Whatever your institution’s policy on student AI use, whether strict prohibition, permitted use with disclosure, or something still being worked out, you face the same underlying challenge: when a submission raises questions about whether a student did their own thinking, you need a legitimate way to find out. AI detectors won’t give you one; research has shown they produce unacceptable false positive rates and carry particular bias against non-native English speakers (Liang et al., 2023; Weber-Wulff et al., 2023). Formal academic integrity processes require a standard of proof that AI-related cases almost never meet.
Learning verification gives you that pathway. Rather than trying to prove how a submission was produced, you ask the student to demonstrate that they understand what they submitted. That question is always within your professional authority, regardless of your institution’s AI policy, because it is a question about learning, not conduct. A student who can account for their work has demonstrated learning. A student who cannot has not, and that is directly relevant to the grade.
What follows is a step-by-step guide to implementing learning verification in any course, including a menu of methods so you can choose what fits your context.
Step 1: Set the Expectation in Your Syllabus
Verification lands very differently when students encounter it for the first time mid-semester versus reading it on day one. Students who see it without warning assume accusation. Students who read it upfront understand it as a teaching expectation.
Frame it as an extension of your assessment philosophy, not a response to AI. Keep it brief and focused on learning.
Sample language: “I am committed to ensuring that your grade accurately reflects your learning. As part of my assessment practice, I may ask you to briefly discuss or explain your submitted work at any point during the semester. This is not a test of memory but an opportunity to demonstrate your thinking and ensure your grade reflects your understanding.”
Step 2: Reinforce the Expectation at Assignment Launch
When introducing major assignments, briefly remind students that you are interested in their thinking process, not just the final product.
Sample language: “As you work on this assignment, focus on developing your own reasoning; I may follow up with a question or two about your thinking after reviewing your submission.”
One sentence, delivered consistently, shifts how students approach their work before any concern arises.
Step 3: Identify Submissions That Raise Questions
You are looking for discrepancies between the submitted work and what you know about the student: writing that sounds unlike their previous work, arguments unusually sophisticated without supporting evidence elsewhere in the course, or responses generic enough to suggest the specific prompt wasn’t genuinely engaged with.
Trust your professional judgment. You know your students.
Step 4: Choose Your Verification Method
This is where learning verification becomes flexible. There is no single right approach; the method should fit the assignment, your course format, and what you need to know. Here are five options, each suited to different contexts.
One-on-One Conversation
The most direct method. Reach out with a neutral message: “I’d like to connect briefly to talk through your ideas on this assignment, about ten minutes.” In the conversation, ask the student to walk you through their thinking: their central argument, how they approached the problem, what they found challenging, what they would add if they had more time. Five to ten minutes is usually enough to determine whether genuine understanding is present.
Written Reflection
Ask the student to submit a short written response explaining their reasoning, the choices they made, and what they learned through the process. This works well for larger courses where scheduling individual conversations is impractical, and it can be built into the assignment as a required final component rather than added after the fact. It shifts the time investment to the student and creates a written record of their thinking alongside the submitted work.
Student-submitted Video
Ask the student to record a brief video of two to five minutes explaining their approach, argument, or key takeaways in their own words. This works particularly well in asynchronous online courses where face-to-face conversation is harder to arrange, and it allows you to review responses on your own schedule. Students who engaged with the material speak naturally and specifically. Students who didn’t typically struggle to move beyond surface-level description.
Draft Submission Trail
Require students to submit earlier drafts alongside the final version, along with a brief note explaining how the work evolved. A genuine draft trail shows development, revision, and the kind of iterative thinking that AI-generated content typically lacks. This method is most effective when built into the assignment design from the start, which makes verification a natural part of the process rather than something triggered by concern.
AI Chat Log Review
For assignments where AI tools are permitted, ask students to submit their chat logs alongside the final work. The log reveals how the student interacted with the tool: what prompts they used, how they evaluated and revised the output, whether they engaged critically or simply accepted what was generated. A student who used AI thoughtfully will have logs that show that thinking. A student who used it to bypass the work typically will not.
Step 5: Reach Out Simply and Neutrally
When a submission raises questions and you are using a method that requires direct contact, keep your message brief and free of any signal of suspicion.
Sample language: “Hi [Name], I’d like to connect briefly to talk through your ideas on [assignment]. Would you be available this week? It should only take about ten minutes.”
You are scheduling an assessment, not opening an investigation. The framing matters both for the student’s experience and for your own professional positioning.
Step 6: Document and Grade
After verification, record the outcome in a single sentence in your grade book: “Verification conducted [date]; student demonstrated [strong/partial/limited] command of submitted work.”
Let the grade reflect what verification revealed. If a student demonstrates genuine understanding, the submitted grade stands. If they cannot account for the work, adjust the grade to reflect the learning that was actually demonstrated. You do not need to prove how the work was produced; you need only to assess what the student knows, which is what grades are for.
Verification requires time, but so does every alternative. A formal academic integrity complaint takes more time per student, involves more institutional friction, and typically arrives at an inconclusive result. A verification conversation or reflection takes five to ten minutes and produces a clear, gradeable outcome. Early in a course the volume of concerns may be higher; it drops substantially once students understand that your course requires demonstrated understanding, not just submitted work.
The Bigger Picture
Faculty have always had the authority to ask what a student knows. Written assessment didn’t replace that authority; it just made exercising it less necessary, because a well-designed assignment could stand in for direct dialogue. AI has changed that equation. When a submitted document can no longer be trusted as evidence of learning, the faculty member’s judgment becomes the instrument of assessment again.
These five methods are not new inventions. They are structured ways to do what good teachers have always done: find out what a student actually understands. The only thing that has changed is how urgently that ability matters. In a course where AI tools are available to students, the grade that reflects demonstrated understanding is the only grade that means anything. Learning verification is how you get there.
AI Disclosure: The author used Claude (Anthropic) as a collaborative drafting and editing tool in the development of this piece. AI contributions included structural suggestions and prose drafting. All arguments, institutional details, and editorial judgments reflect the author’s own expertise and professional experience. The author reviewed, revised, and approved all content prior to submission.
B. Jean Mandernach, PhD, is Executive Director of the Center for Educational Technology and Learning Advancement at Grand Canyon University, where she leads institutional AI policy development and faculty support initiatives.
References
Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), 100779. https://doi.org/10.1016/j.patter.2023.100779
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). https://doi.org/10.1007/s40979-023-00146-z
