Most advice about teachers and AI is about students using it. This is about you using it, and specifically about the hours it can take off a week rather than anything visionary about the future of education.
Start with admin, not pedagogy
The instinct is to try it on the interesting part — lesson design, differentiation, feedback quality. Those are where your judgement matters most, which makes them the worst place to start.
Start with the tasks that are pure time cost and require no professional judgement at all. That is where the hours actually are, and where being wrong costs nothing.
What genuinely saves time
Turning one thing into another format
You have a lesson plan; you need a worksheet, a slide deck, a parent-friendly summary and a version for a student who reads two years below level. That is four documents from one source, and it is the single biggest time saver available.
The content is yours. It is doing the reformatting, which is exactly what it is good at.
First-draft comments and reports
Not final comments. Give it your bullet points on a student — honest, shorthand, the way you would say it to a colleague — and ask for a professional paragraph. Then edit it, because you know the child and it does not.
The blank page is most of the time cost of report season. Removing it while keeping the judgement is the right split.
Differentiation
"Rewrite this passage at three reading levels." "Give me a version of this worksheet with more scaffolding." "What would confuse a student who has not met fractions yet?"
This is work that genuinely improves lessons and that most teachers do not have time to do properly. It is also low-risk, because you review everything before it reaches a child.
Explaining your own subject differently
Ask for five analogies for a concept. Four will be poor. The fifth is sometimes better than the one you have used for years, and you would not have thought of it because you already understand the thing.
The email
Parent emails, cover instructions, department updates. Say what you mean bluntly, ask for the diplomatic version.
What to be careful with
- Anything with a named student in it. Do not paste a child's name, work, or details into a general AI tool. This is the one that matters most, it is a data protection issue rather than a preference, and your school almost certainly has a policy. Anonymise, or do not use it.
- Marking. It cannot assess against your rubric reliably, it does not know what you taught, and a grade is a professional judgement you are accountable for. Feedback phrasing, yes; grades, no.
- Factual content in your subject. It will produce confident errors, and a wrong date in a history worksheet is worse than no worksheet.
- Detecting AI in student work. Detectors are unreliable in both directions. An accusation based on one is not a position you want to defend.
A realistic first month
Pick one repeating task. The weekly newsletter, the same three parent emails, the resource you rewrite every year for a different group.
Get it working properly once and save the prompt. Then pick the next one.
Do not try to redesign your teaching around it — that is how people spend a term on tooling and end up back where they started. The value here is boring and cumulative: an hour or two a week, taken off the parts of the job nobody went into teaching for.
If you want the general version of this, using AI at work without overhauling everything covers the same idea outside a school, and how to write a prompt that actually works is the difference between a usable draft and a bland one.