AI Detection at Swiss Gymnasiums: From Kantonsschule Essays to the Matura

The Gymnasium Between Educational Mission and Control Reflex

Swiss gymnasiums and cantonal schools are the institutions where the AI challenge is greatest. Students are old enough to use ChatGPT competently. Written requirements are high — from argumentative essays in German class through historical analysis to the matura thesis. And the pressure for good grades is real.

At the same time, gymnasiums have a special educational mission: preparing young people for university readiness — meaning enabling independent, critical thinking and work. A pure control approach contradicts this mission. A pure trust approach is naive.

This article charts a path between the extremes: a practical framework for gymnasiums that combines verification, pedagogy, and AI competence.

Taking Stock: Where Does the Swiss Gymnasium Stand in 2026?

An informal survey among German and history teachers at Zurich cantonal schools (fall 2025) painted this picture:

• About 80% of respondents suspect that at least a quarter of their students regularly use AI for school writing.
• Fewer than 30% of schools have a formal, written AI policy.
• About 50% of teachers have tried an AI detector at least once — but only 15% use one regularly.
• The most common response to suspected AI use: oral follow-up in a private conversation.

The picture: awareness is high, structure is low. That's exactly where this article picks up.

The Five-Level Model for Gymnasiums

Based on proven approaches from the University of Zurich, adapted for gymnasium needs:

Level 0 — No AI Use

When: Proctored exams, oral exams, entrance examinations.

Measure: Technical safeguards (no smartphones, no internet). Declaration of independent work.

Level 1 — AI for Research

When: Preparatory tasks, research phases, source discovery.

Rule: AI may be used as search assistance. All claims must trace back to verifiable sources. The research process is documented.

Level 2 — AI for Revision

When: Essays, summaries, protocols.

Rule: AI may be used for spelling, grammar, and simple rephrasing. Not for substantive additions or structural reshaping.

Level 3 — AI as Writing Partner

When: Project work, specialized papers, presentations.

Rule: AI may be used for brainstorming, structuring, and feedback. Final formulation is original work. An AI journal is maintained.

Level 4 — AI Freely Usable

When: Programming tasks, exercises where the outcome matters.

Rule: AI may be used freely. Oral defense possible at any time.

The key: each subject teacher declares at the start of the semester which level applies to which assignments — and communicates this in writing.

Subject-Specific Strategies

German Language

The subject with the largest AI attack surface. Essays, argumentative texts, interpretations — everything ChatGPT excels at. Proven strategies:

• Write in class: Important texts are produced in the classroom, not at home.
• Handwritten drafts: Require a handwritten draft before the digital clean copy.
• Style comparison: Compare with the student's earlier work.
• Require personal references: Assignments demanding personal experiences, opinions, and local connections.
• Oral text discussion: Discuss the text one-on-one — anyone who understands their text can explain it.

History and Geography

Here the problem is less about style and more about content: AI can summarize facts and formulate historical arguments that are superficially correct. The danger lies in historical errors that are hard to spot — because the AI sounds plausible.

• Source work over summaries: Tasks based on concrete sources (documents, maps, statistics) are more AI-resistant than open-ended summaries.
• Local history: "Describe industrialization using the example of your municipality." — That requires real research, not ChatGPT.
• Source criticism: Have students critically analyze AI-generated historical texts — this trains both source competence and AI competence simultaneously.

Mathematics and Sciences

In STEM subjects, the AI challenge is different. The problem is less the text and more the solution: AI can solve math problems, model physics scenarios, balance chemistry equations.

• Require the solution path: Grade not just the result but each documented step.
• Exam variations: Slightly modified problem statements preventing a standard AI solution from being copied verbatim.
• Oral examination elements: "Explain why you chose this approach."

Foreign Languages (French, English, Italian)

Language teaching is particularly affected. Translations, essays, text analyses — all can be AI-created, often at higher quality than the student could deliver.

• Strengthen oral competence: Oral exams and discussions in the foreign language cannot be passed with AI.
• Process tasks: "Write three drafts of this text and document what you changed between drafts."
• In-class translations: Important translation tasks during class, not as homework.

The Matura Thesis: The Acid Test

The matura thesis is the biggest single written achievement at the gymnasium. It spans several months, is written at home, and is orally defended at the end. For AI detection, it's simultaneously the biggest challenge and the best test case.

Why the Matura Thesis Is Especially Vulnerable

• Long working period without supervision.
• High performance pressure (the thesis counts toward the matura grade).
• Many students work truly independently on a large project for the first time.
• The supervisor typically sees only intermediate stages, not the full creation process.

A Proven Procedure for the Matura Thesis

1. Set the AI level: The school defines which level applies to matura theses. Recommendation: Level 3 (AI as writing partner with journal).
2. Require an AI journal: Students continuously document which AI tools they used for what. The journal is submitted with the thesis.
3. Regular check-ins: The supervisor meets with the student at least three times — asking comprehension questions about the content each time.
4. Collect drafts: At least one rough draft and one revision are submitted before the final version.
5. Technical verification: The final version is checked with a compliant detector. Results inform the overall assessment but are never the sole criterion.
6. Oral defense (colloquium): In the colloquium, the student must defend their work on substance — questions about the research process, about specific passages, about sources. Anyone who truly wrote their own work passes this conversation effortlessly.

The Role of the Faculty

An AI strategy only works when the entire faculty buys in. If German class has strict rules but history has no controls, the system's credibility crumbles. Three school-level measures:

1. Faculty-wide agreement: All departments agree on a shared tier model. Implementation is subject-specific, but the framework is uniform.
2. Annual training: Half a day per school year where faculty collectively try AI tools, interpret detector results, and discuss assessment design.
3. Central tool provision: The school provides a compliant detector centrally. Individual teachers don't have to search for tools themselves.

Assessment Design: Making Exams More AI-Secure

Even without detectors, clever assessment design can prevent AI from providing an unfair advantage. Some proven approaches:

1. Two-Stage Assessments

First part: written work at home (AI use according to declared level). Second part: oral exam or in-class deep-dive on the same topic. The grade comes from both parts.

2. Portfolio Assessment

Instead of one large paper, several smaller texts are collected over the semester. The teacher sees the development and can spot style jumps.

3. Reflection Component

Every written assignment includes a brief reflection: "What was difficult? Where did I change my mind? What would I do differently next time?" Such questions are very hard for AI to answer convincingly.

4. Source-Based Tasks

Instead of open questions: tasks based on concrete sources covered in class. "Analyze Source C in light of last Monday's discussion." That requires class participation, not ChatGPT.

5. Peer Feedback Rounds

Students give each other feedback on drafts. This creates a social control mechanism: anyone submitting an obviously AI-generated text gets noticed by peers.

When and How to Use AI Detectors

At the gymnasium, AI detectors make more sense than at primary or lower secondary level because texts are longer, more complex, and thus more reliably analyzable. Still, clear rules apply:

• Not as routine screening: Running every paper through the detector is disproportionate and creates a surveillance climate.
• On concrete suspicion: Style mismatch, atypical phrasing, knowledge jumps — when multiple signals converge, a detector check is justified.
• For matura theses: Systematic use is more justifiable here because the work has high stakes and students are informed in advance.
• Always in combination: Detector plus conversation plus draft comparison. Never detector alone.
• Data protection compliant: Swiss hosting, DPA with vendor, anonymized upload.

AIDetector.ch is particularly suited for gymnasium use: Swiss servers, support for German, French, Italian, and English, sentence-by-sentence analysis, and classroom mode for matura thesis batches.

Conclusion: Structure Creates Freedom

The gymnasium doesn't need a surveillance apparatus. It needs structure: a clear tier model, subject-specific implementation, central tool provision, regular training, and the willingness to treat AI not just as something to control but as a subject of gymnasium education in its own right.

The matura thesis is the acid test. Securing it with an AI journal, regular check-ins, technical verification, and an oral colloquium yields a system that's neither naive nor paranoid — but professional.

Sources

• EDK, Framework Curriculum for Matura Schools (current edition).
• University of Zurich, Five-Level Model for AI Use, 2024.
• Canton of Zurich, Upper Secondary and Vocational Education Office: Recommendations on AI at Gymnasiums, 2024.
• Swiss Matura Commission, Guidelines for the Matura Thesis.
• Canton of Aargau, Department of Education, Culture and Sport: AI in Teaching Handbook, 2025.