoiding detection. This should concern us all.
The Problem with Thresholds
Part of the challenge may lie in our reliance on numerical thresholds. Many institutions have informal expectations regarding acceptable similarity levels. The exact figures vary, but the principle is often the same. Below a certain number is considered acceptable. Above a certain number may trigger concern.
The difficulty is that similarity and plagiarism are not the same thing. A paper could have a low similarity score and still contain unethical appropriation of ideas. Conversely, a document could have a relatively high similarity score for entirely legitimate reasons.
Reference lists, discipline standard phrasing, technical terminology, biographies and properly quoted material can all increase similarity percentages without representing misconduct.
A similarity score, presented as a single value, may give a feeling of comfort, but it hides a lot of complexity. The number may be easy to understand, but it does not necessarily capture what we actually care about.
A Familiar Pattern
The more I think about this issue, the more it resembles many of the other challenges facing research and higher education. Impact factors were designed to help evaluate journals. They are now frequently used as targets.
Citation counts were intended to measure influence. They are now actively optimised.
University rankings were created to provide comparative information. Institutions now invest considerable effort in improving ranking performance.
The same pattern appears repeatedly. A measure is introduced to assess quality. The measure becomes important. People begin optimising for the measure. Eventually, the measure becomes disconnected from the thing it was supposed to represent.
Perhaps plagiarism detection software is experiencing the same phenomenon?
What Should We Really Be Measuring?
The fundamental question is simple. What is the objective? If the objective is to achieve a low similarity score, then plagiarism detection software is performing very well.
If the objective is to encourage original thinking, proper attribution and academic integrity, the answer is less clear.
The challenge is that originality cannot be reduced to a percentage. Integrity cannot be captured by a numerical threshold. And genuine scholarship involves far more than ensuring a document falls below an arbitrary similarity score.
The danger is that we focus on what is easy to measure rather than what actually matters.
Final Thoughts
I am not arguing that plagiarism detection software is ineffective. Nor am I suggesting that universities, publishers and journals should stop using it. These tools remain valuable and often identify genuine cases of misconduct.
However, we should recognise that any compliance system can eventually become something that people learn to game. The question is no longer whether plagiarism detection software works. The more interesting question may be whether it is increasingly being used for the opposite purpose to the one for which it was originally developed.
Have plagiarism detectors become plagiarism enablers? I suspect the answer is more complicated than many of us would like to admit.
About the Author
Professor Graham Kendall is the Vice Chancellor at GlobalNxt University, Malaysia, an Emeritus Professor at the University of Nottingham and a Visiting Professor at Hong Kong Metropolitan University. He has published more than 300 peer-reviewed papers and has spent over 25 years researching and writing within academia. Through his Publishing with Integrity initiative, he explores issues relating to research integrity, publication ethics, research evaluation and the responsible use of AI in scholarly communication.