Plagiarism and AI Thresholds in Academic Theses: Towards a Clear Standard

This article examines plagiarism detection and AI-assisted writing in academic theses, focusing on how universities worldwide handle similarity reports. It introduces a proposed standard: less than 10% similarity = Acceptable, 10–15% = Needs Evaluation, and above 15% = Fail. Drawing from real practices at leading universities in Europe, North America, Asia, and the Middle East, the article highlights the lack of explicit thresholds in many institutions, the challenges posed by artificial intelligence, and the need for clear and transparent rules. The findings show that while universities use plagiarism detection software extensively, policies often leave decisions to supervisors, creating inconsistencies. The article concludes by recommending the adoption of clear thresholds to ensure fairness and academic integrity.

Introduction

Universities across the globe share a common goal: to maintain academic integrity and ensure that research work reflects originality and intellectual honesty. Plagiarism, defined as using another person’s ideas or words without proper acknowledgment, undermines this goal. The growing use of plagiarism detection software, combined with the rise of AI-assisted writing tools, has created new challenges for academic institutions.

Most universities today use similarity detection tools such as Turnitin or iThenticate to evaluate theses before acceptance or publication. However, many institutions do not publish clear numerical thresholds, leaving students and supervisors uncertain about acceptable similarity levels. This article proposes a practical standard—less than 10% = Acceptable, 10–15% = Needs Evaluation, above 15% = Fail—and examines how it aligns with real practices in international higher education.

Literature Review

Previous research shows wide variations in plagiarism policies across countries and universities. Some studies report that similarity below 10% is often ignored because minor overlaps are unavoidable, especially in methodology sections or literature reviews. Similarity scores between 10–15% often trigger requests for explanation or revision, while scores above 15% frequently lead to more serious actions, including rejection or academic misconduct investigations.

However, very few universities publish these thresholds explicitly. For example:

• Several European universities require plagiarism checks for all theses but allow supervisors to interpret the reports rather than impose fixed percentages.

• In North America, many institutions treat similarity above 20% as serious but provide no universal standard.

• In Asia and the Middle East, practices vary: some universities demand less than 10% similarity, while others have no public thresholds at all.

Artificial intelligence adds another layer of complexity. AI tools often produce text that resembles existing content, potentially inflating similarity scores even when no intentional plagiarism occurs.

Methodology

This article draws upon three main sources:

1. Policy Documents – Academic regulations from universities in Switzerland, the United Kingdom, the United States, Australia, and Asia were reviewed for explicit plagiarism thresholds.

2. Similarity Report Analysis – Sample data from fifty master’s and doctoral theses across different disciplines were analyzed to see how similarity percentages were handled in practice.

3. Faculty Interviews – Academic staff at several universities provided insights into how supervisors interpret similarity scores and manage AI-assisted writing.

The proposed threshold standard (<10%, 10–15%, >15%) was applied to real thesis cases to test its practicality.

Analysis

The review produced three major observations:

1. Lack of Explicit Thresholds Many universities require plagiarism checks but leave interpretation entirely to supervisors. This leads to inconsistent decisions, especially in borderline cases. For example, a thesis with 12% similarity might be acceptable at one university but require revision at another.

2. Practical Use of Thresholds In reality, most theses with less than 10% similarity are accepted without major concern. Theses with 10–15% often receive revision requests, while those with above 15% frequently face serious academic review. This suggests that the proposed standard already reflects common practice, even if not officially stated.

3. Impact of AI-Assisted Writing Supervisors report that AI-generated content can raise similarity scores because AI often uses common phrases or patterns found in existing publications. Universities are beginning to require students to disclose any AI assistance to avoid accusations of academic dishonesty.

Findings

From the analysis, the following key findings emerged:

• Universities Use Software but Not Clear Cutoffs: While plagiarism detection tools are universal, explicit similarity thresholds are rare.

• Supervisors Carry Responsibility: In most cases, supervisors decide whether overlap is acceptable, leading to variations across faculties and institutions.

• Proposed Threshold is Practical: Applying the <10%, 10–15%, >15% standard to real thesis cases aligned well with decisions actually taken in many institutions.

• AI Requires New Guidelines: Universities need policies explaining how AI-generated text should be disclosed, evaluated, and distinguished from plagiarism.

Conclusion

Plagiarism detection has become an essential part of thesis evaluation worldwide. Yet, the absence of clear, published thresholds creates uncertainty for students and academic staff alike. Based on real practices observed in multiple universities, this article recommends the following standard:

• Less than 10% similarity: Acceptable, no action required.

• 10–15% similarity: Needs evaluation; supervisors may request explanation or revision.

• Above 15% similarity: Unacceptable; may require major revisions or lead to failure.

Such thresholds would improve transparency, reduce inconsistent decisions, and help universities manage both plagiarism and AI-assisted writing fairly. Future policies should also include AI disclosure requirements and faculty training to interpret similarity reports correctly.

References

• Brown, L., & Patel, R. (2019). Academic Integrity in the Digital Age. Oxford University Press.

• Davis, S., & Gómez, M. (2021). “AI-Assisted Writing and Similarity Detection in Graduate Theses.” Journal of Higher Education Research, 45(2), 123–142.

• Harrington, J., & Lee, S. (2018). “Thresholds for Plagiarism: Policy and Practice.” Journal of Academic Ethics, 16(1), 45–60.

• Smith, A., Johnson, P., & Chen, Y. (2020). “University Policies on Plagiarism Across the Globe.” International Journal of Educational Policy, 12(3), 210–228.