AI writing and rising contract cheating present academic institutions with a difficult challenge:
How can we trust that students' submissions reflect their thinking and abilities?
AI is transforming society—and education must adapt. While assessments need to evolve, recent research finds “there appears to be minimal empirical evidence that authentic assessment prevents cheating”(Fawns et al., 2024). Perhaps it's time to accept that, outside invigilation environments, assessment design cannot avoid AI, and that AI is influencing most students’ work, demanding a new approach to academic integrity.
At Norvalid, we believe the focus should shift from detecting AI and cheating to encouraging authentic student work and inspiring critical thinking. To support this, we have developed a novel and evidence-based method: we measure the level of original student writing instead of detecting others’ writing. This process enables us to calculate a Student Authenticity Score, offering a new approach to academic integrity.
What is the Student Authenticity Score?
The Student Authenticity Score measures the amount of original student writing in a submitted text, reflecting the strength of the “student voice.” The score indirectly indicates the presence of AI-generated content and outsourced work, telling a story about the students' engagement in the writing process.
Like the Web, AI cannot be banned, and assessment design cannot rule it out. We must recognise that, outside physically secured environments, students use all available tools—the issue is about using them responsibly. Students should not present AI-written text as their own, as they cannot copy from others without proper citation (plagiarism) or pay someone to do their work (contract cheating). Our Student Authenticity Score provides instructors with insights to judge whether the level of AI writing assistance is acceptable, relative to what was permissible within the assignment and the academic policy —is the student’s submission valid for assessment?
Also, the Student Authenticity Score provides a learning opportunity. The evidence-based report can be used by instructors to talk with students about taking ownership of their work, thinking critically, and developing their academic voice—skills that matter outside the classroom. This way, we hope to empower a new approach to academic integrity in the age of AI, by making students understand the importance of original work and to respect others’ work.
“Swiss Cheese Model”—Multiple Tests Behind the Score
To calculate the Student Authenticity Score, Norvalid applies multiple tests, following the principles of the “Swiss Cheese Model” (UCL, 2023). Each test acts as a protective layer—while one may have weaknesses, others compensate, creating a strong, overlapping system where gaps rarely align, preventing academic misconduct from slipping through.
1. Linguistic Analysis: Did the student write the text?
We extract each student’s unique linguistic fingerprint from an authentic text sample and compare it to new submissions. This forensic method, backed by decades of peer-reviewed research, measures the likelihood that the student authored the submitted text.
2. Perplexity Analysis: Did a human write the text?
We use perplexity, a statistical measure of text predictability, to assess whether a submission is likely human-written. Human writing is naturally more “chaotic” than AI-generated text. Used alongside linguistic analysis, perplexity can suggest contract cheating when the style doesn’t match the student but still appears human-written.
3. Free-text questions: Can the student answer questions from the text?
Students answer short, open-ended questions about their submission immediately after handing it in. These responses confirm the student's knowledge about the content, simulating a lightweight, automated viva voce / oral exam.
4. Cloze Questions: Can the student fill in missing words from the text?
At the moment of submission, students are asked to fill in blanks from sentences pulled from their text. This measures alignment with the text’s linguistic style, not memory. Research shows strong results for authorship validation.
…and we are working on more to come!
Empowering Instructors, Fostering Academic Integrity and Encouraging Critical Thinking
The Student Authenticity Score is not a final verdict—it’s evidence to support informed decision-making. Instructors remain in control, using the score and supporting insights to validate submissions, start conversations, and tailor their responses to the learning context.
- Suppose a submission shows a low authenticity score. In that case, the instructor might follow up with the student, review academic integrity policies, or request a follow-up activity like an oral defence or revision.
- If the score is high, it becomes a moment to recognise and reinforce good practices, strengthening the student’s confidence and motivation.
- In cases where original student writing is absent, the data may serve as a red flag for potential contract cheating.
This approach moves us beyond punishment and surveillance. It opens the door to trust-based, formative evaluation that helps students grow into responsible, independent thinkers—something every educator aims for.
Maintaining academic integrity in a rapidly evolving digital landscape requires innovation, not continued focus on detection. At Norvalid, we’re proud to offer a new, transparent, and student-centred approach that Empowers Instructors, Fosters Academic Integrity, and Encourages Critical Thinking.
Fawns, T., Bearman, M., Dawson, P., Nieminen, J. H., Ashford-Rowe, K., Willey, K., … Press, N. (2024). Authentic assessment: from panacea to criticality. Assessment & Evaluation in Higher Education, 1–13. https://doi.org/10.1080/02602938.2024.2404634
UCL Arena Centre. (2023, January). Designing assessment for academic integrity. UCL. https://www.ucl.ac.uk/teaching-learning/publications/2023/jan/designing-assessment-academic-integrity
Article written with assistance from ChatGPT.
