Teach students to question and verify AI outputs
A class can study AI without every student opening an account. A teacher demonstration, printed examples or an approved school tool can support the same discussion. Choose the format for the students, school policy, accessibility needs and available connection.
This is a proposed sequence, not a validated curriculum for every age or subject. UNESCO’s guidance on generative AI in education addresses privacy, age-appropriate use and institutional preparation. Check the actual provider’s age requirements and required permissions before use; a suggested age in a lesson is not permission to create an account.
Prepare before the lesson
- Define permitted assistance and what students must do themselves.
- Use fictional business examples and public or authorised teaching material. Do not upload identifiable student work, health or disability records, contact details or private family information.
- Provide an equivalent non-AI activity without penalising students who cannot or should not use the service.
- Check every answer key, source and practical activity yourself. Do not let a model design unsupervised experiments.
- Explain how students should disclose assistance and where to ask for help.
A four-part teaching sequence
1. Compare a claim with its source
Use a short approved reading and a prepared AI summary. Students identify supported statements, omissions and invented details. An output may be correct; do not require students to “find one error” when none is established.
2. Provide clear context without private data
Use the context worksheet with an imaginary local shop. Students supply audience, task, units and a brief approved price list. Compare a vague request with a specific one and discuss the limits of both.
3. Revise a bounded draft
Choose a country or sector example such as a visitor-information sheet or a product description. Students check each factual statement, rewrite unclear language and list what needs human confirmation. Nothing is sent to a business or published as part of the exercise without separate permission.
4. Explain a failure and its consequence
Use the risk guide to examine an invented source, mistranslation or unauthorised action. Students propose a test and explain when the tool should stop. A classroom checklist is not a deployment audit.
Subject examples
- History: compare a summary with several assigned sources. Identify disputed claims rather than inventing the voice of an absent community.
- Science: critique a worked explanation against the textbook. Any practical experiment must come from teacher-approved safety procedures.
- Languages: compare translations with a competent speaker’s version, including local register and ambiguous terms.
- Mathematics: check a worked solution independently and explain each correction.
- Arts: discuss provenance, permissions and representation using authorised material.
- Computing: inspect a small sandboxed example without credentials, real accounts or production access.
Assess learning, not detector scores
Ask students to explain their reasoning, verify sources and revise their work under the school’s policy. AI-generated-work detectors are not proof of misconduct. Do not use a general model to decide grades, discipline, admission or support eligibility.
Collect feedback on understanding, access and checking effort. A fluent assignment or fast completion does not by itself demonstrate learning. Adjust the lesson rather than treating one tool as essential.
UNESCO source checked 11 November 2026. Native-language and curriculum approval remain with the responsible educators.
Created by Adrian Dunkley | MaestrosAI | maestroailabs.com | ceo@maestrosai.com. Editorial review: 11 November 2026.