It usually happens late at night. A submission doesn’t sit right. The writing is too polished, the argument too clean, the voice not quite the one you’ve been reading all semester. You know you’re not supposed to run it through a detector. You know the score isn’t proof. Your institution has probably said as much explicitly. You know the same tools that flag your struggling student’s work will misclassify fluent non-native speakers at alarming rates. You run it anyway.
Not because you trust the tool. Because you don’t know what else to do.
The number comes back. Too high to ignore, not high enough to act on. You take a screenshot. You start drafting documentation. And then you hit the wall every faculty member in this position eventually hits: the evidence isn’t sufficient, the process is unclear, and an hour later you have nothing to show for it except the same unanswered question and much less patience.
This is not a failure of effort. It’s a failure of framework. The detection approach asks faculty to prove AI use, a problem that is by design very difficult to solve, using tools that research has repeatedly shown aren’t up to the task. What nobody has clearly offered is a different question to ask instead.
That question exists. It takes less time than what you’re doing now, produces a clear, actionable result, and does what detection never could. It tells you whether your student learned something.
- Spotlight guide: Boosting AI literacy across your institution
- ‘If we make AI the enemy then surely it must become one’
- GenAI has turned teachers into auditors
The mindset shift
The detection mindset asks: did this student use AI? It’s a question about authorship, and it leads you into an investigative role most faculty are neither trained nor equipped to fill. The evidence is ambiguous, the standard of proof is high, and the outcome is usually inconclusive. The entire process is adversarial. It positions you against your student before you’ve had a single conversation.
The verification mindset asks something different: does this student understand what they submitted? That question is entirely within your professional authority to answer. It doesn’t require a detector or a formal complaint. It requires a conversation with the student, and in most cases, a short one.
This is not a workaround or a compromise. It is a more rigorous form of assessment than a plagiarism score. A student who can explain their argument, discuss their sources, and walk you through their reasoning has demonstrated learning. A student who cannot do those things has not, and that is relevant to their grade regardless of how the work was produced.
At Grand Canyon University, we built this approach into our institutional AI framework under the name learning verification, with explicit policy backing for faculty who use it. The most consistent feedback from faculty who have tried it is that they wish they’d had it sooner.
An unexpected benefit
Verification conversations are not only an assessment tool. They are a teaching opportunity. When you ask a student to explain their thinking, you are engaging them in a conversation about what AI can and cannot do, where their own thinking adds value AI cannot replicate, and what it means to own academic work in an AI-integrated world. Students come away with a more nuanced understanding of AI’s role in their learning than any policy statement or honour code acknowledgement could produce.
This matters. Research finds most institutions lack systematic AI literacy programmes, and the gap between what students need and what institutions currently provide keeps widening. The verification conversation, embedded in your existing assessment practice, creates that literacy in context, around work the student actually produced. A student who struggles to explain a submission they didn’t really write is experiencing, directly and personally, the limits of using AI as a substitute for learning. That is a more powerful lesson than anything on a syllabus.
What this requires of you
Everything here depends on one thing: accepting that your job, when AI use becomes a concern, is not to prove misconduct. It is to assess learning.
That shift feels like a loss to some faculty, as though moving away from detection means lowering standards. The opposite is true. Asking a student to demonstrate understanding is a higher standard than running their work through a tool and hoping the score holds up. It is more honest and more defensible.
You are never going to win the detection arms race. The tools students have access to will always be one step ahead, and the documentation burden will always exceed the evidentiary results. The faculty who manage the challenges of AI most effectively will not be the ones who find better detectors. They will be the ones who stay focused on what their courses were always for: learning, demonstrated by the student, confirmed by the teacher.
That is the trade that verification offers. Detection promised certainty it could never deliver, in exchange for hours of work that rarely produced a usable outcome. Verification offers something smaller and more honest: a conversation, a question, and an answer you can stand behind. It may be less dramatic than a detector score but it is the only thing that actually tells you whether your student learned.
A follow-up piece will cover the practical mechanics, including the time question every faculty member asks first.
B. Jean Mandernach is the executive director of the Center for Educational Technology and Learning Advancement at Grand Canyon University.
If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.
comment