Most advice about AI in teaching stops at "try ChatGPT." That's not advice — that's a shrug.
The real skill isn't access to a tool. It's knowing what to ask it. A vague prompt gives you a vague, generic answer you'll never use. A precise one gives you something you can take into class tomorrow.
Below are five prompts I keep coming back to. They work in any AI assistant — the wording matters far more than the brand. Steal them, adapt the bracketed parts, and make them yours.
1. The "Explain It Three Ways" prompt
When a concept lands flat in lecture, the problem usually isn't the students. It's that you only have one explanation, and it works for the third of the room that already thinks like you.
"I'm teaching [concept] to [year level] students in [discipline]. Give me three different explanations of it: one using a concrete real-world analogy, one that's visual and spatial, and one that builds step by step from first principles. Keep each under 100 words."
Why it works: you walk in with three on-ramps instead of one. When the first explanation gets blank stares, you have two more ready — and you'll often find one of them explains the idea better to you, too.
2. The "Devil's Advocate" prompt
Reading responses get repetitive when every student agrees with the author. This flips the discussion open.
"Here is the core argument from [reading/author]. Generate the three strongest objections a thoughtful critic would raise — steelman each one, don't strawman it. Then note which objection is hardest to answer and why."
Why it works: you get a ready-made set of counter-arguments to seed discussion, and the "steelman" instruction stops the AI from producing weak, easily-dismissed objections. Students engage far more when the opposing view is genuinely strong.
3. The "Misconception Detector" prompt
You can't address misunderstandings you can't predict. This surfaces them before the exam does.
"I'm about to teach [topic]. List the five most common misconceptions students at [level] hold about this, why each one feels intuitively correct, and one diagnostic question I could ask to reveal whether a student holds it."
Why it works: it turns your assessment from "did they memorize it?" into "do they actually understand it?" The diagnostic questions are the gold here — drop one into a lecture as a quick poll and you'll instantly see where the room actually stands.
4. The "Feedback Skeleton" prompt
Grading 60 essays drains the energy you need for the feedback that actually changes student work. Use AI for the scaffold, not the judgment.
"Here is my rubric: [paste rubric]. Read this student essay and draft feedback organized by each rubric criterion. For each, note one specific strength and one specific area to improve, citing the exact passage. Do not assign a grade — I'll do that."
Why it works: you stay the one making the evaluative call (this matters — both ethically and pedagogically), but you skip the blank-page fatigue. You review and personalize the draft instead of writing every comment from scratch. One caution: never paste identifiable student work into a tool your institution hasn't approved for it. Check your data-protection rules first.
5. The "Cold Open" prompt
The first ninety seconds of a class decide whether attention is yours or the phone's. This prompt manufactures a hook.
"Give me five ways to open a lecture on [topic] that would make [discipline] students lean in: a surprising statistic, a counterintuitive question, a short scenario, a real recent example, and a 'what would you do?' dilemma. Each should take under a minute to deliver."
Why it works: it attacks the attention problem at exactly the moment it's most fragile. You don't have to use AI's exact words — you're after the angle. Often the surprising statistic or the dilemma is the spark you'd never have reached for on your own.