The misunderstanding about AI in education
When people talk about AI in teaching, the conversation almost always revolves around two topics: students cheating with ChatGPT and teachers generating quizzes with ChatGPT.
That's like using a Ferrari to do the grocery run. It works, but it's not exactly the best use.
AI can do much more for pedagogy — if you move beyond the obvious uses. Here are 10 concrete uses, ranked from simplest to most transformative.
Level 1: AI saves time
These uses don't fundamentally change pedagogy. They accelerate what you were already doing.
1. Generate exercise variants
The problem: Creating 4 versions of the same exercise to prevent copying takes an hour.
The AI use: Ask an LLM to generate variants from a model exercise. Same structure, different data. In 2 minutes, you have 10 versions.
Tools: ChatGPT, Claude, Mistral — any LLM.
Real gain: 45 minutes per exercise session.
2. Summarize and adapt content
The problem: You have a 30-page research article for sophomores who have neither the time nor the level.
The AI use: Ask for a 500-word summary at sophomore level, with key concepts in bold and a glossary of technical terms.
Tools: ChatGPT, Claude. Always verify the result — LLMs can distort nuances.
Real gain: 1 hour of rewriting avoided.
3. Pre-grade papers
The problem: 80 papers to grade over the weekend. Half the time goes to recurring comments ("lacks structure," "no sources").
The AI use: Submit papers to an LLM with your grading rubric. It identifies strengths and weaknesses, drafts preliminary comments. You review, adjust, and sign off.
Tools: ChatGPT with a rubric prompt. Platforms like Gradescope integrate AI.
Real gain: 40-60% of grading time. Time saved is reinvested in personalized comments on papers that deserve them.
Level 2: AI personalizes learning
These uses start changing the student experience.
4. Instant personalized feedback
The problem: A student submits interim work. They won't get feedback for 2 weeks — too late to course-correct.
The AI use: A chatbot trained on your rubric gives immediate feedback on interim work. Not a grade — suggestions: "your introduction lacks a problem statement," "you didn't use the data from table 3."
Tools: ChatGPT with a structured prompt, or dedicated tools (Khanmigo, Sana Labs).
Impact: Students iterate in real time instead of waiting for delayed feedback. Formative feedback increases final results by 20-30% (Black & Wiliam, 1998).
5. Adaptive pathways
The problem: In a group of 40, some students already master the basics. Others are lost. The lecture ignores both.
The AI use: An adaptive system adjusts difficulty and content based on student responses. Those who've mastered move forward. Those who are stuck get additional explanations.
Tools: Area9 Rhapsode, Smart Sparrow, or a simple decision tree in a Wooclap quiz.
Impact: Each student works at their level. Class time is reserved for what only the teacher can provide.
6. Translation and accessibility
The problem: You have international students struggling with technical language. Or students with disabilities who need alternative formats.
The AI use: Instant translation of your materials (DeepL), transcription of your lectures (Whisper/Otter), generation of simplified or plain-language versions.
Tools: DeepL, Whisper, ChatGPT.
Impact: Accessibility is no longer a separate project — it's an automatic byproduct of your course.
Level 3: AI transforms pedagogy
These uses fundamentally change what's possible in the classroom.
7. The Socratic tutor
The problem: You'd like every student to have a personal tutor who guides them through questions, not answers. But you have 120 students and 3 hours of tutorial.
The AI use: A chatbot configured to never give the answer, but always ask the next question. "You say the problem is cash flow. What makes you think that? Have you looked at the payment terms?"
Tools: ChatGPT with a Socratic system prompt, Khanmigo.
Impact: Students develop their reasoning instead of seeking the right answer. The cognitive load is on the student, not the chatbot.
8. Realistic data generation
The problem: For an accounting or statistics exercise, you need realistic data. Inventing a coherent dataset takes hours.
The AI use: Ask the AI to generate a dataset with precise constraints: "a 50-employee industrial SME, €3M revenue, cash flow difficulties, with ratios consistent with the metalwork sector." The AI produces a realistic dataset in 30 seconds.
Tools: ChatGPT, Claude. For complete, coherent financial data: MEτiS generates entire companies with chart of accounts, income statements, and org charts.
Impact: Each group can work on different data. No more "I found last year's answer key."
9. Pedagogical characters
The problem: You'd like your students to interrogate an expert, an unhappy client, an HR director facing a conflict — but you have neither the actors nor the budget.
The AI use: AI characters configured with a personality, specific knowledge, and limits. They respond based on what they "know" — and don't volunteer everything. The student must ask the right questions.
Tools: ChatGPT with a role prompt, or dedicated platforms like MEτiS that manage consistency across multiple characters.
Impact: Students shift from reading to conversation. They form hypotheses, test them, and adjust. This is the core of inquiry-based learning.
10. Complete immersive simulation
The problem: You want to combine everything — realistic data, characters with different perspectives, time-distributed clues, tracking of every interaction, and process assessment, not just product assessment.
The AI use: A simulation platform where AI generates a complete world (company, literary work, historical event) with characters holding partial knowledge. Students investigate freely. The instructor tracks everything in real time and assesses the journey as much as the deliverable.
Tools: MEτiS — create a world in 5 minutes, launch a session, track each team's investigation.
Impact: This is the convergence of all previous uses: realistic data (#8), characters (#9), real-time feedback (#4), process assessment (#5), and AI-proof assessment by construction (#10). The result: each team has a unique experience, and the debrief is the richest pedagogical moment of the session.
The usage matrix
| Use | Effort | Impact | Changes pedagogy? | |-----|--------|--------|-------------------| | 1. Exercise variants | ⭐ | ⭐ | No | | 2. Adapted summaries | ⭐ | ⭐ | No | | 3. Pre-grading | ⭐⭐ | ⭐⭐ | No | | 4. Instant feedback | ⭐⭐ | ⭐⭐⭐ | Yes | | 5. Adaptive pathways | ⭐⭐⭐ | ⭐⭐⭐ | Yes | | 6. Translation/accessibility | ⭐ | ⭐⭐ | No | | 7. Socratic tutor | ⭐⭐ | ⭐⭐⭐⭐ | Yes | | 8. Realistic data | ⭐ | ⭐⭐⭐ | Yes | | 9. Pedagogical characters | ⭐⭐ | ⭐⭐⭐⭐ | Yes | | 10. Immersive simulation | ⭐⭐ | ⭐⭐⭐⭐⭐ | Yes |
Where to start?
If your problem is time: start with uses 1-3. Immediate gains, no pedagogical change.
If your problem is engagement: jump to uses 7-10. The Socratic tutor or pedagogical characters transform the student experience.
If your problem is cheating: use 10 (immersive simulation) makes cheating structurally impossible. Read our guide: 5 Assessment Methods ChatGPT Can't Beat.
If you want to change everything: start with one immersive simulation in one course. 5 minutes of preparation, 2 hours of experience. Observe the difference. Then expand.
The question is no longer "should I use AI in teaching?" The question is: what can you do now that you couldn't do before?