Plagiarism and AI Thresholds in Academic Theses: A Practical Framework for Quality, Integrity, and Fair Evaluation

Academic theses are expected to show originality, critical thinking, and ethical scholarship. As digital tools and artificial intelligence become more common in higher education, universities are paying closer attention to plagiarism detection and AI-generated content in student work. This paper proposes a simple evaluation framework for academic theses based on three thresholds: less than 10% similarity as acceptable, 10–15% as needing evaluation, and above 15% as fail. The article argues that similarity reports should not be treated as automatic proof of misconduct, but as tools for academic judgment. It also discusses the need to assess AI use carefully, especially in research writing, editing, and literature support. Examples from international universities and universities in Kenya show that institutions are increasingly combining software-based checks with human review. The paper concludes that clear thresholds, combined with academic evaluation, can improve fairness, strengthen research quality, and support students in producing honest and credible theses.

Introduction

Academic integrity remains one of the most important pillars of higher education. A thesis is not only a final academic document; it is also evidence of a student’s ability to think independently, engage with literature, and contribute responsibly to knowledge. In recent years, plagiarism detection software and AI-writing tools have changed the way universities monitor academic work.

This change has created both opportunities and concerns. On one hand, technology helps institutions detect copied material, improve referencing, and protect academic standards. On the other hand, similarity scores can be misunderstood, and AI-generated text may appear original while still weakening the student’s own intellectual contribution.

This article presents a practical standard for thesis evaluation: Less than 10% = Acceptable, 10–15% = Needs Evaluation, Above 15% = Fail. The purpose is not to punish students, but to create a fair and transparent structure that encourages good academic practice.

Literature Review

Plagiarism is broadly understood as the use of another person’s words, ideas, data, or structure without proper acknowledgment. Academic writing scholars have long argued that plagiarism is not always caused by dishonesty alone. In some cases, it results from weak research skills, poor paraphrasing, language limitations, or limited understanding of citation rules.

Similarity detection software has therefore become widely used in thesis examination. However, scholars also note that similarity percentages do not automatically equal plagiarism. A high score may come from quotations, references, standard terminology, or repeated methodological phrases. For this reason, interpretation by qualified supervisors and examiners is essential.

The rise of artificial intelligence adds a new dimension. AI tools may assist with grammar, language clarity, summarization, and idea organization. Yet if students depend too heavily on such systems, the originality of the thesis may be reduced. International universities are increasingly emphasizing that AI should support learning, not replace authorship. In Kenya as well, the conversation around responsible AI use in research is growing, particularly within graduate education and thesis preparation.

Methodology

This article uses a qualitative policy-oriented approach. It reviews academic integrity principles from higher education practice and applies them to thesis evaluation. It also considers emerging institutional approaches in both international and Kenyan university contexts, where plagiarism checks and AI-related guidance are becoming more visible.

The proposed threshold model is designed as a practical framework for thesis review:

• Less than 10% = Acceptable

• 10–15% = Needs Evaluation

• Above 15% = Fail

The model assumes that plagiarism software is a screening tool, not a final judge. Human academic review remains necessary in all cases.

AnalysisThe first threshold, less than 10%, is suitable for most well-prepared theses. At this level, the work is generally likely to reflect proper paraphrasing, correct citation, and independent writing. Minor overlap may still exist in titles, technical phrases, methodology descriptions, or reference sections, but this does not normally threaten the integrity of the thesis.

The second threshold, 10–15%, requires careful academic evaluation. This range should not lead to automatic failure. Instead, the examiner should ask several questions: Where does the overlap appear? Is it concentrated in the literature review, methodology, or conclusion? Is the text properly cited? Does the student demonstrate original analysis? A thesis in this category may still be acceptable after revision if the overlap is not deceptive and can be corrected.

The third threshold, above 15%, signals serious concern and should normally lead to failure or major resubmission. At this level, the overlap may suggest excessive dependence on copied material, weak paraphrasing, or overuse of AI-generated content. The academic concern is not only percentage size, but also whether the student’s own voice and research contribution are visible.

AI thresholds should be treated with similar caution. If AI is used only for language polishing or formatting support and is openly disclosed, the academic risk may be low. But if a student uses AI to generate large sections of the thesis, the work may appear polished while lacking real authorship. In such cases, examiners should assess coherence, depth, source accuracy, argument development, and oral defense performance.

Examples from international universities show that academic integrity policies now often include AI-generated material within the broader discussion of plagiarism and unauthorized assistance. In Kenya, universities are also strengthening thesis writing guidance, anti-plagiarism procedures, and responsible research standards. These developments suggest that a combined model of software screening and expert academic judgment is both timely and useful.

Findings

This article identifies five main findings. First, a fixed threshold system helps universities communicate expectations clearly. Second, similarity reports are most useful when interpreted by trained academics rather than treated mechanically. Third, the middle category of 10–15% is important because it allows fairness and context-based review. Fourth, AI use should be evaluated not only by quantity but also by purpose, disclosure, and impact on authorship. Fifth, institutions that combine clear policy, student training, and human evaluation are more likely to protect both academic quality and student confidence.

Conclusion

Plagiarism and AI thresholds in academic theses should be managed through clarity, fairness, and educational purpose. A simple three-part framework—less than 10% acceptable, 10–15% needs evaluation, and above 15% fail—offers a practical model for universities, supervisors, and examiners. Such a framework helps students understand expectations while protecting the value of original research.

In today’s academic environment, integrity is not only about avoiding misconduct. It is also about building a culture of responsible writing, transparent scholarship, and thoughtful use of new technologies. When universities apply clear standards with human judgment, they support both academic excellence and student development in a positive and constructive way.

References

• Bretag, T. (2016). Handbook of Academic Integrity. Singapore: Springer.

• Carroll, J. (2007). A Handbook for Deterring Plagiarism in Higher Education. Oxford: Oxford Centre for Staff and Learning Development.

• Eaton, S. E. (2022). Plagiarism in Higher Education: Tackling Tough Topics in Academic Integrity. Santa Barbara: Libraries Unlimited.

• Pecorari, D. (2013). Teaching to Avoid Plagiarism: How to Promote Good Source Use. Maidenhead: Open University Press.

• Sutherland-Smith, W. (2008). Plagiarism, the Internet and Student Learning: Improving Academic Integrity. New York: Routledge.

• Foltynek, T., Meuschke, N., & Gipp, B. (2020). Academic plagiarism detection: A systematic literature review. ACM Computing Surveys, 52(6), 1–42.

• Perkins, M. (2023). Academic integrity considerations of AI large language models in higher education. Journal of University Teaching and Learning Practice, 20(5), 1–17.

Hashtags #AcademicIntegrity #ThesisWriting #PlagiarismPolicy #ResponsibleAI #HigherEducation #ResearchQuality