AI detection software promised a straightforward solution to a genuinely difficult problem, but its track record has been inconsistent enough that many departments have grown cautious about relying on it for high-stakes decisions. False positive rates remain a real concern, particularly for multilingual writers and students whose natural prose style happens to resemble patterns the detector associates with generated text. A single flagged essay can trigger a disciplinary process that damages trust between a teacher and student even when the underlying accusation turns out to be wrong. That risk has pushed a growing number of departments toward a different strategy: designing assignments that are harder to complete meaningfully with AI in the first place, rather than trying to catch misuse after submission.

Assignment redesign generally moves away from generic prompts that any large language model can answer competently without specific context. A prompt asking students to analyze a widely discussed novel in general terms is exactly the kind of task a generative model handles well, since it draws on broad training data rather than anything unique to the classroom. A prompt that asks students to connect that same novel to a specific class discussion, a personal reading experience, or a piece of evidence from an in-class annotation exercise is much harder to answer convincingly without having actually done the work. This kind of specificity does not eliminate misuse entirely, but it raises the effort required to fake competence to the point where doing the assignment honestly is often the more efficient path.

Process-based assessment is the other major shift departments are making. Rather than grading only a finished essay, teachers are building in checkpoints, an outline, an annotated source list, a rough draft with visible revision, that make the writing process itself part of the evaluated work. A student cannot easily fabricate a believable draft-to-final revision history using AI alone, especially when that history includes specific choices tied to earlier class discussions or teacher feedback. This approach does add grading touchpoints throughout a unit rather than concentrating all evaluation at the end, which some teachers initially resist, but many find it actually reduces the anxiety of a single high-stakes final submission.

Building Assignments That Reward the Process

Assignments that reward process over polish tend to look different from traditional essay prompts in a few consistent ways. They often require students to cite specific in-class moments, whether a discussion, a source shared only in class, or a peer's comment during workshop, in ways that would be difficult to fabricate without having attended and participated. They may also ask students to reflect explicitly on their own revision choices, explaining why they cut a paragraph or restructured an argument, which requires genuine engagement with an earlier draft rather than generation of a finished product from scratch. None of these techniques make AI misuse impossible, but they shift the incentive structure so that doing the work honestly becomes the path of least resistance rather than the harder option.

• Require citation of specific in-class discussions, sources, or peer feedback
• Grade process artifacts like outlines and annotated drafts alongside the final essay
• Ask for explicit reflection on revision choices between draft and final version
• Avoid generic prompts that any large language model can answer without classroom context
• Treat detection flags as a starting conversation rather than a final verdict

Where AI Fits Into a Student's Legitimate Writing Process

Redesigning assignments for integrity works best alongside clear, explicit expectations about what AI use is and is not acceptable for a given task, since ambiguity is what tends to create both accidental violations and false accusations. Some teachers allow AI for brainstorming or grammar checking but prohibit it for drafting actual sentences, while others take a stricter no-AI stance for specific high-stakes assignments and a more permissive one for lower-stakes practice work. Whatever the policy, students need it stated plainly at the start of an assignment, not implied or left to individual interpretation, because students genuinely differ in what they assume is allowed without explicit guidance. Departments that publish a consistent, grade-level-wide AI use policy tend to see fewer disputes than those where the rule changes from teacher to teacher.

It is worth distinguishing between a student using AI to write their essay and a teacher using AI to help grade it, since these are entirely different uses with different implications. A teacher applying an AI first-pass score against their own rubric, then reviewing and personalizing that feedback before it reaches the student, is using the technology to handle a mechanical, repetitive task while keeping human judgment over the final evaluation. That is a very different relationship to AI than a student submitting generated prose as their own original argument. Departments building AI use policy for students often find it useful to be explicit about this distinction, since students sometimes reasonably ask why a tool that saves the teacher time is off limits to them, and the honest answer is that the two uses serve entirely different purposes in the writing and grading process.