What these tools do
Generative AI tools produce text, images or code by predicting likely output from patterns in large amounts of training data. They do not look facts up the way a search engine does unless the tool is built to search, and even then they can misread or misquote what they find. That one fact explains most of their strengths and weaknesses.
Sensible uses for teachers
- First drafts of routine material. Rubric wording, parent letter templates, exit-ticket questions. You still edit every line.
- Differentiation. Rewriting a passage at a simpler reading level, or adding a glossary of key words, then checking that nothing important was lost.
- Practice generators. Extra worked examples or quiz questions on a topic you have already taught, reviewed before students see them.
- Brainstorming. Lesson hooks, discussion prompts or alternative explanations when your usual one is not landing.
- Modelling critique. Showing the class an AI answer and asking them to find the errors is a strong critical-thinking exercise in itself.
Known limits
- Fabrication. Tools can invent references, quotations, dates and statistics that look entirely plausible. Never pass on a fact you have not checked against a reliable source.
- Bias. Output can reflect stereotypes present in training data. Read generated examples, names and images with that in mind.
- Out of date. A model may not know about recent changes to curricula, exam specifications or local rules.
- Uneven quality by subject. Output may be fluent in one area and unreliable in another, especially multi-step maths and specialist topics.
- No knowledge of your students. It cannot judge what your class already understands. That remains your job.
Academic integrity
Blanket bans are hard to enforce and blanket permission undermines learning. Many schools now set expectations assignment by assignment. A simple three-level scale works well:
| Level | What students may do | Example task |
|---|---|---|
| No AI | Work entirely on their own | In-class writing, tests, personal reflections |
| AI for support | Brainstorm or get feedback, but write the final work themselves and say how they used AI | Research projects, essay planning |
| AI as part of the task | Use AI openly and critique or improve its output | Comparing an AI summary with the source text |
Designing for process helps more than policing. Ask for planning notes and drafts, build in short conferences where students explain their reasoning, and do some writing in class.
Tools that claim to detect AI-written text can produce false positives, and they should never be the only evidence in a misconduct decision. Follow your school's integrity policy and talk with the student first.
Protecting student privacy
- Use approved tools. Check which AI tools your school or district has reviewed. Consumer accounts may store and use what you type under very different terms.
- Keep personal data out. Do not paste names, grades, behaviour notes, health information or anything that identifies a student into a general-purpose chatbot.
- Check age terms. Many AI services set minimum ages or require parental permission for younger users. Read the terms before asking students to sign up.
- Tell families. Explain which tools are used in class and why. See the student privacy basics for the legal background.
A low-risk way to start
Try one teacher-facing use for a month, such as drafting differentiated reading questions, and keep a note of how much editing each output needed. That gives you real evidence about time saved before you bring any tool in front of students.