Stopping the AI ‘Witch Hunt’: From Policing to Pedagogy

The world of education is at a crossroads. On one side, AI tools like ChatGPT act as 24/7 private tutors that provide revolutionary personalized learning frontiers. On the other side, a “Witch Hunt” has emerged, driven by a strong reliance on AI detection software. However, these tools often function more like probability guesses than reliable forensic instruments, creating a culture of distrust that hits foreign language learners the hardest.¹ Furthermore, we must consider the long-term impact; while AI can quickly polish a paper, it may not help students build lasting writing skills. In fact, students who rely too heavily on AI tend to stagnate, showing a lack of motivation to write even when using these tools.²

In Foreign Language Teaching (FLT), the hunt for AI can feel like a trap. Research shows that AI detectors are statistically more likely to flag writing by non-native speakers because the structured, formulaic English taught in EFL curricula mirrors the patterns used by LLMs.³ When a student successfully uses the Academic Word List, they are paradoxically more likely to be accused of cheating.

Using AI detectors to catch cheating is unfair and doesn’t work. Because AI is getting better at writing like humans, these detectors often make mistakes. This creates a bad environment where a student who actually improves their writing might be accused of cheating instead of being praised. To fix this, teachers should stop acting like “police” trying to catch students and instead focus on helping them through the actual steps of the writing process.

Shifting from Policing to Pedagogy

Rather than judging a finished paper—which could have been made by a machine—teachers are encouraged to look at how a project grows. By reviewing early drafts and Version History, we can see the student’s hard work step-by-step. Talking to our students about their progress helps us value how they got to the final result, allowing us to build back the trust that was lost.

According to recent frameworks, many educators are currently navigating Five Stages of Grief regarding AI—moving from Denial and Anger (the Witch Hunt) toward Acceptance and alternative assignment design.⁵ Acceptance means shifting the focus to process-oriented learning, where the evolution of a student’s ideas becomes as visible and valuable as the final product. This transition is a core focus of the AI2Improve Project, which aims to support teachers in navigating these pedagogical shifts.

As we transition toward acceptance, here are some emerging approaches educators are exploring to foster authentic writing in the classroom:

1. Process vs. Product: Grading the Journey

Instead of giving 100% of the grade to the final essay, break it into breadcrumbs:

• The Breadcrumb Method: Allocate 10% for a brainstorm map, 20% for an outline, and 20% for a first draft. The final product is only worth 50%. However, this can also be generated by AI. So, to move away from policing and toward authentic engagement, teachers can adopt a Reflective Writing Cycle. This model focuses on the student’s agency and their ability to critically evaluate AI feedback:
• The Structural Dialogue: Before a single word is typed, the student orally explains their essay’s architecture to the teacher. This brief conversation ensures the core ideas are original and provides a baseline for the student’s thinking.
• Supervised Drafting: The initial drafting takes place in a controlled environment (in-class or during a proctored session). This establishes the student’s authentic voice and ensures the human foundation of the work is solid.
• AI as a Feedback Coach: Instead of using AI to generate content, students use it as a critical friend. They prompt the AI to critique their draft for clarity, tone, or counter-arguments.
• The Rationale Submission: The final submission includes the essay plus a Reflective Rationale. The student must explicitly list which AI suggestions they accepted and—more importantly—which they rejected and why.
• Version History Check: Use cloud tools like Google Docs to verify Edit History. Four hours of typing is a human fingerprint; a single paste of 1,000 words is a red flag. However, the goal of the AI2Improve Project is to move beyond this detective role. So, instead of just checking history for proof of guilt, we can encourage Process Documentation as a standard part of academic growth:
• The Living Document: Encourage students to work in a shared space (like Google Docs or Teams) not for surveillance, but for real-time feedback. This allows the teacher to drop in and leave a “Well done on this paragraph!” comment while the ideas are still fresh.
• The Evolution Log: Rather than looking at timestamps, ask students to highlight one section they significantly revised and explain why it changed. This celebrates the “messy” part of writing.
• Milestone Snapshots: Instead of a single final submission, have students submit snapshots of their progress—a rough outline, a key argument, and then the final piece. This makes the final product a natural conclusion to a visible journey.
• Reflection Note: Have students write a paragraph explaining one big change they made between drafts (e.g., “I replaced simple verbs with academic ones”).

2. Oral Defense (Viva Voce): The 5-Minute Chat

Miller⁶ suggests that because it is so easy for AI to write a perfect essay at home, we need new ways to test students. One effective method is the Oral Defense (Viva Voce). In this setup, students talk about their work in person. A short conversation is often more accurate than any AI detector—it checks whether the student actually understands what they handed in.⁷ For a language classroom, this is actually a great thing: it moves the focus away from a “perfect” final product and puts it back on the student’s real ability to speak and explain their own ideas. By shifting to oral defenses, we move beyond simply ‘policing’ AI. Instead, we encourage responsible and critical integration, where students may use technology to gather ideas but must maintain the independent mastery required to explain and defend those ideas in their own voice:

• The “Deep Dive” Question: Pick a complex sentence from their essay and ask, “You used the word ‘paradoxical’ here. Can you explain what that means in your own words?”
• The “Source” Question: Point to a specific argument and ask, “Where did you find this information, and why did you think it was important for your essay?”
• The “What’s Next” Question: Ask, “If you had two more weeks to work on this paper, what else would you add?”

Why this works better than an AI Detector?

• It promotes autonomy: It tells the student, “You can use tools, but you must own the knowledge.”
• It builds “Sentipensante” learning: It connects the intellectual output (the essay) with the human emotional presence and voice (the conversation).
• It’s Pedagogical, not Administrative: It treats AI as a part of the writing process that requires human oversight, rather than just a “cheating machine”.

3. In-Class Baselines: Finding the “Signature”

A “baseline” is a sample of writing done in person without any tech. It acts as a fingerprint to compare with take-home work.

• The 15-Minute Flash Write: On the first day of class, give a simple prompt: “Write about a book or movie that changed your mind.” No phones or laptops allowed.
• The Handwritten Journal: Once a week, have students write a 100-word response to a lesson in a physical notebook during class time.
• Style Comparison: If a student turns in a “perfect” essay at home, but their in-class baseline shows they usually struggle with past tense verbs or have a specific way of starting sentences, the teacher has a clear, non-aggressive way to start a conversation. Example: “Your essay is excellent! It looks very different from your in-class writing?”⁵

Conclusion

The challenge before us is not merely technical, but deeply pedagogical, and the transformation of our classrooms begins with the empowerment of the educator. Let us move beyond the paralyzing era of suspicion and, through the AI2Improve Project, build a future where teachers are no longer observers of the digital shift, but its architects. Our mission is to equip teachers with the literacy and strategies needed to navigate this new landscape—not just to manage AI as a tool, but to reclaim their role as mentors who inspire an authentic human voice in a digital world. By investing in our own professional growth today, we ensure that the bond of trust remains the foundation of every student’s journey toward true understanding.⁹

Ready to transform your classroom? Click here to access our AI2Improve Teacher Training modules and lead the shift from policing to pedagogy.

References

1. Liang, W., Yuksekgonul, M., Postolalsky, Y., Liao, J. P., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), 100779.
2. Berutu, H., Dewi, U., & Daulay, S. H. (2024). What AI-based writing assistant actually improved: Writing quality or writing skills? [Unpublished manuscript/Working paper]. (Original work published 2023).
3. 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
4. Sadasivan, V. S., Kumar, A., Balasubramanian, S., Wang, W., & Feizi, S. (2023). Can AI-generated text be reliably detected? arXiv preprint arXiv:2303.11156. https://doi.org/10.48550/arXiv.2303.11156
5. Cox, K., Das, R., Draucker, S., Nadeau, A., Nesbit, K., & Thierauf, D. (2025). Death of the Essay? Generative AI, Literature Teachers’ Five Stages of Grief, and Alternative Assignment Design in Victorian Studies. Victorian Network, 12.
6. Miller, M. (2023). AI for educators: Learning strategies, teacher efficiencies, and a vision for an artificial intelligence-integrated future. Ditch That Textbook.
7. Perkins, M. (2023). Academic integrity considerations of AI Large Language Models in the post-pandemic era. International Journal for Educational Integrity, 19(7).
8. Vásquez Cortez, L. H., Aguas Veloz, J. F., Villota Oyarvide, W. R., Tacle Humanante, P. M., & Tapia Ramírez, C. S. (2026). Artificial intelligence in higher education 5.0: Ethical implications, pedagogical innovation and personalized learning. Data and Metadata. https://doi.org/10.56294/dm20261295
9. Wang, C., & Tian, Z. (Eds.). (2025). Rethinking writing education in the age of generative AI (1st ed.). Routledge.