How to use AI as a teacher, without losing your evenings to it

A teacher types "write me a lesson plan on fractions" into a chatbot. The result is generic and unusable. The teacher concludes that the tool does not understand teaching, and stops there. That conclusion is wrong, but the experience behind it is common enough to be worth examining. A lesson plan is one of the most context-dependent documents a teacher produces. A model that has not been told the year group, the period length, or what the class covered last lesson has no way to guess correctly.

The habit that changes the output

The fix is not a better tool. It is four lines, written once and reused for the rest of the term: the year group, the class size, the period length, the resources actually available, and the one or two things this class typically struggles with. Paste that at the top of every planning conversation for that class. The output stops being generic, because the model is no longer guessing.

This single change matters more than any prompting technique. A model given real constraints produces something a teacher could walk into a classroom with tomorrow morning. A model given a bare topic produces something that gets rewritten from scratch, and that is precisely the experience that makes teachers give up on the tool after one try.

Where the time is actually recovered

• Producing the same worksheet at three ability levels, rather than writing it three times by hand.
• Turning a quick judgement, such as "the content is sound, the structure is the problem," into three sentences of specific written feedback.
• Drafting a calm, professional reply to a difficult parent email instead of losing twenty minutes to a first draft written while irritated.
• Writing exam questions and rubrics quickly enough to revise them properly, rather than shipping a first draft for lack of time.

The line that does not move

The grade belongs to the teacher. A model can produce a plausible-looking mark, but the teacher remains accountable for it, and "the AI decided" is not an answer that holds up in a parent meeting or a formal appeal. What can be delegated safely is the writing-up of a judgement the teacher has already made. The judgement itself cannot be.

The teacher decides the arc of the lesson. The model writes the materials. That division of labour is the entire skill.

The part rarely mentioned: student data

A named student essay pasted into a consumer AI tool discloses a minor's personal data, and it is almost certainly a breach of school policy before it is anything else. The habit worth building early: remove the name, and describe the judgement rather than pasting the work, wherever that is possible. It is both safer and, in most cases, faster.

What about students using it?

AI detectors are not reliable enough to justify accusing a student on the basis of a score. A blanket ban mainly converts disclosed use into undisclosed use; it does not reduce use itself. A more durable approach sets a specific line for each task, and redesigns the assessments that AI can complete without effort, since those were often testing recall rather than the skill the assessment was meant to measure.

Where to start

The free CourseParse course covers the fundamentals: how these tools actually work, why generic prompts produce generic answers, and how to check output rather than trust it by default. It is worth completing first if those habits are not yet automatic.

The classroom-specific version continues from there: lesson planning, differentiation, marking, assessment design, and a workable policy for students using AI, built around real classroom artefacts rather than generic examples. AI for Teachers picks up exactly where this article ends.