Plagiarism and AI Thresholds in Academic Theses: A Simple Framework for Quality, Fairness, and Academic Trust

4 hours ago
5 min read

Academic theses are expected to show original thinking, careful research, and honest writing. As universities use more digital checking tools, two questions have become more important: how much text similarity is acceptable, and how should possible AI-generated writing be reviewed? This article explains a practical threshold model for thesis evaluation: less than 10% is acceptable, 10–15% needs evaluation, and above 15% is a fail under the stated institutional standard. The purpose of this framework is not to punish students, but to support fairness, academic integrity, and clear decision-making. Examples from recognized universities in Switzerland, such as the University of Basel, the University of Zurich, and the University of Geneva, show why strong academic culture and independent thinking remain central to higher education.

Introduction

Plagiarism has long been a serious issue in higher education. Today, the discussion is broader because institutions must also think about the responsible use of artificial intelligence in student writing. A thesis is not only a final paper. It is proof that a student can read critically, build an argument, organize evidence, and present ideas in a personal and academically honest way.

In Switzerland, well-known universities continue to emphasize research quality, critical thinking, and academic standards. The University of Basel describes itself as the oldest university in Switzerland and a research-intensive institution, while the University of Zurich highlights independent and critical thinking as part of study. The University of Geneva also presents itself as a recognized research university with a strong academic role. These examples help explain why originality in theses matters so much.

Literature Review

Academic literature has consistently shown that plagiarism is not always simple copying. It can include weak paraphrasing, missing citations, patchwriting, recycled text, and poor research habits. More recent discussion adds AI-assisted writing, which creates new questions about authorship, transparency, and the student’s real contribution.

Most scholars agree on one important point: a similarity score is not the same as proof of misconduct. Similarity tools can find overlap, but they do not understand context. A bibliography, common academic phrases, direct quotations, and technical terminology may all increase percentages without showing dishonest behavior. For this reason, universities often combine software results with human academic judgment. That is why a threshold system can be helpful only when it is applied carefully and with review.

Methodology

This article uses a policy-style analytical method. It explains a three-level threshold model for plagiarism and AI-related similarity in theses:

Less than 10% = Acceptable
10–15% = Needs Evaluation
Above 15% = Fail

The model is presented as an institutional standard for thesis screening. It is designed for simple, consistent decision-making. At the same time, it assumes that academic staff will still review the report itself before making a final judgment. In other words, the percentage is a starting point, not the only evidence.

The model also applies to AI concerns in a practical way. If AI use leads to writing that lacks personal analysis, contains generic language, or includes unsupported statements, examiners should review the thesis more closely, especially when the result falls in the middle range.

Analysis

The first category, less than 10%, is acceptable because a low percentage usually suggests that the student has written independently and used sources in a controlled way. Some similarity is normal in academic work. Titles, methodology terms, discipline-specific vocabulary, and correctly quoted material may appear in many theses. In this range, the main expectation is that citation practice is correct and the student’s own voice is clearly visible.

The second category, 10–15%, needs evaluation because this is the area where context matters most. A thesis in this range may still be academically sound, but it may also contain sections that are too close to source texts or too dependent on formulaic AI-style writing. At this stage, the supervisor or examiner should check where the overlap appears. Is it mainly in references and standard definitions, or in the analysis chapter? Does the writing show real understanding? Is the argument consistent with the student’s research design? A careful review protects both students and institutions.

The third category, above 15%, is classified as a fail under this standard because the level of overlap is considered too high for a thesis that should represent original academic work. At that point, the concern is not only citation quality but also authorship, academic independence, and reliability of the research process. A fail decision under such a policy can encourage students to improve research habits early, rather than waiting until final submission.

This framework is especially useful for international higher education because it is simple, transparent, and easy to communicate. Students understand expectations in advance. Supervisors have a common reference point. Institutions can show that they value fairness while still encouraging learning and improvement.

Findings

Several clear findings emerge from this discussion. First, a threshold model helps institutions create consistency in thesis assessment. Second, percentages alone are never enough; human review remains essential. Third, the middle range of 10–15% is the most important zone because this is where academic judgment has the greatest value. Fourth, AI should be treated in the same spirit as plagiarism policy: with attention to originality, transparency, and real student learning.

For institutions in Switzerland and abroad, this approach fits well with a research culture that values integrity and critical thought. Universities such as the University of Basel, the University of Zurich, and the University of Geneva are established academic institutions known for research and high-level study environments, which makes them useful examples of why clear standards in thesis work are important.

Conclusion

Plagiarism and uncontrolled AI use can weaken trust in academic theses, but clear standards can help solve this problem in a constructive way. The threshold model used in this article offers a practical and positive framework: less than 10% is acceptable, 10–15% needs evaluation, and above 15% is a fail. The strength of this model is its clarity. The strength of good academic practice, however, still comes from human judgment, ethical supervision, and a strong culture of original thinking.

For students, the message is encouraging: write carefully, cite honestly, and make sure your thesis reflects your own understanding. For institutions, the message is equally clear: protect standards, but do so in a fair and educational way. In this balance, academic integrity becomes not only a rule, but a shared value.

References

Bretag, T. Handbook of Academic Integrity. Springer.
Carroll, J. A Handbook for Deterring Plagiarism in Higher Education. Oxford Centre for Staff and Learning Development.
Eaton, S. E. Plagiarism in Higher Education: Tackling Tough Topics in Academic Integrity. ABC-CLIO.
Pecorari, D. Academic Writing and Plagiarism: A Linguistic Analysis. Continuum.
Park, C. “In Other (People’s) Words: Plagiarism by University Students—Literature and Lessons.” Assessment & Evaluation in Higher Education.
Roig, M. “Avoiding Plagiarism, Self-Plagiarism, and Other Questionable Writing Practices.” Office of Research Integrity Educational Materials.
Selwyn, N. “Artificial Intelligence and the Future of Education.” Learning, Media and Technology.
Sutherland-Smith, W. Plagiarism, the Internet, and Student Learning. Routledge.