AI Skills for Educators: Teach Better with AI Tools

Teaching has a large preparation layer and a small, dense delivery layer. The preparation is where AI earns its place: variants of a worksheet, examples at three difficulty levels, a rubric draft, the same explanation for a learner who didn't get the first one. The delivery layer — noticing that a room has stopped following, deciding whether a wrong answer is a misconception or a slip — isn't going anywhere.

The mistake worth avoiding is applying AI to the delivery layer because it's the visible one.

Skill one: describe the misconception, not the topic

"Explain photosynthesis" produces a competent textbook paragraph, which your learners already have.

"Explain photosynthesis to someone who thinks plants get their mass from soil" produces something useful — because it targets the specific wrong model in the learner's head rather than restating the correct one.

Teaching is rarely about presenting correct information. It's about displacing an incorrect one. Prompts that name the misconception outperform prompts that name the topic by a wide margin.

Skill two: generate the wrong answers too

Writing plausible distractors for a multiple-choice question is genuinely hard and genuinely time-consuming — they have to be wrong for a reason a learner might hold, not just wrong.

This is one of the best AI uses in education. Ask for four distractors and, for each, the misconception a learner would have to hold to choose it. You'll reject some. The ones you keep are better than what you'd have written at 6pm on a Sunday.

Skill three: build templates, not one-off materials

Most educators regenerate similar materials every term and start from scratch each time.

The alternative: once a worksheet works, describe what made it work — structure, difficulty ramp, how many worked examples before independent practice, what the final question demands. That description is a template. Next term you supply the topic.

This compounds. Prompting from scratch doesn't.

Skill four: keep the assessment judgement

AI can mark against a rubric consistently and quickly. What it can't do is notice that a learner's wrong answer reveals sophisticated thinking, or that a technically-correct answer was copied.

The workable split: AI handles the mechanical pass and drafts feedback. You review anything at the boundaries — very high, very low, and anything anomalous. That's where the teaching information is.

Farida — the explanation that finally worked

Farida teaches accounting at a further education college. Every year the same cohort of students hit the same wall on accruals, and every year she explained it the same way, more slowly.

She asked AI for five explanations of accruals aimed at someone who believes an expense happens when money leaves the account. The fifth used a gym membership: you pay in January, you consume it monthly, and the cost belongs to the months you used it.

It wasn't sophisticated. It worked, because it matched the specific wrong model the students actually held — one she'd stopped being able to see, because she'd been fluent for twenty years.

Nathan — the worksheet that became a system

Nathan teaches secondary science. He'd been generating practice worksheets with AI each week — decent output, entirely disposable, starting over every time.

Halfway through a term he took the three worksheets that had worked best and asked what they had in common. The answer was concrete: two worked examples before independent practice, difficulty rising every third question, a final question requiring transfer rather than repetition, and never more than twelve items.

He turned that into a standing instruction. Now he supplies a topic and gets something usable in one pass rather than three.

The time saved was real. The bigger gain was that his worksheets became consistent, so students stopped having to work out the format each week.

The one thing

Use AI on the preparation layer, hard. Variants, distractors, alternative explanations, rubric drafts, differentiation — all of it, and build templates so the effort compounds.

Keep the delivery layer and the boundary cases in assessment. That's where teaching actually happens, and it depends on reading a room and a wrong answer — neither of which is in the file.