AI for teachers: keep the judgement, lose the handwriting.

The honest division of labour

Every useful AI-for-teachers workflow starts with the same sentence: AI drafts, you decide. A model is very fast at producing a first pass — a rubric-aligned comment, a paragraph of feedback, a worksheet outline, a rewrite of a paragraph at a different reading level. It is not fast, and never will be, at knowing whether the child who wrote that paragraph is having a good week. Keep the judgement human. Push the handwriting downstream.

The reason to be strict about this isn't philosophical — it's practical. The moment you stop reading, two things happen: the marks stop being yours, and you stop noticing the patterns that actually improve your teaching. AI is a leverage tool for teachers who read. It is a catastrophe for teachers who don't.

Marking: the biggest single win

If you only automate one thing this term, automate essay marking. A class set of thirty discursive essays takes a competent English or LO teacher somewhere in the region of six to seven hours to mark properly — rubric-aligned commentary, individual feedback, mark reconciliation. It is the single largest recoverable block of time in the profession, and it happens roughly once a fortnight.

Modern rubric-aware tools can draft the commentary — stance, balance, evidence, register — in about a minute per essay. You read the essay alongside the draft, adjust the language, correct the mark where you disagree, and move on. Realistically it collapses a lost Sunday into an after-school evening. Nothing about the standard drops; the meaningful judgement stays exactly where it always was.

Worksheets and planning

Worksheets are the other quiet time sink — especially in foundation and intermediate phase, where you need three difficulty tiers of the same content by Wednesday. AI is genuinely good at generating differentiated variants of an exercise you already have: same skill, different reading level, or same reading level with a different scaffold.

Before you generate anything from scratch, though, check whether the worksheet already exists. Skill Squad's Froly hosts over a thousand free educational printables — the fastest AI workflow is the one you don't have to run. Use AI to fill the gaps, not to redo the work someone already did.

The POPIA rules, in plain English

POPIA (South Africa's Protection of Personal Information Act) applies the moment you're processing personal information about a learner — their name, their work, their marks, their photograph. Four rules cover most classroom AI use:

1. Never paste learner names into a public chatbot. Use "Learner A", "Learner B", or the first initial. Strip surnames and class codes.
2. Only use tools with a written no-training policy. If the terms don't explicitly say your inputs won't be used to train the model, assume they will be.
3. Check for a deletion-on-request option. POPIA gives learners (and their guardians) a right to request their data be removed. Your tool needs to be able to honour that.
4. The teacher signs off — always. AI output is a draft. The mark on the report is yours.

These aren't a compliance ceiling; they're a floor. Anything less and you're carrying risk that isn't yours to carry.