Academic Integrity Without Detectors
Why AI detection tools are an unreliable basis for accusations, and what works instead: process evidence, conversations, and assignments that make learning visible.
The Problem With AI Detectors
AI detection tools promise to tell you whether text was written by AI. Applying discernment to the detectors themselves shows serious problems:
- False positives. Detectors flag human-written work as AI-generated. Research has found they disproportionately flag writing by non-native English speakers, whose simpler phrasing can resemble AI output (Liang et al., 2023, Patterns).
- Easy to evade. Light editing, paraphrasing, or asking AI to write “less like an AI” often defeats them, so the students most determined to cheat are the least likely to be caught.
- No evidence trail. A score from a black box can’t be explained or verified, and it isn’t proof of anything.
- Unstable. As AI models change, detector accuracy changes too. OpenAI withdrew its own AI text classifier in 2023, citing its low rate of accuracy.
A false accusation of cheating can seriously harm a student. That’s a diligence issue: you are accountable for decisions you make based on a tool’s output, just as with any other AI system.
Bottom line: a detector score should never be the basis for an accusation. At most, it’s a reason to look more closely using the approaches below.
What Works Better
1. Make the Process Visible
Assignments that show how work developed make both learning and integrity easier to see:
- Drafts, outlines, and notes submitted along the way
- Version history in shared documents
- Short reflections: “What was the hardest decision in this draft?”
- AI use statements, where AI is permitted
2. Talk With the Student
If a submission doesn’t match what you know of a student’s work, start with a conversation, not an accusation:
- “Walk me through how you developed this argument.”
- “What did you mean by this paragraph?”
- “What sources did you use, and how did you find them?”
Students who did the work can usually discuss it. A conversation also gives a student who misused AI a chance to be honest, and you a chance to teach.
3. Design for Learning That’s Hard to Outsource
Assignments that connect to class discussion, personal experience, local context, or in-class work are harder to hand off to AI, and usually more meaningful. For a full approach to redesigning objectives and assessment, see Teaching Above the Line.
4. Follow a Fair Process
Know your institution’s academic integrity process and follow it. Document what you observed, not what a tool reported. Treat a first, unclear case as a teaching moment when policy allows.
Common mistake: emailing a student “Our AI detector shows 87% AI-generated, so you’ll receive a zero.”
The score can’t be verified, may be wrong, and may reflect bias against certain writing styles. Start from what you can actually observe, such as the absence of drafts or a mismatch with in-class writing, and talk with the student.
- Choose an assignment where you’ve worried about AI misuse.
- Add two process checkpoints (for example, an outline and a draft, or a proposal and a reflection).
- Add one element tied to your class: a discussion, a local example, or the student’s own experience.
- Write three conversation questions you’d use if a submission seemed inconsistent with a student’s work.
- Ask an AI assistant to try completing your redesigned assignment, then look at what it produced. Which parts could it do convincingly, and which couldn’t it? Adjust if needed.
1. Why shouldn't an AI detector score be the basis for an accusation?
Unreliable, unexplainable scores can wrongly harm students. Decisions should rest on evidence you can observe and discuss.
2. A strong essay doesn't match a student's in-class writing. What's the best first step?
A conversation gathers real evidence, gives the student a fair chance, and often becomes a teaching moment.
3. Which assignment change best makes learning visible?
Process checkpoints show how thinking developed, which supports both learning and integrity.
Up Next
Module 5 brings everything together: you’ll plan a unit that uses AI fluently in your teaching and teaches fluency to your students.