Teaching Students to Think With AI, Not Because of It

A year ago, I wrote about students using AI as a crutch; about the shortcut that bypasses productive struggle; about the illusion of learning that comes from pasting a question into Gemini, Claude, or ChatGPT and copying the result. I stand by every word of that post. But I’ve spent the past six months in classrooms watching what happens when teachers do something different, and I’ve come to believe that the framing of “AI vs. thinking” misses something important.

The problem was never really that students were using AI. The problem was that they didn’t know how to use it well, and neither, honestly, do most of us. I still see students struggling and teachers developing lessons per volume for consumption instead of by quality.

What a bad prompt looks like

Ask a student to compare the economic policies of the New Deal with modern progressive proposals, and you know what they’ll type into the chat window? Exactly that sentence. They’ll paste the assignment prompt, verbatim, and accept whatever comes back. What they get is a plausible-sounding answer that touches every required keyword and says almost nothing. It satisfies the rubric without generating a single original thought.

That’s not an AI problem; it’s a literacy problem. We’ve always had students who could locate a Wikipedia article and rephrase it. AI is just faster and more fluent at producing that same surface-level response. The student who prompts poorly is doing the same thing that student has always done: outsourcing the cognitive work at the exact moment the cognitive work was the point.

But here’s what I’ve come to understand: prompting is a skill. It is, in fact, a skill that looks a great deal like the thinking skills we already claim to teach.

What a good prompt requires

A student who prompts well has to do something that a student who prompts badly never does: they have to think before they type. A meaningful AI query requires the student to narrow the question, supply relevant context, define the terms they actually care about, and anticipate the kinds of responses that would and wouldn’t be useful. That process is, almost word for word, the prewriting process we teach in every English class. It is the analytical framework we use in every history seminar.

Consider what a better prompt for that New Deal comparison actually requires. The student needs to know what the New Deal’s core economic mechanisms were (deficit spending, direct employment, financial regulation) and be able to name at least one contemporary policy that parallels each one. They need to recognize which dimension of comparison is most historically interesting. They have to make a choice about what they actually want to argue, because a vague prompt returns a vague answer, and a vague answer is useless.

That is prior knowledge, disciplinary reasoning, and argumentative intent working together before the student ever hits send. The student who can write that prompt has already done a significant share of the intellectual work the assignment was designed to produce.

The threshold question

What I’ve started calling the “threshold question” is the pivot point in this whole conversation: has the student already developed enough foundational knowledge and skill to use AI productively? Or are they using AI to avoid developing it in the first place?

Those are very different situations, and they demand very different responses from us as educators.

For a student who has read the sources, participated in the seminar, and built a working understanding of the historical period, a well-placed AI query can be a genuine thinking tool; a way to test an argument, generate counterexamples, or push through a place where they’re stuck. That’s the use I want to model and encourage.

For a student who hasn’t done the reading, hasn’t taken the notes, and has no framework for evaluating what the AI gives back, that same query is just outsourcing. The AI fills a gap that should have been filled by the student’s own encounter with the material, and the student walks away having learned nothing except that the tool works quickly.

This is why “AI literacy” can’t just mean “how to use the tools.” It has to include a prior conversation about when and why to use them, and what the student needs to bring to the table before the tool is worth reaching for.

What I’ve actually tried

The most effective shift I’ve made is moving AI use to the revision and stress-testing phase rather than the drafting phase. Students write their first response without AI, then query the model specifically to challenge their reasoning: “Here is my argument. What is the strongest counterargument? What evidence would undermine this claim?” That’s a use that requires having an argument first, which means the AI can only run if the student has already done the foundational thinking. It also teaches something genuinely useful: how to argue against yourself, which is the core of good historical writing.

I’ve also started using what I call “bad prompt, better prompt” as a classroom activity. Students share AI queries side by side: the lazy version and the revised version. We read the responses together and analyze what changes. The lesson isn’t technical; it’s epistemic. Why does a specific question get a more useful answer than a general one? What does it mean to give an AI enough context to be genuinely helpful? How do you know when an answer is too easy and needs to be pushed further? These conversations sound like writing workshop; they are writing workshop, just with a new kind of text to examine.

The year-end honest assessment

I don’t think we’ve solved this. I think most students still default to the bad prompt when the assignment feels tedious and the deadline is close. I think the illusion of learning is still alive and well, dressed up in smoother prose than it used to wear.

But I no longer believe the right goal is to keep AI out of the thinking process. The right goal is to raise the threshold for what counts as thinking, and to make clear to students that a tool is only as good as the mind directing it. In a world where information is genuinely free and instant, the scarce resource is judgment: knowing what question to ask, evaluating whether the answer is any good, and caring enough about the difference to push further.

That’s what we’ve always been teaching, or trying to. AI just makes it more visible when we’re succeeding and more visible when we’re not.

This is part of my AI in the Classroom series where I write about how AI is changing education and what I have learned from using it. Read the rest of this series here.