The Best Response to Student AI Use Is Not Detection, It Is Dialog

Synthesis: Mandernach (2026), a practitioner account from Grand Canyon University, argues that the best institutional response to student AI use is learning verification through dialog — not detection. Because AI detectors are unreliable (and disproportionately flag nonnative English speakers) and formal misconduct processes require detection-grade proof that is functionally unavailable, faculty are left with "suspicion without recourse" that corrodes Trust. GCU instead shifted the question from "Did this student use AI?" to "Can this student demonstrate understanding of what they submitted?" — an Assessment approach faculty are already qualified to run. Implemented institution-wide in fall 2025, learning verification restores faculty authority and changes what assignments require, making AI a less attractive shortcut.

Key Findings

• Detection is unreliable and formal processes stall. AI detectors produce unacceptable error rates and are disproportionately likely to flag writing by nonnative English speakers; Code of Conduct processes require proof that is functionally unavailable with AI, so cases stall, students are not found responsible, and faculty feel dismissed.
• Reframe the problem: AI use is a confounding variable. AI use is not the true concern — it is one of several things (ghostwriting, overstepping tutors, contract cheating) that disrupt the inferential link between a submitted artifact and actual student understanding. The real question is whether the student mastered the learning objectives.
• Learning verification in practice. A syllabus statement normalizes that students may be asked to explain their work at any point (framed as Pedagogies and Teaching Strategies, not surveillance); verification conversations (5–10 minutes) focus on the material, not the tool — an extension of demonstrating understanding. Alternatives for large/asynchronous courses: written reflections, short student videos, early draft submissions, and (when used) AI chat logs.
• Trust and relationships shift. Students initially approached verification with anxiety (assuming they were accused), but when communicated well they began disclosing AI use more openly and engaging with assignments as genuine learning tasks rather than products. The framework treats AI use as acceptable when the student can demonstrate understanding.
• Faculty and institutional challenges. Faculty's top concern was time, which softened once 5–10 minute conversations proved cleaner than formal proceedings. The harder persistent challenge is faculty who are philosophically unwilling to treat verified AI use as neutral — a gap addressed through brown-bag conversations rather than policy training. Formal, systematic evaluation of outcomes is still in development.

What this means for practice

• Instructors. Rebuild the integrity question around learning rather than tool use: ask "Can this student demonstrate understanding of what they submitted?" instead of "Did this student use AI?", because AI use is only one of several confounds — alongside ghostwriting, tutors overstepping, and contract cheating — sitting between a submitted artifact and actual understanding.
• Instructors. Say it in the syllabus, in advance, that students may be asked to explain their work at any point, and frame it as pedagogy rather than surveillance; this reframing is what makes learning verification acceptable to students rather than a threat.
• Instructors. Hold short verification conversations — five to 10 minutes is usually enough — that stay on the material, not the tool. Faculty at the implementing institution found these cleaner and less time-consuming than formal misconduct proceedings, and the practice is an extension of demonstrating understanding rather than a new procedure.
• Instructors. Scale the same logic in large or asynchronous courses with written reflections, short student videos, early draft submissions, or AI chat logs, which show how a student interacted with a tool and whether they critically evaluated its output.
• Instructors. Announce the purpose before the first conversation: students initially read a request to explain their work as an accusation, and the author reports that when the rationale was communicated well students disclosed AI use more openly and treated assignments as learning tasks rather than products.

Limitations

• This is a practitioner account from one institution, Grand Canyon University, where learning verification was implemented institution-wide in fall 2025; no comparison group, pre/post measure, or student outcome data are reported.
• The author states that formal, systematic evaluation is still in development — the mechanisms under exploration are supplemental questions on end-of-course student evaluations and a dedicated faculty survey — so the reported effects on faculty authority and student engagement rest on unsolicited faculty feedback rather than measurement.
• The account identifies a problem the framework does not yet solve: faculty who are philosophically unwilling to treat verified AI use as neutral, which the author describes as the place where implementation "is most honestly still in progress" and which is being addressed through brown-bag conversation rather than policy.
• The policy was launched with the expectation that it would need revision after its first full year, and no first-year implementation data or student outcome evidence are presented.

Citation

Mandernach, B.J. (2026). The best response to student AI use is not detection, it is dialog. Change: The Magazine of Higher Learning, 58(4), 28–33.