Beyond AI Detection
Supporting consistency, fairness, and evidence of student learning
Authors: Lorelei Anselmo, MEd, Ali Mikaeili, Ed.D (Student)
Last modified: October 10, 2026
Why this matters
Evidence of student learning is foundational to effective teaching and assessment. The goal is to evaluate what students know and can do, how they developed that knowledge and skill, and whether they have met the intended learning outcomes. In a context where human and technological contributions may be increasingly intertwined, assessment should not depend on trying to prove that every word was produced without AI assistance (Eaton, 2023; 2025).
Although AI detection tools promise a quick answer to questions about authorship, current research indicates that these technologies have substantial accuracy and validity limitations. Importantly, this research indicates that their outputs should not be relied upon to determine whether an academic integrity violation has occurred (Bassett et al., 2026; Perkins et al., 2024).
AI detection results are probabilistic estimates, not independently verifiable evidence of who authored a text, which tools were used, or whether any use of AI was unauthorized (Bassett et al., 2026; Perkins et al., 2024). Instead of relying on detection scores, instructors are encouraged to focus on evidence of student learning, consistency and fairness, clearly communicated expectations, and supportive conversations when concerns arise.
*The University of Calgary does not currently provide or centrally support an AI detection tool.
Related content
- Item 1
- Item 2
- Item 3
- Item 4
AI detection results are not reliable evidence of academic misconduct
Detection tools estimate whether particular textual patterns resemble AI-generated writing, but they cannot independently establish authorship, identify how a text was produced, or determine whether a student used unauthorized assistance (Bassett et al., 2026; Kumar et al., 2024).
False positives create equity and fairness concerns
Human-authored writing can be misclassified as AI-generated, and large-scale evaluations of AI detection tools have documented false positive results even when texts were written entirely by humans. Research also indicates that non-native English and multilingual writers may be disproportionately flagged by some detection tools, raising additional concerns about equity and fairness in academic integrity processes (Giray et al., 2026; Liang et al., 2023; Weber-Wulff et al., 2023).
Detection results can be vulnerable to circumvention
The accuracy of detectors may decline substantially when AI-generated text is modified through relatively simple techniques. This further limits their suitability for determining whether an academic integrity violation occurred (Perkins et al., 2024).
Uploading student work to external tools may create additional risks
Depending on the tool and the information submitted, concerns may include privacy, informed consent, security, copyright, and intellectual property. Institutions must also ensure that the collection, use, disclosure, and protection of personal information comply with Alberta’s Protection of Privacy Act (POPA; SA 2024, c P-28.5) and applicable institutional privacy requirements.
Academic integrity concerns must be addressed through established institutional processes
When a concern arises, instructors should consult the Student Academic Misconduct Policy, the Student Academic Misconduct Procedure, and the appropriate academic decision-maker within their faculty (University of Calgary, 2019a, 2019b).
Responding to concerns about student work
If a concern arises, the first step is to contact the academic leader responsible for academic misconduct in your faculty, and that contact should come before any conversation with the student. The guidance below is written for instructors of record. It is not a process for teaching assistants, who should bring anything they notice to the course instructor rather than raising it with the student directly. Where a conversation does take place, it sits inside that faculty procedure as a way of understanding a student’s process. It does not replace the procedure and is not a means of reaching a finding.
If something about a piece of student work gives you pause, begin by reviewing the assignment instructions, the course expectations regarding AI use, and the available evidence of learning. Every faculty has a standard operating procedure for academic misconduct.
A direct and supportive conversation may help clarify the student’s process.
These questions are intended to support understanding. They are not a substitute for the formal procedures required when an allegation of academic misconduct is pursued.
- Can you walk me through how you approached this assignment, from start to finish?
- What sources, notes, technologies, or other supports did you use?
- How did you decide which ideas, examples, or evidence to include?
- Is there a part of the work you would feel comfortable expanding on or explaining?
- Do you have earlier drafts, outlines, notes, version histories, or other evidence of your process?
- How does this work connect with the course materials or discussions?
- Was there anything about the assignment instructions, timing, or scope that was unclear or particularly challenging?
- If you used generative AI or another writing tool, how did you use it, and how did you review or revise its output?
Ask these questions with curiosity rather than suspicion. A conversation may provide useful context about a student’s learning process, but confidence, fluency, writing style, nervousness, or an ability to respond immediately should not be treated as proof of either misconduct or innocence.
Consider accessibility, language, cultural, and communication needs. Provide an appropriate format and reasonable opportunity for the student to explain their process. If the concern remains, follow the University’s established policy and procedure rather than conducting an independent investigation or making a determination based only on the conversation.
Many concerns can be prevented by setting expectations early. The Taylor Institute resource Communicating Generative AI Use With Your Students offers example course-outline clauses for AI-free, AI-scaffolded, and AI-integrated assignments. Each clause pairs a permitted and not-permitted table with the course learning outcome it serves, so students can see both what is allowed and why.
Building consistency and fairness
Consistency and fairness are supported through transparent expectations, consistent processes, and assessment practices that make learning visible. Consider the following:
Clearly communicate whether, when, and how students may use generative AI in the course and in each assessment
Distinguish among AI-free, AI-scaffolded, and AI-integrated approaches to each assessment
Explain how students should acknowledge, disclose, or document permitted AI assistance
Ensure that expectations are communicated before students begin an assessment
Focus on evidence of learning rather than attempting to infer authorship from writing style alone
Explain how questions about academic integrity will be addressed and where students can seek advice or support
Review more than one source of relevant information before deciding whether a formal referral is warranted
Avoid treating an AI detection score, writing-style change, or isolated textual feature as proof of unauthorized AI use
Apply institutional policy and procedure consistently, using the linked policy and procedure in the UCalgary Policies and Resources section below
Consider the potential effects of an unverified allegation on a student’s learning, well-being, and confidence in the educational process
Fairness does not mean overlooking legitimate concerns. It means addressing those concerns through transparent, consistent, and evidence-informed processes rather than through assumptions or unverifiable technological outputs (Bassett et al., 2026).
Assessment and course design considerations
Assessment design can provide richer and more meaningful evidence of learning than an AI detection score. Process-oriented and authentic approaches may also help clarify expectations, make student learning more visible, and support constructive conversations about responsible AI use (Kofinas et al., 2025; Kumar et al., 2024).
Depending on the course context and learning outcomes, instructors might consider:
- collecting proposals, outlines, drafts, annotations, or reflections at meaningful points;
- asking students to explain selected decisions made during the learning process;
- connecting assignments to course learning outcomes and course-specific discussions, experiences, data, or contexts;
- incorporating opportunities for feedback and revision;
- inviting students to document how permitted technologies contributed to their work;
- designing tasks that require disciplinary judgment, application, interpretation, or reflection; and
- providing accessible alternatives when oral explanations, presentations, or live check-ins are used.
These practices should not be implemented merely as surveillance mechanisms. Their primary purpose should be to support learning, feedback, transparency, and alignment with the intended learning outcomes.
Guiding questions for course and assessment design
These questions are designed to work alongside the UCalgary Assessment Principles and the Taylor Institute’s assessment and generative AI resources, both linked in the section below, which offer fuller guidance on aligning assessment with intended learning outcomes.
- How might students demonstrate their learning process, not only submit a final product?
- What evidence of learning could be collected at meaningful points during the course?
- Are expectations about AI-free, AI-scaffolded, or AI-integrated use clear for this assessment?
- Are the expectations aligned with the intended course learning outcomes?
- How will students acknowledge or disclose permitted AI assistance?
- Does this assessment reward a polished final product more than it rewards the thinking, reasoning, and learning behind it?
- How does the grading for this assessment align with the knowledge and skills expected in the course learning outcomes?
- What accessible alternatives are available if an oral explanation or live check-in is used?
- How can transparency and dialogue reduce reliance on attempts to verify authorship?
- How will concerns be addressed fairly if the submitted work does not appear to align with other available evidence of learning?
UCalgary policies and resources
Use these resources to inform instructional decisions, assessment design, AI literacy, and responses to possible academic integrity concerns.
You do not have to navigate these decisions alone. If you would like to talk through assessment design, course outline language about generative AI, or ways to make evidence of learning more visible in your assessments, contact the Taylor Institute for Teaching and Learning.
References
Bassett, M. A., Bradshaw, W., Bornsztejn, H., Hogg, A., Murdoch, K., Pearce, B., & Webber, C. (2026). Heads we win, tails you lose: AI detectors in education. Journal of Higher Education Policy and Management. Advance online publication. https://doi.org/10.1080/1360080X.2026.2622146
Eaton, S. E. (2023). Postplagiarism: Transdisciplinary ethics and integrity in the age of artificial intelligence and neurotechnology. International Journal for Educational Integrity, 19, Article 23. https://doi.org/10.1007/s40979-023-00144-1
Eaton, S. E. (2025). Global trends in education: Artificial intelligence, postplagiarism, and future-focused learning for 2025 and beyond: 2024–2025 Werklund Distinguished Research Lecture. International Journal for Educational Integrity, 21, Article 12. https://doi.org/10.1007/s40979-025-00187-6
Giray, L., Roe, J., & Diesta Espiritu, J. (2026). AI writing detectors are ineffective, unreliable and harmful. English Teaching: Practice & Critique. Advance online publication. https://doi.org/10.1108/ETPC-07-2025-0155
Kofinas, A. K., Tsay, C., & Pike, D. (2025). The impact of generative AI on academic integrity of authentic assessments within a higher education context. British Journal of Educational Technology, 56, 2522–2549. https://doi.org/10.1111/bjet.13585
Kumar, R., Eaton, S. E., Mindzak, M., & Morrison, R. (2024). Academic integrity and artificial intelligence: An overview. In S. E. Eaton (Ed.), Second handbook of academic integrity (pp. 1583–1596). Springer. https://doi.org/10.1007/978-3-031-54144-5_153
Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), Article 100779. https://doi.org/10.1016/j.patter.2023.100779
Perkins, M., Roe, J., Vu, B. H., Postma, D., Hickerson, D., McGaughran, J., Khuat, H. Q., & Chan, J. (2024). Simple techniques to bypass GenAI text detectors: Implications for inclusive education. International Journal of Educational Technology in Higher Education, 21, Article 53. https://doi.org/10.1186/s41239-024-00487-w
Protection of Privacy Act, SA 2024, c P-28.5. View the current statute
University of Calgary. (2019a). Student academic misconduct policy. View the Student Academic Misconduct Policy
University of Calgary. (2019b). Student academic misconduct procedure. View the Student Academic Misconduct Procedure
Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S., Foltýnek, T., Guerrero-Dib, J., Popoola, O., Šigut, P., & Waddington, L. (2023). Testing of detection tools for AI-generated text. International Journal for Educational Integrity, 19(1), 26. https://doi.org/10.1007/s40979-023-00146-z