The Footprint of AI in Academic Work: A Reflection from Master’s Thesis Advising

Artificial intelligence has opened new possibilities in academia, but it also raises significant challenges for originality, authenticity, and the assessment of university papers. This article offers a practical perspective on the signs that may help identify AI-generated or AI-assisted texts, as well as teaching strategies to promote the responsible, ethical, and formative use of technology.

Introduction

The rise of artificial intelligence (AI) in academia has opened up new opportunities, but it has also created ethical and practical challenges. One critical challenge we face as faculty members is the increasing tendency for students to use AI tools to draft assignments, raising serious doubts about the originality and authenticity of the submitted content. This phenomenon may compromise not only the assessment process, but also the genuine educational development of students, who must develop critical skills. The question therefore arises: is it possible to reliably detect whether an academic paper has been generated by AI?

As a university instructor, researcher, and lawyer specializing in Compliance, I have witnessed how artificial intelligence has transformed both academic and professional environments. Although AI is a powerful tool with the potential to transform education, it also poses serious challenges for academic integrity. Why? Because more and more students are turning to AI-based technologies to generate academic papers, making it harder to distinguish between genuine effort and automated content.

In my experience, it is essential not only to detect these cases, but also to guide students toward the ethical use of technological tools. This article offers a guide for faculty members and academic advisors on how to identify student submissions created with AI, as well as teaching strategies to encourage originality among students.

Key Indicators for Detecting AI-Generated Text

AI often generates well-organized texts, but these submissions may lack certain key elements that faculty members should identify through rigorous assessment. These indicators should be assessed together and in context; none of them, on its own, proves that a paper was generated by AI.

Changes in Structure and Linguistic Style

One of the first signs that an academic paper may have been created by AI can be found in its structure and style. AI tends to generate texts with formal, well-structured language, but these texts often lack the human variability that comes from reflection or critical reasoning. Some indicators include:

Lack of depth: AI-generated submissions often lack critical and in-depth analysis, limiting themselves to superficial information. A student who is genuinely engaged in research usually provides more reflective comments linked to their understanding of the subject.

An AI-generated paper may look impressive in form, but it may lack in-depth analysis or critical argumentation on the topic. Instead of offering detailed analysis, the text may provide “correct” answers without an original or meaningful interpretation. Comparing the submission with the student’s previous contributions in class can help detect these discrepancies.

Uniformity in style and tone: AI-generated texts often have an extremely homogeneous structure. They do not show natural transitions between sections or variations in tone that would suggest critical or personal thinking by the student. Academic papers written by students usually reflect a more dynamic style, with shifts in tone and complexity as the author deepens the analysis.

A practical example would be a student submitting a criminal law paper on criminal responsibility for international crimes where the critical analysis is flat and lacks clear argumentative development. This may be an indication of AI use.

Absence of personal experience: AI has no genuine personal experience, although it can simulate or fabricate personal anecdotes. Student papers that avoid mentioning the student’s personal context or opinions may raise suspicions.

A vital component of any academic paper is the student’s ability to synthesize the topic with their own professional experiences or perspectives. AI, despite its power, cannot genuinely possess the authenticity of these contributions, even when it can imitate their form. A practical example can be found in legal papers, where a student may contribute practical observations from professional or academic experience. The authenticity of these contributions should be assessed through coherence, context, oral explanation, and the student’s prior academic trajectory.

Superficial argumentation: Although AI systems have advanced considerably, they may still produce superficial or descriptive answers, particularly when a task requires nuanced legal or academic reasoning.

A practical example: in a paper on corporate data protection compliance, an analysis that merely describes legal obligations without exploring more complex legal issues could suggest the use of AI.

Cohesion and coherence: AI can produce long and apparently coherent texts, but it may struggle to maintain a logical argument throughout an entire paper. This can result in a submission that appears solid at first glance, but that on closer analysis may lack a clear logical structure.

Practical example: a paper on the right to be forgotten online that jumps from one topic to another without clear conclusions, or that fails to connect sections coherently, could be an AI-generated text.

Generic citations and references: The sources cited by AI are often generic, widely known, or easily accessible. For example, it may cite classic works instead of recent or relevant research in the specific field. This may contrast with a student who has immersed themselves in current and specific studies.

AI can generate texts with well-structured citations, but it often limits itself to easily accessible, well-known, or generic sources. A solid paper requires specific references, such as recent case studies or specialized research in the field of study. Checking the depth of the sources is key to detecting whether a student submission has been prepared with AI.

We must also bear in mind that AI often has difficulties integrating citations accurately. Although it can generate references, they are often incorrect or irrelevant to the context of the paper. Students who use AI often submit papers with references that, when reviewed, do not exist or have no relation to the central topic.

For example, a paper that incorrectly mentions the General Data Protection Regulation (GDPR) in an unrelated context, or where the cited sources do not correspond to legitimate publications, may be a significant warning sign that the text has been generated or heavily assisted by AI.

Originality Analysis

AI-detection and plagiarism-analysis tools may support an originality review, but their results should never be treated as conclusive evidence on their own. Many AI systems tend to produce texts that reuse common structures or standard phrases. Tools such as the following can help identify patterns that may deserve closer academic review:

Winston AI: A cloud-based tool that uses machine learning to flag possible AI-generated content and plagiarism.

Originality.AI: A tool combining AI-detection features and plagiarism checking, which may support originality reviews in academic contexts.

GLTR.IO: A visual and forensic analysis tool that helps examine linguistic patterns in short texts.

GPTZero: A tool designed to flag text that may have been generated or assisted by AI.

CheckGPT: Analyzes and determines whether a text was created using AI.

Turnitin: Expanded its well-known plagiarism detection software to include AI-writing indicators, whose results require human interpretation and may produce false positives.

Undetectable AI: Uses text analysis techniques to flag possible automated content.

OpenAI Detector: OpenAI’s former AI text classifier, withdrawn in July 2023 because of its low accuracy. It is retained here as a historical reference to the evolution of AI-detection tools.

Metadata analysis tools: By examining the creation and editing data of a file, it is possible to identify clues as to whether the content was generated or edited using AI.

These technological tools, which can assist in detecting the use of AI in academic submissions, may be helpful for faculty members. Nevertheless, human analysis remains essential. Detection tools should be used as supporting indicators within a broader academic assessment that includes human review, comparison with prior submissions, oral discussion, and analysis of the student’s learning process.

It is also essential to foster a culture of academic integrity. Students must be aware of the consequences of improper use of AI and of the importance of developing their own critical thinking.

Mismatch in the Student’s Level of Knowledge

One of the most important indicators is the disparity between the level of knowledge demonstrated by the student in class and the level reflected in their academic papers. As an instructor, I have found that students who rely on AI tend to submit papers that do not reflect their level of command of the subject.

Complexity Without Justification

AI tends to generate responses that appear advanced in terminology and structure, but lack justification or practical examples. A solid academic text requires the student to justify concepts with examples or concrete applications. If a paper seems too technical for the student’s level, this is a warning sign.

Lack of Depth in the Analysis

As mentioned above, AI-generated submissions may appear very sophisticated on the surface, but they may fail in deeper analysis. A practical example of this phenomenon occurs when a critical analysis is requested and the student submits a text that presents many ideas but develops none of them in particular. In these cases, inviting the student to an interview to discuss the paper can reveal inconsistencies.

Teaching Strategies to Mitigate the Use of AI

In my teaching experience, I have found that it is crucial not only to detect the use of AI, but also to implement preventive strategies that encourage originality.

That said, we must understand that when used responsibly, technology can help faculty members teach more effectively and enable students to deepen their learning. Technology is in continuous development and growth; it is already available to everyone and often free of charge. Resisting or opposing it is not only a futile effort, but it also isolates academia from broader technological and societal evolution. We must therefore encourage learning, continuous assessment, and process-based assessment.

In other words, instead of basing assessment on a single final paper, the process can be divided into partial submissions, such as drafts, summaries, or outlines. In this way, students show their progress gradually, making it more difficult for them to rely on AI to write the entire paper. This approach also allows for continuous feedback.

Example: In a paper on digital law, asking for a detailed outline of the argument before the full draft forces the student to structure their thinking independently and show critical progress from the early stages.

Another alternative would be the design of personalized assignments; that is, implementing an effective strategy by designing tasks that require students to connect academic content with their own experiences or critical reflections. For example, in a law course, students can be asked to analyze a recent case and explain how they would apply theory to the facts. This type of task makes it harder for AI to generate adequate answers.

Another highly effective measure, in my view, would be to conduct occasional or routine complementary oral assessments.

Inviting the student to defend or explain certain aspects of their paper can reveal whether they really understood the topic or simply relied on AI. A practical example would be to conduct a brief interview on the key points of their research paper.

Talking with the student about the submitted paper makes it possible to detect inconsistencies between the knowledge expressed in writing and the knowledge demonstrated orally. For example, if a student submits a paper that seems too technical or advanced, asking about the reasoning behind certain decisions may help identify whether they really understood the content.

Another very effective alternative is to compare the style and level of the student’s current submission with their previous papers. In most cases, this is very revealing. If there is a significant improvement without justification in class, the use of AI may be suspected.

The use of controlled writing environments may also be another alternative to prevent students from abusing AI. A practical example that can be used in class would be to create supervised writing sessions in face-to-face or online classes, where students write within a limited time and without access to external tools. This approach helps ensure that the content is authentic and reflects their real level of knowledge.

For example, during a class we can ask students to write a 500-word critical reflection on a recent article or judicial decision, without access to digital tools.

Assessment based on reflection and metacognition can also be a good alternative. This means asking students to include a personal reflection section on the research and writing process, describing how they reached their conclusions and what challenges they faced. This introspection reveals the student’s understanding and learning process, which is difficult to simulate with AI.

Example: At the end of a paper on data protection, ask for a 300-word reflection on how current legislation affects the student’s daily life or the lives of people around them.

Group projects with individual tasks can also be used: we can assign group projects in which each student has a specific and individual task. This reduces the temptation to use AI, since each member of the group must demonstrate their personal contribution to the overall project.

For example, if students analyze European data protection regulations, each student could be responsible for researching a specific article or section of the General Data Protection Regulation (GDPR), and then presenting how it affects different sectors.

Problem-based learning techniques (PBL) involve proposing complex problems or ethical dilemmas that students must solve. AI tends to generate more generic responses to problems that require lateral thinking and innovative solutions.

Example: Present a scenario in which a company has violated a citizen’s right to be forgotten and ask students to develop a legal action plan for defending the case before the courts.

Peer review can be another alternative. We can include in the assessment a phase in which students review and comment on their classmates’ papers. This process encourages greater awareness of content quality and requires each student to defend and explain their ideas to their peers.

Example: After submitting an analysis of a court case, have students exchange papers and assess the quality of the argumentation and originality, providing constructive comments.

As faculty members and educators, we have a commitment to excellent teaching and, as a result, we must promote academic ethics.

It is essential to educate students on the ethical use of technology. Instead of banning AI tools, we should teach them how to use them responsibly, for example to improve grammar or text structure, but not to create complete content. Transparency regarding the use of digital tools must be a key component of academic policies, both for students and for faculty members themselves.

Conclusion

Detecting the use of AI in academic papers has become a growing challenge for faculty members. Nevertheless, an appropriate combination of technological tools, critical analysis, and teaching strategies makes it possible to identify these cases and, even more importantly, guide students toward ethical and responsible use of technology. The instructor’s expert judgment remains irreplaceable in this process. As educators, we must not only protect academic integrity, but also prepare our students for an increasingly digital and complex environment.

However, we cannot afford to fall behind the pace of technological development. What we understand and control today may change radically tomorrow. If we do not stay up to date, we run the risk of becoming obsolete, which would affect us not only as professionals but also the quality of education we provide. Our responsibility therefore also lies in adapting continuously, so that we can continue to be effective guides in a constantly changing world.

Marcos Romero Perin

Reflections from my experience as an instructor, researcher, and lawyer passionate about new technologies

References

Anijovich, R. (2017). La evaluación formativa en la enseñanza superior. Voces de la Educación, 2(1), 31-38

Bearman, M. and Ajjawi, R. (2023). Learning to work with the black box: Pedagogy for a world with artificial intelligence. British Journal of Educational Technology, 54(5), 1160-1173. https://doi.org/10.1111/bjet.13337

Coll Salvador, C., Díaz Barriga Arceo, F., Engel Rocamora, A. and Salinas Ibáñez, J. (2023). Evidencias de aprendizaje en prácticas educativas mediadas por tecnologías digitales. Revista Iberoamericana de Educación a Distancia, 26(2), 9-25. https://doi.org/10.5944/ried.26.2.37293

Dai, Z., Xiong, J., Zhao, L. and Zhu, X. (2023). Smart classroom learning environment preferences of higher education teachers and students in China: An ecological perspective. Heliyon, 9(6), e16769. https://doi.org/10.1016/j.heliyon.2023.e16769

Pérez, S. (February 15, 2023). Consejos y herramientas para evitar el plagio en tus trabajos. Biblioteca Universitat Oberta de Catalunya. https://biblioteca.uoc.edu/es/actualidad/noticia/Consejos-y-herramientas-para-evitar-el-plagio-en-tus-trabajos/

Courses

Blockchain Basico (March 2024, FCOI03)

Inteligencia artificial aplicada a la gestión de procesos (June 2024, CR25-05/2023/5615ENL-1/107 Madrid).

Web AI Tools

Winston AI (March 2024) https://gowinston.ai/

Originality AI (March 2024) https://www.originality.ai

https://gptzero.me

CheckGpt (April 2024) https://www.yeschat.ai/

Turnitin (June 2024) https://www.turnitin.com

Undetectable AI (March 2024) https://undetectable.ai

GLTR IO (March 2024) https://gltr.io

OpenAI Detector (June 2024) https://openai.com/detector

Quillbot (July 2024) https://quillbot.com/ai-content-detector

This text forms part of Marcos Romero Perin’s academic and professional archive on artificial intelligence, university teaching, technology applied to law, and a culture of integrity in digital environments.